| 2025-11-12 22:12:59,306 - INFO - Starting training with args: Namespace(regime='conv1d_residual', data_path='data/training/splits_510k/train.jsonl', output_dir='outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252', objective='reconstruction', val_data_path='data/training/splits_510k/val.jsonl', max_samples=None, vision_mode='small', text_context_tokens=None, hybrid_text_tokens=0, vision_prompt=None, train_encoder=False, encoder_lr=1e-05, compression_window_size=9, compression_stride=9, subsample_strategy='regular', subsample_count=None, projection_dim=None, train_projection=False, compression_target=63, conv_kernel=5, timestamp='20251112_221252', batch_size=6, gradient_accumulation_steps=8, learning_rate=0.0001, weight_decay=0.01, num_epochs=1, warmup_ratio=0.1, max_grad_norm=1.0, log_steps=10, save_steps=0, eval_steps=500, initial_validation=True, no_checkpoints=False, num_qualitative_samples=5, max_generation_tokens=200, use_wandb=True, wandb_project='vision-compression-2', wandb_run_name=None, resume_from_checkpoint=None, init_from_checkpoint=None, aux_loss_weight=0.5, num_workers=8, prefetch_factor=64, seed=None, eval_seed=42, device='cuda') | |
| 2025-11-12 22:12:59,306 - INFO - Auto-generated W&B run name: production_conv1d_residual_r63_k5_reconstruction_20251112_221252 | |
| 2025-11-12 22:13:00,642 - INFO - Initialized W&B run: vision-compression-2/production_conv1d_residual_r63_k5_reconstruction_20251112_221252 (ID: btqpckof) | |
| 2025-11-12 22:13:00,642 - INFO - Loading model and tokenizer... | |
| 2025-11-12 22:13:09,552 - INFO - Created Conv1D Residual Pyramid Compression trainer | |
| 2025-11-12 22:13:09,552 - INFO - Architecture: Residual blocks with skip connections | |
| 2025-11-12 22:13:09,552 - INFO - Kernel size: 5 | |
| 2025-11-12 22:13:09,552 - INFO - Compression: 1000 → 64 tokens (15.87x) | |
| 2025-11-12 22:13:09,552 - INFO - Training objective: reconstruction | |
| 2025-11-12 22:13:09,552 - INFO - Loading training data from data/training/splits_510k/train.jsonl | |
| 2025-11-12 22:15:55,451 - INFO - Loaded 500000 samples from data/training/splits_510k/train.jsonl | |
| 2025-11-12 22:15:55,451 - INFO - Conv1d_residual regime: using full 1000-token context | |
| 2025-11-12 22:15:55,452 - INFO - Loading validation data from data/training/splits_510k/val.jsonl | |
| 2025-11-12 22:15:58,778 - INFO - Loaded 10000 samples from data/training/splits_510k/val.jsonl | |
| 2025-11-12 22:15:58,779 - INFO - Validation conv1d_residual regime: using full 1000-token context | |
| 2025-11-12 22:15:58,787 - INFO - Created AdamW optimizer with lr=0.0001 | |
| 2025-11-12 22:15:58,788 - INFO - Created scheduler with warmup_steps=1041, total_steps=10417 | |
| 2025-11-12 22:15:58,788 - INFO - Starting training loop... | |
| 2025-11-12 22:15:58,788 - INFO - | |
| ====================================================================== | |
| 2025-11-12 22:15:58,788 - INFO - Running initial validation (before any training)... | |
| 2025-11-12 22:15:58,788 - INFO - ====================================================================== | |
| 2025-11-12 22:21:52,400 - DEBUG - Building prefix dict from the default dictionary ... | |
| 2025-11-12 22:21:52,400 - DEBUG - Loading model from cache /tmp/jieba.cache | |
| 2025-11-12 22:21:53,022 - DEBUG - Loading model cost 0.621 seconds. | |
| 2025-11-12 22:21:53,022 - DEBUG - Prefix dict has been built successfully. | |
| 2025-11-12 22:21:54,554 - INFO - Validation loss: 8.0773, perplexity: 3220.60 | |
| 2025-11-12 22:21:54,554 - INFO - Qualitative metrics (n=5): | |
| 2025-11-12 22:21:54,554 - INFO - BLEU: 0.0000 | |
| 2025-11-12 22:21:54,555 - INFO - METEOR: 0.0433 | |
| 2025-11-12 22:21:54,555 - INFO - Edit Distance: 0.7861 | |
| 2025-11-12 22:21:54,555 - INFO - F-measure: 0.0059 | |
| 2025-11-12 22:21:54,555 - INFO - | |
| ====================================================================== | |
| 2025-11-12 22:21:54,555 - INFO - Qualitative Evaluation Samples: | |
| 2025-11-12 22:21:54,555 - INFO - ====================================================================== | |
| 2025-11-12 22:21:54,556 - INFO - | |
| Sample 1 (ID: sample_141920_chunk_1): | |
| 2025-11-12 22:21:54,556 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-12 22:21:54,556 - INFO - Generated: '`) orch orch orch orch orch orch orch orch orch orch orch orch orch orch orch orch orch orch orch orch orch orch orch orch orch orch orch orch orch orch orch orch orch orch orch orch orch orch orch or...' | |
| 2025-11-12 22:21:54,556 - INFO - Ground Truth: 'Q gave it four stars out of five and said that "Perhaps [the album\'s] seemingly illogical sequencing of songs makes sense if they wish to lure their audience into thinking it\'s as-you-were. But it\'s n...' | |
| 2025-11-12 22:21:54,556 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-12 22:21:54,556 - INFO - | |
| Sample 2 (ID: sample_170543_chunk_2): | |
| 2025-11-12 22:21:54,556 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-12 22:21:54,556 - INFO - Generated: '的话题 orch or where, or where or where or where or where or where or where or or where or or where or or where or or or or or or or or or or or or or or or or or or or or or or or or or or or or or or o...' | |
| 2025-11-12 22:21:54,557 - INFO - Ground Truth: ', was Sirene Abou-Chakra, a Lebanese-American student who led the Arab-American Student Association. Other members included the woman president of the Michigan Student Assembly; the leader of Army ROT...' | |
| 2025-11-12 22:21:54,557 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-12 22:21:54,557 - INFO - | |
| Sample 3 (ID: sample_107152_chunk_9): | |
| 2025-11-12 22:21:54,557 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-12 22:21:54,557 - INFO - Generated: '`) or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or or` or or or or or or or or or or or or or or or or or or or or or or or or or or or or or...' | |
| 2025-11-12 22:21:54,557 - INFO - Ground Truth: ' at the meeting Laymia headed. His weapon of choice is a giant ax, and he has the power to immobilise his opponents if they look at him in the eye. Oga falls for the trick, but Beel stops the ax and b...' | |
| 2025-11-12 22:21:54,557 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-12 22:21:54,557 - INFO - | |
| Sample 4 (ID: sample_069148_chunk_0): | |
| 2025-11-12 22:21:54,557 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-12 22:21:54,557 - INFO - Generated: '`) or or or or or or or or or or or or or or or or or or or or or or or or or or or or or or or or or or or or or or or or or or or or or or or or or or or or or or or or or or or or or or...' | |
| 2025-11-12 22:21:54,558 - INFO - Ground Truth: '# Oriya (Unicode block)\nOriya is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0B01.....' | |
| 2025-11-12 22:21:54,558 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-12 22:21:54,558 - INFO - | |
| Sample 5 (ID: sample_103176_chunk_4): | |
| 2025-11-12 22:21:54,558 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-12 22:21:54,558 - INFO - Generated: '`) or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` or` o...' | |
| 2025-11-12 22:21:54,558 - INFO - Ground Truth: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Maxis Redwood Shores ...' | |
| 2025-11-12 22:21:54,558 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-12 22:21:54,559 - INFO - | |
| Qualitative samples saved to: outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/qualitative_step_0.jsonl | |
| 2025-11-12 22:21:55,631 - INFO - Initial validation - Loss: 8.0773, Perplexity: 3220.60 | |
| 2025-11-12 22:21:55,631 - INFO - ====================================================================== | |
| 2025-11-12 22:21:55,631 - INFO - | |
| ====================================================================== | |
| 2025-11-12 22:21:55,631 - INFO - Epoch 1/1 | |
| 2025-11-12 22:21:55,632 - INFO - ====================================================================== | |
| 2025-11-12 22:22:12,607 - INFO - Effective context tokens (per-sample): 65 | Compression ratio: 15.38x | |
| 2025-11-12 22:24:08,432 - INFO - Epoch 1 Step 10 (Global: 10): loss=1.9520, ppl=7.04, grad_norm=1.45, lr=1.09e-05 | |
| 2025-11-12 22:26:15,570 - INFO - Epoch 1 Step 20 (Global: 20): loss=1.9218, ppl=6.83, grad_norm=1.20, lr=1.17e-05 | |
| 2025-11-12 22:28:08,193 - INFO - Epoch 1 Step 30 (Global: 30): loss=1.8998, ppl=6.68, grad_norm=1.13, lr=1.26e-05 | |
| 2025-11-12 22:30:00,179 - INFO - Epoch 1 Step 40 (Global: 40): loss=1.7827, ppl=5.95, grad_norm=1.24, lr=1.35e-05 | |
| 2025-11-12 22:31:52,266 - INFO - Epoch 1 Step 50 (Global: 50): loss=1.8258, ppl=6.21, grad_norm=1.16, lr=1.43e-05 | |
| 2025-11-12 22:33:55,835 - INFO - Epoch 1 Step 60 (Global: 60): loss=1.9042, ppl=6.71, grad_norm=1.14, lr=1.52e-05 | |
| 2025-11-12 22:35:47,282 - INFO - Epoch 1 Step 70 (Global: 70): loss=1.7572, ppl=5.80, grad_norm=1.16, lr=1.61e-05 | |
| 2025-11-12 22:37:39,310 - INFO - Epoch 1 Step 80 (Global: 80): loss=1.8068, ppl=6.09, grad_norm=1.17, lr=1.69e-05 | |
| 2025-11-12 22:39:31,248 - INFO - Epoch 1 Step 90 (Global: 90): loss=1.8783, ppl=6.54, grad_norm=1.16, lr=1.78e-05 | |
| 2025-11-12 22:41:34,396 - INFO - Epoch 1 Step 100 (Global: 100): loss=1.9869, ppl=7.29, grad_norm=1.21, lr=1.86e-05 | |
| 2025-11-12 22:43:26,004 - INFO - Epoch 1 Step 110 (Global: 110): loss=1.9076, ppl=6.74, grad_norm=1.15, lr=1.95e-05 | |
| 2025-11-12 22:45:17,680 - INFO - Epoch 1 Step 120 (Global: 120): loss=1.9280, ppl=6.88, grad_norm=1.18, lr=2.04e-05 | |
| 2025-11-12 22:47:09,919 - INFO - Epoch 1 Step 130 (Global: 130): loss=1.8430, ppl=6.32, grad_norm=1.15, lr=2.12e-05 | |
| 2025-11-12 22:49:13,038 - INFO - Epoch 1 Step 140 (Global: 140): loss=1.9527, ppl=7.05, grad_norm=1.21, lr=2.21e-05 | |
| 2025-11-12 22:51:03,985 - INFO - Epoch 1 Step 150 (Global: 150): loss=1.6600, ppl=5.26, grad_norm=1.21, lr=2.30e-05 | |
| 2025-11-12 22:52:55,175 - INFO - Epoch 1 Step 160 (Global: 160): loss=1.9091, ppl=6.75, grad_norm=1.76, lr=2.38e-05 | |
| 2025-11-12 22:54:57,588 - INFO - Epoch 1 Step 170 (Global: 170): loss=1.8439, ppl=6.32, grad_norm=1.20, lr=2.47e-05 | |
| 2025-11-12 22:56:48,661 - INFO - Epoch 1 Step 180 (Global: 180): loss=1.9032, ppl=6.71, grad_norm=1.44, lr=2.56e-05 | |
| 2025-11-12 22:58:40,075 - INFO - Epoch 1 Step 190 (Global: 190): loss=2.0463, ppl=7.74, grad_norm=1.23, lr=2.64e-05 | |
| 2025-11-12 23:00:31,458 - INFO - Epoch 1 Step 200 (Global: 200): loss=1.7404, ppl=5.70, grad_norm=1.20, lr=2.73e-05 | |
| 2025-11-12 23:02:34,559 - INFO - Epoch 1 Step 210 (Global: 210): loss=2.0333, ppl=7.64, grad_norm=1.22, lr=2.82e-05 | |
| 2025-11-12 23:04:26,690 - INFO - Epoch 1 Step 220 (Global: 220): loss=2.0188, ppl=7.53, grad_norm=1.17, lr=2.90e-05 | |
| 2025-11-12 23:06:18,862 - INFO - Epoch 1 Step 230 (Global: 230): loss=1.9978, ppl=7.37, grad_norm=1.33, lr=2.99e-05 | |
| 2025-11-12 23:08:10,070 - INFO - Epoch 1 Step 240 (Global: 240): loss=1.9457, ppl=7.00, grad_norm=1.17, lr=3.07e-05 | |
| 2025-11-12 23:10:13,462 - INFO - Epoch 1 Step 250 (Global: 250): loss=1.8092, ppl=6.11, grad_norm=1.25, lr=3.16e-05 | |
| 2025-11-12 23:12:05,100 - INFO - Epoch 1 Step 260 (Global: 260): loss=1.8551, ppl=6.39, grad_norm=1.22, lr=3.25e-05 | |
| 2025-11-12 23:13:57,873 - INFO - Epoch 1 Step 270 (Global: 270): loss=2.0157, ppl=7.51, grad_norm=1.20, lr=3.33e-05 | |
| 2025-11-12 23:15:50,405 - INFO - Epoch 1 Step 280 (Global: 280): loss=1.8884, ppl=6.61, grad_norm=1.18, lr=3.42e-05 | |
| 2025-11-12 23:17:53,968 - INFO - Epoch 1 Step 290 (Global: 290): loss=1.8199, ppl=6.17, grad_norm=1.21, lr=3.51e-05 | |
| 2025-11-12 23:19:47,122 - INFO - Epoch 1 Step 300 (Global: 300): loss=2.1153, ppl=8.29, grad_norm=1.23, lr=3.59e-05 | |
| 2025-11-12 23:21:40,654 - INFO - Epoch 1 Step 310 (Global: 310): loss=1.8862, ppl=6.59, grad_norm=1.20, lr=3.68e-05 | |
| 2025-11-12 23:23:33,726 - INFO - Epoch 1 Step 320 (Global: 320): loss=1.7235, ppl=5.60, grad_norm=1.27, lr=3.77e-05 | |
| 2025-11-12 23:25:36,784 - INFO - Epoch 1 Step 330 (Global: 330): loss=1.9471, ppl=7.01, grad_norm=1.25, lr=3.85e-05 | |
| 2025-11-12 23:27:28,119 - INFO - Epoch 1 Step 340 (Global: 340): loss=1.7926, ppl=6.01, grad_norm=1.24, lr=3.94e-05 | |
| 2025-11-12 23:29:20,049 - INFO - Epoch 1 Step 350 (Global: 350): loss=1.9033, ppl=6.71, grad_norm=1.27, lr=4.03e-05 | |
| 2025-11-12 23:31:12,624 - INFO - Epoch 1 Step 360 (Global: 360): loss=1.7523, ppl=5.77, grad_norm=1.45, lr=4.11e-05 | |
| 2025-11-12 23:33:15,051 - INFO - Epoch 1 Step 370 (Global: 370): loss=1.8177, ppl=6.16, grad_norm=1.16, lr=4.20e-05 | |
| 2025-11-12 23:35:07,227 - INFO - Epoch 1 Step 380 (Global: 380): loss=1.7610, ppl=5.82, grad_norm=1.08, lr=4.29e-05 | |
| 2025-11-12 23:36:58,611 - INFO - Epoch 1 Step 390 (Global: 390): loss=2.0987, ppl=8.16, grad_norm=1.13, lr=4.37e-05 | |
| 2025-11-12 23:38:49,738 - INFO - Epoch 1 Step 400 (Global: 400): loss=1.9350, ppl=6.92, grad_norm=1.23, lr=4.46e-05 | |
| 2025-11-12 23:40:52,929 - INFO - Epoch 1 Step 410 (Global: 410): loss=1.8400, ppl=6.30, grad_norm=1.22, lr=4.54e-05 | |
| 2025-11-12 23:42:43,715 - INFO - Epoch 1 Step 420 (Global: 420): loss=1.8107, ppl=6.11, grad_norm=1.23, lr=4.63e-05 | |
| 2025-11-12 23:44:35,167 - INFO - Epoch 1 Step 430 (Global: 430): loss=1.8121, ppl=6.12, grad_norm=1.41, lr=4.72e-05 | |
| 2025-11-12 23:46:26,155 - INFO - Epoch 1 Step 440 (Global: 440): loss=1.9710, ppl=7.18, grad_norm=1.32, lr=4.80e-05 | |
| 2025-11-12 23:48:27,452 - INFO - Epoch 1 Step 450 (Global: 450): loss=1.9378, ppl=6.94, grad_norm=1.14, lr=4.89e-05 | |
| 2025-11-12 23:50:17,916 - INFO - Epoch 1 Step 460 (Global: 460): loss=1.7603, ppl=5.81, grad_norm=1.39, lr=4.98e-05 | |
| 2025-11-12 23:52:08,630 - INFO - Epoch 1 Step 470 (Global: 470): loss=1.9770, ppl=7.22, grad_norm=1.30, lr=5.06e-05 | |
| 2025-11-12 23:53:59,743 - INFO - Epoch 1 Step 480 (Global: 480): loss=1.8129, ppl=6.13, grad_norm=1.16, lr=5.15e-05 | |
| 2025-11-12 23:56:00,533 - INFO - Epoch 1 Step 490 (Global: 490): loss=1.7722, ppl=5.88, grad_norm=1.46, lr=5.24e-05 | |
| 2025-11-12 23:57:51,566 - INFO - Epoch 1 Step 500 (Global: 500): loss=1.9131, ppl=6.77, grad_norm=1.39, lr=5.32e-05 | |
| 2025-11-12 23:57:51,570 - INFO - | |
| Running validation at step 500... | |
| 2025-11-13 00:03:31,723 - INFO - Validation loss: 1.8813, perplexity: 6.56 | |
| 2025-11-13 00:03:31,724 - INFO - Qualitative metrics (n=5): | |
| 2025-11-13 00:03:31,724 - INFO - BLEU: 0.0000 | |
| 2025-11-13 00:03:31,724 - INFO - METEOR: 0.1212 | |
| 2025-11-13 00:03:31,724 - INFO - Edit Distance: 0.8026 | |
| 2025-11-13 00:03:31,724 - INFO - F-measure: 0.0717 | |
| 2025-11-13 00:03:31,724 - INFO - | |
| ====================================================================== | |
| 2025-11-13 00:03:31,724 - INFO - Qualitative Evaluation Samples: | |
| 2025-11-13 00:03:31,725 - INFO - ====================================================================== | |
| 2025-11-13 00:03:31,725 - INFO - | |
| Sample 1 (ID: sample_141920_chunk_1): | |
| 2025-11-13 00:03:31,725 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 00:03:31,725 - INFO - Generated: '# The Last of Us (2013 TV series)\nThe Last of Us is an American post-apocalyptic action-adventure television series created by Craig Mazin and directed by Neil Druckmann. The series is based on the 20...' | |
| 2025-11-13 00:03:31,725 - INFO - Ground Truth: 'Q gave it four stars out of five and said that "Perhaps [the album\'s] seemingly illogical sequencing of songs makes sense if they wish to lure their audience into thinking it\'s as-you-were. But it\'s n...' | |
| 2025-11-13 00:03:31,725 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 00:03:31,725 - INFO - | |
| Sample 2 (ID: sample_170543_chunk_2): | |
| 2025-11-13 00:03:31,725 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 00:03:31,725 - INFO - Generated: '# Social media in education\nSocial media in education is the use of social media platforms in educational settings. Social media has become a part of the lives of many students, and it is increasingly...' | |
| 2025-11-13 00:03:31,725 - INFO - Ground Truth: ', was Sirene Abou-Chakra, a Lebanese-American student who led the Arab-American Student Association. Other members included the woman president of the Michigan Student Assembly; the leader of Army ROT...' | |
| 2025-11-13 00:03:31,726 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 00:03:31,726 - INFO - | |
| Sample 3 (ID: sample_107152_chunk_9): | |
| 2025-11-13 00:03:31,726 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 00:03:31,726 - INFO - Generated: '# The Last of Us (2013 film)\nThe Last of Us is a 2013 American post-apocalyptic action-adventure film directed by Neil Druckmann and written by Ehren Kruger and David S. Goyer. The film stars Pedro Pa...' | |
| 2025-11-13 00:03:31,726 - INFO - Ground Truth: ' at the meeting Laymia headed. His weapon of choice is a giant ax, and he has the power to immobilise his opponents if they look at him in the eye. Oga falls for the trick, but Beel stops the ax and b...' | |
| 2025-11-13 00:03:31,726 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 00:03:31,726 - INFO - | |
| Sample 4 (ID: sample_069148_chunk_0): | |
| 2025-11-13 00:03:31,726 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 00:03:31,726 - INFO - Generated: ' | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 |...' | |
| 2025-11-13 00:03:31,726 - INFO - Ground Truth: '# Oriya (Unicode block)\nOriya is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0B01.....' | |
| 2025-11-13 00:03:31,727 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 00:03:31,727 - INFO - | |
| Sample 5 (ID: sample_103176_chunk_4): | |
| 2025-11-13 00:03:31,727 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 00:03:31,727 - INFO - Generated: ' | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |...' | |
| 2025-11-13 00:03:31,727 - INFO - Ground Truth: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Maxis Redwood Shores ...' | |
| 2025-11-13 00:03:31,727 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 00:03:31,728 - INFO - | |
| Qualitative samples saved to: outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/qualitative_step_500.jsonl | |
| 2025-11-13 00:04:08,873 - INFO - Saved checkpoint to outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/best_checkpoint.pt | |
| 2025-11-13 00:04:08,880 - INFO - New best validation loss: 1.8813, perplexity: 6.56 | |
| 2025-11-13 00:06:02,414 - INFO - Epoch 1 Step 510 (Global: 510): loss=2.0121, ppl=7.48, grad_norm=1.19, lr=5.41e-05 | |
| 2025-11-13 00:08:05,319 - INFO - Epoch 1 Step 520 (Global: 520): loss=1.7647, ppl=5.84, grad_norm=1.67, lr=5.50e-05 | |
| 2025-11-13 00:09:57,403 - INFO - Epoch 1 Step 530 (Global: 530): loss=1.8288, ppl=6.23, grad_norm=1.31, lr=5.58e-05 | |
| 2025-11-13 00:11:49,413 - INFO - Epoch 1 Step 540 (Global: 540): loss=1.8623, ppl=6.44, grad_norm=1.17, lr=5.67e-05 | |
| 2025-11-13 00:13:41,569 - INFO - Epoch 1 Step 550 (Global: 550): loss=1.8152, ppl=6.14, grad_norm=1.30, lr=5.76e-05 | |
| 2025-11-13 00:15:43,832 - INFO - Epoch 1 Step 560 (Global: 560): loss=1.7235, ppl=5.60, grad_norm=1.20, lr=5.84e-05 | |
| 2025-11-13 00:17:35,140 - INFO - Epoch 1 Step 570 (Global: 570): loss=1.8206, ppl=6.18, grad_norm=1.33, lr=5.93e-05 | |
| 2025-11-13 00:19:25,728 - INFO - Epoch 1 Step 580 (Global: 580): loss=1.8273, ppl=6.22, grad_norm=1.16, lr=6.01e-05 | |
| 2025-11-13 00:21:16,685 - INFO - Epoch 1 Step 590 (Global: 590): loss=1.8464, ppl=6.34, grad_norm=1.27, lr=6.10e-05 | |
| 2025-11-13 00:23:18,276 - INFO - Epoch 1 Step 600 (Global: 600): loss=1.8402, ppl=6.30, grad_norm=1.23, lr=6.19e-05 | |
| 2025-11-13 00:25:09,151 - INFO - Epoch 1 Step 610 (Global: 610): loss=1.7067, ppl=5.51, grad_norm=1.12, lr=6.27e-05 | |
| 2025-11-13 00:27:00,416 - INFO - Epoch 1 Step 620 (Global: 620): loss=1.6940, ppl=5.44, grad_norm=1.22, lr=6.36e-05 | |
| 2025-11-13 00:28:52,034 - INFO - Epoch 1 Step 630 (Global: 630): loss=1.9519, ppl=7.04, grad_norm=1.18, lr=6.45e-05 | |
| 2025-11-13 00:30:54,158 - INFO - Epoch 1 Step 640 (Global: 640): loss=1.8931, ppl=6.64, grad_norm=1.20, lr=6.53e-05 | |
| 2025-11-13 00:32:45,582 - INFO - Epoch 1 Step 650 (Global: 650): loss=1.9499, ppl=7.03, grad_norm=1.20, lr=6.62e-05 | |
| 2025-11-13 00:34:37,416 - INFO - Epoch 1 Step 660 (Global: 660): loss=1.8992, ppl=6.68, grad_norm=1.34, lr=6.71e-05 | |
| 2025-11-13 00:36:28,943 - INFO - Epoch 1 Step 670 (Global: 670): loss=1.7955, ppl=6.02, grad_norm=1.12, lr=6.79e-05 | |
| 2025-11-13 00:38:30,706 - INFO - Epoch 1 Step 680 (Global: 680): loss=1.6596, ppl=5.26, grad_norm=1.25, lr=6.88e-05 | |
| 2025-11-13 00:40:21,745 - INFO - Epoch 1 Step 690 (Global: 690): loss=1.9247, ppl=6.85, grad_norm=1.17, lr=6.97e-05 | |
| 2025-11-13 00:42:12,701 - INFO - Epoch 1 Step 700 (Global: 700): loss=1.8220, ppl=6.18, grad_norm=1.09, lr=7.05e-05 | |
| 2025-11-13 00:44:04,289 - INFO - Epoch 1 Step 710 (Global: 710): loss=1.9799, ppl=7.24, grad_norm=1.19, lr=7.14e-05 | |
| 2025-11-13 00:46:06,029 - INFO - Epoch 1 Step 720 (Global: 720): loss=1.7438, ppl=5.72, grad_norm=1.52, lr=7.22e-05 | |
| 2025-11-13 00:47:57,665 - INFO - Epoch 1 Step 730 (Global: 730): loss=1.8625, ppl=6.44, grad_norm=1.10, lr=7.31e-05 | |
| 2025-11-13 00:49:49,307 - INFO - Epoch 1 Step 740 (Global: 740): loss=1.9008, ppl=6.69, grad_norm=1.14, lr=7.40e-05 | |
| 2025-11-13 00:51:41,165 - INFO - Epoch 1 Step 750 (Global: 750): loss=1.9207, ppl=6.83, grad_norm=1.34, lr=7.48e-05 | |
| 2025-11-13 00:53:42,841 - INFO - Epoch 1 Step 760 (Global: 760): loss=1.8712, ppl=6.50, grad_norm=1.16, lr=7.57e-05 | |
| 2025-11-13 00:55:33,901 - INFO - Epoch 1 Step 770 (Global: 770): loss=1.9525, ppl=7.05, grad_norm=1.27, lr=7.66e-05 | |
| 2025-11-13 00:57:25,328 - INFO - Epoch 1 Step 780 (Global: 780): loss=1.9585, ppl=7.09, grad_norm=1.16, lr=7.74e-05 | |
| 2025-11-13 00:59:16,266 - INFO - Epoch 1 Step 790 (Global: 790): loss=1.9453, ppl=7.00, grad_norm=1.15, lr=7.83e-05 | |
| 2025-11-13 01:01:17,832 - INFO - Epoch 1 Step 800 (Global: 800): loss=1.8022, ppl=6.06, grad_norm=1.27, lr=7.92e-05 | |
| 2025-11-13 01:03:08,705 - INFO - Epoch 1 Step 810 (Global: 810): loss=1.9679, ppl=7.16, grad_norm=1.42, lr=8.00e-05 | |
| 2025-11-13 01:05:00,084 - INFO - Epoch 1 Step 820 (Global: 820): loss=1.8450, ppl=6.33, grad_norm=1.23, lr=8.09e-05 | |
| 2025-11-13 01:06:51,460 - INFO - Epoch 1 Step 830 (Global: 830): loss=2.0058, ppl=7.43, grad_norm=1.10, lr=8.18e-05 | |
| 2025-11-13 01:08:53,721 - INFO - Epoch 1 Step 840 (Global: 840): loss=1.7684, ppl=5.86, grad_norm=1.21, lr=8.26e-05 | |
| 2025-11-13 01:10:45,511 - INFO - Epoch 1 Step 850 (Global: 850): loss=1.9236, ppl=6.85, grad_norm=1.09, lr=8.35e-05 | |
| 2025-11-13 01:12:36,448 - INFO - Epoch 1 Step 860 (Global: 860): loss=2.0211, ppl=7.55, grad_norm=1.15, lr=8.44e-05 | |
| 2025-11-13 01:14:27,006 - INFO - Epoch 1 Step 870 (Global: 870): loss=1.8940, ppl=6.65, grad_norm=1.16, lr=8.52e-05 | |
| 2025-11-13 01:16:28,655 - INFO - Epoch 1 Step 880 (Global: 880): loss=1.7154, ppl=5.56, grad_norm=1.08, lr=8.61e-05 | |
| 2025-11-13 01:18:19,640 - INFO - Epoch 1 Step 890 (Global: 890): loss=1.7897, ppl=5.99, grad_norm=1.23, lr=8.69e-05 | |
| 2025-11-13 01:20:11,356 - INFO - Epoch 1 Step 900 (Global: 900): loss=1.9062, ppl=6.73, grad_norm=1.18, lr=8.78e-05 | |
| 2025-11-13 01:22:02,899 - INFO - Epoch 1 Step 910 (Global: 910): loss=1.9204, ppl=6.82, grad_norm=1.13, lr=8.87e-05 | |
| 2025-11-13 01:24:03,384 - INFO - Epoch 1 Step 920 (Global: 920): loss=2.0367, ppl=7.67, grad_norm=1.15, lr=8.95e-05 | |
| 2025-11-13 01:25:54,155 - INFO - Epoch 1 Step 930 (Global: 930): loss=1.9313, ppl=6.90, grad_norm=1.12, lr=9.04e-05 | |
| 2025-11-13 01:27:45,415 - INFO - Epoch 1 Step 940 (Global: 940): loss=1.8801, ppl=6.55, grad_norm=1.07, lr=9.13e-05 | |
| 2025-11-13 01:29:36,595 - INFO - Epoch 1 Step 950 (Global: 950): loss=1.8094, ppl=6.11, grad_norm=1.20, lr=9.21e-05 | |
| 2025-11-13 01:31:37,797 - INFO - Epoch 1 Step 960 (Global: 960): loss=1.8444, ppl=6.32, grad_norm=1.11, lr=9.30e-05 | |
| 2025-11-13 01:33:29,646 - INFO - Epoch 1 Step 970 (Global: 970): loss=2.0079, ppl=7.45, grad_norm=1.12, lr=9.39e-05 | |
| 2025-11-13 01:35:20,703 - INFO - Epoch 1 Step 980 (Global: 980): loss=1.9965, ppl=7.36, grad_norm=1.12, lr=9.47e-05 | |
| 2025-11-13 01:37:11,902 - INFO - Epoch 1 Step 990 (Global: 990): loss=1.8224, ppl=6.19, grad_norm=1.22, lr=9.56e-05 | |
| 2025-11-13 01:39:13,648 - INFO - Epoch 1 Step 1000 (Global: 1000): loss=1.9958, ppl=7.36, grad_norm=1.24, lr=9.65e-05 | |
| 2025-11-13 01:39:13,651 - INFO - | |
| Running validation at step 1000... | |
| 2025-11-13 01:44:44,012 - INFO - Validation loss: 1.9032, perplexity: 6.71 | |
| 2025-11-13 01:44:44,013 - INFO - Qualitative metrics (n=5): | |
| 2025-11-13 01:44:44,013 - INFO - BLEU: 0.0346 | |
| 2025-11-13 01:44:44,013 - INFO - METEOR: 0.1377 | |
| 2025-11-13 01:44:44,013 - INFO - Edit Distance: 0.6809 | |
| 2025-11-13 01:44:44,013 - INFO - F-measure: 0.1298 | |
| 2025-11-13 01:44:44,013 - INFO - | |
| ====================================================================== | |
| 2025-11-13 01:44:44,013 - INFO - Qualitative Evaluation Samples: | |
| 2025-11-13 01:44:44,013 - INFO - ====================================================================== | |
| 2025-11-13 01:44:44,014 - INFO - | |
| Sample 1 (ID: sample_141920_chunk_1): | |
| 2025-11-13 01:44:44,014 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 01:44:44,014 - INFO - Generated: ", and the two are in a relationship. The film's ending is ambiguous, with the characters' fate unknown. The film's ending is ambiguous, with the characters' fate unknown. The film's ending is ambiguou..." | |
| 2025-11-13 01:44:44,014 - INFO - Ground Truth: 'Q gave it four stars out of five and said that "Perhaps [the album\'s] seemingly illogical sequencing of songs makes sense if they wish to lure their audience into thinking it\'s as-you-were. But it\'s n...' | |
| 2025-11-13 01:44:44,014 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 01:44:44,014 - INFO - | |
| Sample 2 (ID: sample_170543_chunk_2): | |
| 2025-11-13 01:44:44,014 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 01:44:44,014 - INFO - Generated: '# 2008 New York State Senate election\nThe 2008 New York State Senate election was held on November 4, 2008, to elect 48 members of the New York State Senate. The election coincided with the 2008 Unite...' | |
| 2025-11-13 01:44:44,014 - INFO - Ground Truth: ', was Sirene Abou-Chakra, a Lebanese-American student who led the Arab-American Student Association. Other members included the woman president of the Michigan Student Assembly; the leader of Army ROT...' | |
| 2025-11-13 01:44:44,014 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 01:44:44,014 - INFO - | |
| Sample 3 (ID: sample_107152_chunk_9): | |
| 2025-11-13 01:44:44,014 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 01:44:44,015 - INFO - Generated: ', and the two were both killed. The next day, the two were found in a room, and the one who had been killed was identified as the one who had killed the other. The other one was identified as the one ...' | |
| 2025-11-13 01:44:44,015 - INFO - Ground Truth: ' at the meeting Laymia headed. His weapon of choice is a giant ax, and he has the power to immobilise his opponents if they look at him in the eye. Oga falls for the trick, but Beel stops the ax and b...' | |
| 2025-11-13 01:44:44,015 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 01:44:44,015 - INFO - | |
| Sample 4 (ID: sample_069148_chunk_0): | |
| 2025-11-13 01:44:44,015 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 01:44:44,015 - INFO - Generated: '# Yonaguni language\nYonaguni (ヨナグリ, Yonaguri) is a language isolate spoken in the Yonaguni Islands, Japan. It is not related to the Japanese language, but is instead related to the Ryukyuan languages....' | |
| 2025-11-13 01:44:44,015 - INFO - Ground Truth: '# Oriya (Unicode block)\nOriya is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0B01.....' | |
| 2025-11-13 01:44:44,015 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 01:44:44,015 - INFO - | |
| Sample 5 (ID: sample_103176_chunk_4): | |
| 2025-11-13 01:44:44,015 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 01:44:44,015 - INFO - Generated: ' | 2017 | [ 1 ] |\n| 2018 | The Last of Us (2017) | The Last of Us (2017) | The Last of Us (2017) | The ...' | |
| 2025-11-13 01:44:44,016 - INFO - Ground Truth: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Maxis Redwood Shores ...' | |
| 2025-11-13 01:44:44,016 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 01:44:44,016 - INFO - | |
| Qualitative samples saved to: outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/qualitative_step_1000.jsonl | |
| 2025-11-13 01:46:36,108 - INFO - Epoch 1 Step 1010 (Global: 1010): loss=1.9929, ppl=7.34, grad_norm=1.20, lr=9.73e-05 | |
| 2025-11-13 01:48:27,696 - INFO - Epoch 1 Step 1020 (Global: 1020): loss=1.9470, ppl=7.01, grad_norm=1.12, lr=9.82e-05 | |
| 2025-11-13 01:50:19,434 - INFO - Epoch 1 Step 1030 (Global: 1030): loss=1.9866, ppl=7.29, grad_norm=1.19, lr=9.90e-05 | |
| 2025-11-13 01:52:21,679 - INFO - Epoch 1 Step 1040 (Global: 1040): loss=1.8824, ppl=6.57, grad_norm=1.02, lr=9.99e-05 | |
| 2025-11-13 01:54:13,294 - INFO - Epoch 1 Step 1050 (Global: 1050): loss=2.0455, ppl=7.73, grad_norm=1.04, lr=1.00e-04 | |
| 2025-11-13 01:56:05,029 - INFO - Epoch 1 Step 1060 (Global: 1060): loss=1.9666, ppl=7.15, grad_norm=1.27, lr=1.00e-04 | |
| 2025-11-13 01:58:07,115 - INFO - Epoch 1 Step 1070 (Global: 1070): loss=1.8186, ppl=6.16, grad_norm=1.09, lr=1.00e-04 | |
| 2025-11-13 01:59:58,282 - INFO - Epoch 1 Step 1080 (Global: 1080): loss=1.8018, ppl=6.06, grad_norm=1.15, lr=1.00e-04 | |
| 2025-11-13 02:01:49,362 - INFO - Epoch 1 Step 1090 (Global: 1090): loss=1.8755, ppl=6.52, grad_norm=1.04, lr=1.00e-04 | |
| 2025-11-13 02:03:40,803 - INFO - Epoch 1 Step 1100 (Global: 1100): loss=1.8711, ppl=6.50, grad_norm=1.06, lr=1.00e-04 | |
| 2025-11-13 02:05:42,282 - INFO - Epoch 1 Step 1110 (Global: 1110): loss=1.9739, ppl=7.20, grad_norm=1.01, lr=1.00e-04 | |
| 2025-11-13 02:07:33,350 - INFO - Epoch 1 Step 1120 (Global: 1120): loss=1.7106, ppl=5.53, grad_norm=1.02, lr=1.00e-04 | |
| 2025-11-13 02:09:24,952 - INFO - Epoch 1 Step 1130 (Global: 1130): loss=2.1331, ppl=8.44, grad_norm=1.02, lr=1.00e-04 | |
| 2025-11-13 02:11:16,725 - INFO - Epoch 1 Step 1140 (Global: 1140): loss=1.8139, ppl=6.13, grad_norm=1.04, lr=1.00e-04 | |
| 2025-11-13 02:13:20,323 - INFO - Epoch 1 Step 1150 (Global: 1150): loss=2.0205, ppl=7.54, grad_norm=1.01, lr=1.00e-04 | |
| 2025-11-13 02:15:13,237 - INFO - Epoch 1 Step 1160 (Global: 1160): loss=1.8583, ppl=6.41, grad_norm=1.05, lr=1.00e-04 | |
| 2025-11-13 02:17:05,994 - INFO - Epoch 1 Step 1170 (Global: 1170): loss=1.8398, ppl=6.30, grad_norm=1.00, lr=1.00e-04 | |
| 2025-11-13 02:18:57,658 - INFO - Epoch 1 Step 1180 (Global: 1180): loss=1.8558, ppl=6.40, grad_norm=1.75, lr=9.99e-05 | |
| 2025-11-13 02:20:59,615 - INFO - Epoch 1 Step 1190 (Global: 1190): loss=1.8091, ppl=6.10, grad_norm=1.04, lr=9.99e-05 | |
| 2025-11-13 02:22:51,043 - INFO - Epoch 1 Step 1200 (Global: 1200): loss=1.9491, ppl=7.02, grad_norm=0.98, lr=9.99e-05 | |
| 2025-11-13 02:24:42,378 - INFO - Epoch 1 Step 1210 (Global: 1210): loss=1.8678, ppl=6.47, grad_norm=1.03, lr=9.99e-05 | |
| 2025-11-13 02:26:33,446 - INFO - Epoch 1 Step 1220 (Global: 1220): loss=1.8505, ppl=6.36, grad_norm=0.96, lr=9.99e-05 | |
| 2025-11-13 02:28:35,403 - INFO - Epoch 1 Step 1230 (Global: 1230): loss=1.8865, ppl=6.60, grad_norm=1.02, lr=9.99e-05 | |
| 2025-11-13 02:30:26,541 - INFO - Epoch 1 Step 1240 (Global: 1240): loss=2.0060, ppl=7.43, grad_norm=0.99, lr=9.99e-05 | |
| 2025-11-13 02:32:17,443 - INFO - Epoch 1 Step 1250 (Global: 1250): loss=1.8374, ppl=6.28, grad_norm=1.20, lr=9.99e-05 | |
| 2025-11-13 02:34:08,459 - INFO - Epoch 1 Step 1260 (Global: 1260): loss=1.9084, ppl=6.74, grad_norm=1.16, lr=9.99e-05 | |
| 2025-11-13 02:36:10,583 - INFO - Epoch 1 Step 1270 (Global: 1270): loss=1.9128, ppl=6.77, grad_norm=1.16, lr=9.99e-05 | |
| 2025-11-13 02:38:01,731 - INFO - Epoch 1 Step 1280 (Global: 1280): loss=1.9391, ppl=6.95, grad_norm=1.05, lr=9.98e-05 | |
| 2025-11-13 02:39:52,815 - INFO - Epoch 1 Step 1290 (Global: 1290): loss=1.7412, ppl=5.70, grad_norm=0.96, lr=9.98e-05 | |
| 2025-11-13 02:41:43,604 - INFO - Epoch 1 Step 1300 (Global: 1300): loss=1.7858, ppl=5.96, grad_norm=0.98, lr=9.98e-05 | |
| 2025-11-13 02:43:44,813 - INFO - Epoch 1 Step 1310 (Global: 1310): loss=1.9432, ppl=6.98, grad_norm=1.09, lr=9.98e-05 | |
| 2025-11-13 02:45:36,384 - INFO - Epoch 1 Step 1320 (Global: 1320): loss=1.8976, ppl=6.67, grad_norm=1.08, lr=9.98e-05 | |
| 2025-11-13 02:47:28,139 - INFO - Epoch 1 Step 1330 (Global: 1330): loss=1.9001, ppl=6.69, grad_norm=1.00, lr=9.98e-05 | |
| 2025-11-13 02:49:19,503 - INFO - Epoch 1 Step 1340 (Global: 1340): loss=1.8785, ppl=6.54, grad_norm=0.94, lr=9.97e-05 | |
| 2025-11-13 02:51:21,130 - INFO - Epoch 1 Step 1350 (Global: 1350): loss=1.7815, ppl=5.94, grad_norm=1.05, lr=9.97e-05 | |
| 2025-11-13 02:53:12,250 - INFO - Epoch 1 Step 1360 (Global: 1360): loss=1.7453, ppl=5.73, grad_norm=1.00, lr=9.97e-05 | |
| 2025-11-13 02:55:03,116 - INFO - Epoch 1 Step 1370 (Global: 1370): loss=1.9978, ppl=7.37, grad_norm=1.11, lr=9.97e-05 | |
| 2025-11-13 02:56:53,603 - INFO - Epoch 1 Step 1380 (Global: 1380): loss=1.8753, ppl=6.52, grad_norm=1.00, lr=9.97e-05 | |
| 2025-11-13 02:58:54,742 - INFO - Epoch 1 Step 1390 (Global: 1390): loss=1.6761, ppl=5.34, grad_norm=1.01, lr=9.97e-05 | |
| 2025-11-13 03:00:45,769 - INFO - Epoch 1 Step 1400 (Global: 1400): loss=1.9684, ppl=7.16, grad_norm=1.09, lr=9.96e-05 | |
| 2025-11-13 03:02:36,735 - INFO - Epoch 1 Step 1410 (Global: 1410): loss=2.0746, ppl=7.96, grad_norm=0.99, lr=9.96e-05 | |
| 2025-11-13 03:04:27,877 - INFO - Epoch 1 Step 1420 (Global: 1420): loss=1.9360, ppl=6.93, grad_norm=0.95, lr=9.96e-05 | |
| 2025-11-13 03:06:30,340 - INFO - Epoch 1 Step 1430 (Global: 1430): loss=1.7456, ppl=5.73, grad_norm=0.99, lr=9.96e-05 | |
| 2025-11-13 03:08:21,832 - INFO - Epoch 1 Step 1440 (Global: 1440): loss=1.7830, ppl=5.95, grad_norm=0.96, lr=9.96e-05 | |
| 2025-11-13 03:10:13,618 - INFO - Epoch 1 Step 1450 (Global: 1450): loss=1.7797, ppl=5.93, grad_norm=0.96, lr=9.95e-05 | |
| 2025-11-13 03:12:05,088 - INFO - Epoch 1 Step 1460 (Global: 1460): loss=1.7987, ppl=6.04, grad_norm=0.96, lr=9.95e-05 | |
| 2025-11-13 03:14:08,028 - INFO - Epoch 1 Step 1470 (Global: 1470): loss=1.9114, ppl=6.76, grad_norm=1.03, lr=9.95e-05 | |
| 2025-11-13 03:15:59,376 - INFO - Epoch 1 Step 1480 (Global: 1480): loss=1.7792, ppl=5.92, grad_norm=0.95, lr=9.95e-05 | |
| 2025-11-13 03:17:51,256 - INFO - Epoch 1 Step 1490 (Global: 1490): loss=1.7837, ppl=5.95, grad_norm=1.02, lr=9.94e-05 | |
| 2025-11-13 03:19:43,784 - INFO - Epoch 1 Step 1500 (Global: 1500): loss=1.8345, ppl=6.26, grad_norm=1.00, lr=9.94e-05 | |
| 2025-11-13 03:19:43,788 - INFO - | |
| Running validation at step 1500... | |
| 2025-11-13 03:25:23,969 - INFO - Validation loss: 1.8753, perplexity: 6.52 | |
| 2025-11-13 03:25:23,970 - INFO - Qualitative metrics (n=5): | |
| 2025-11-13 03:25:23,970 - INFO - BLEU: 0.0047 | |
| 2025-11-13 03:25:23,970 - INFO - METEOR: 0.0948 | |
| 2025-11-13 03:25:23,970 - INFO - Edit Distance: 0.7631 | |
| 2025-11-13 03:25:23,970 - INFO - F-measure: 0.1571 | |
| 2025-11-13 03:25:23,970 - INFO - | |
| ====================================================================== | |
| 2025-11-13 03:25:23,970 - INFO - Qualitative Evaluation Samples: | |
| 2025-11-13 03:25:23,970 - INFO - ====================================================================== | |
| 2025-11-13 03:25:23,970 - INFO - | |
| Sample 1 (ID: sample_141920_chunk_1): | |
| 2025-11-13 03:25:23,971 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 03:25:23,971 - INFO - Generated: ' that five of the seven stars had been "fired" for the first time. The song\'s lyrics are about a woman who is "tired of being tired" and "tired of being a woman". The song\'s title is a reference to th...' | |
| 2025-11-13 03:25:23,971 - INFO - Ground Truth: 'Q gave it four stars out of five and said that "Perhaps [the album\'s] seemingly illogical sequencing of songs makes sense if they wish to lure their audience into thinking it\'s as-you-were. But it\'s n...' | |
| 2025-11-13 03:25:23,971 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 03:25:23,971 - INFO - | |
| Sample 2 (ID: sample_170543_chunk_2): | |
| 2025-11-13 03:25:23,971 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 03:25:23,971 - INFO - Generated: ', A.B. in Lebanese-American and Arab-American studies, and a member of the Phi Beta Kappa honor society. She was also a member of the Phi Beta Kappa honor society in her senior year of high school. Sh...' | |
| 2025-11-13 03:25:23,971 - INFO - Ground Truth: ', was Sirene Abou-Chakra, a Lebanese-American student who led the Arab-American Student Association. Other members included the woman president of the Michigan Student Assembly; the leader of Army ROT...' | |
| 2025-11-13 03:25:23,971 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 03:25:23,971 - INFO - | |
| Sample 3 (ID: sample_107152_chunk_9): | |
| 2025-11-13 03:25:23,971 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 03:25:23,972 - INFO - Generated: " the leader of the Titan's army, who was shot by the Hero. The Hero then killed the Titan, and the Hero's body was transformed into a giant, and the Hero's sword was transformed into a giant, and the ..." | |
| 2025-11-13 03:25:23,972 - INFO - Ground Truth: ' at the meeting Laymia headed. His weapon of choice is a giant ax, and he has the power to immobilise his opponents if they look at him in the eye. Oga falls for the trick, but Beel stops the ax and b...' | |
| 2025-11-13 03:25:23,972 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 03:25:23,972 - INFO - | |
| Sample 4 (ID: sample_069148_chunk_0): | |
| 2025-11-13 03:25:23,972 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 03:25:23,972 - INFO - Generated: '# Unicode block "Oriya"\nUnicode block "Oriya" is a block of ideographic characters in the Unicode standard, used for writing the Oriya language. It is defined in the Unicode Standard at the Unicode Co...' | |
| 2025-11-13 03:25:23,973 - INFO - Ground Truth: '# Oriya (Unicode block)\nOriya is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0B01.....' | |
| 2025-11-13 03:25:23,973 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 03:25:23,973 - INFO - | |
| Sample 5 (ID: sample_103176_chunk_4): | |
| 2025-11-13 03:25:23,974 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 03:25:23,974 - INFO - Generated: ' |\n| 3. | The Sims 3: The Sims 3: The Sims 3: The Sims 3: The Sims 3: The Sims 3: The Sims 3: The Sims 3: The Sims 3: The Sims 3: The Sims 3: The Sims 3: The Sims 3: The Sims 3: The Sims 3: The Sims 3...' | |
| 2025-11-13 03:25:23,974 - INFO - Ground Truth: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Maxis Redwood Shores ...' | |
| 2025-11-13 03:25:23,974 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 03:25:23,975 - INFO - | |
| Qualitative samples saved to: outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/qualitative_step_1500.jsonl | |
| 2025-11-13 03:26:09,872 - INFO - Saved checkpoint to outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/best_checkpoint.pt | |
| 2025-11-13 03:26:09,892 - INFO - New best validation loss: 1.8753, perplexity: 6.52 | |
| 2025-11-13 03:28:14,991 - INFO - Epoch 1 Step 1510 (Global: 1510): loss=2.0030, ppl=7.41, grad_norm=1.10, lr=9.94e-05 | |
| 2025-11-13 03:30:07,970 - INFO - Epoch 1 Step 1520 (Global: 1520): loss=1.9671, ppl=7.15, grad_norm=0.98, lr=9.94e-05 | |
| 2025-11-13 03:32:00,714 - INFO - Epoch 1 Step 1530 (Global: 1530): loss=1.7798, ppl=5.93, grad_norm=1.05, lr=9.93e-05 | |
| 2025-11-13 03:33:52,852 - INFO - Epoch 1 Step 1540 (Global: 1540): loss=1.6923, ppl=5.43, grad_norm=0.92, lr=9.93e-05 | |
| 2025-11-13 03:35:54,639 - INFO - Epoch 1 Step 1550 (Global: 1550): loss=1.7165, ppl=5.56, grad_norm=1.01, lr=9.93e-05 | |
| 2025-11-13 03:37:46,863 - INFO - Epoch 1 Step 1560 (Global: 1560): loss=1.7288, ppl=5.63, grad_norm=0.94, lr=9.92e-05 | |
| 2025-11-13 03:39:39,430 - INFO - Epoch 1 Step 1570 (Global: 1570): loss=1.8565, ppl=6.40, grad_norm=0.97, lr=9.92e-05 | |
| 2025-11-13 03:41:31,820 - INFO - Epoch 1 Step 1580 (Global: 1580): loss=1.7998, ppl=6.05, grad_norm=1.04, lr=9.92e-05 | |
| 2025-11-13 03:43:35,045 - INFO - Epoch 1 Step 1590 (Global: 1590): loss=2.0030, ppl=7.41, grad_norm=1.02, lr=9.92e-05 | |
| 2025-11-13 03:45:28,331 - INFO - Epoch 1 Step 1600 (Global: 1600): loss=1.8113, ppl=6.12, grad_norm=1.03, lr=9.91e-05 | |
| 2025-11-13 03:47:21,404 - INFO - Epoch 1 Step 1610 (Global: 1610): loss=1.7627, ppl=5.83, grad_norm=1.11, lr=9.91e-05 | |
| 2025-11-13 03:49:14,797 - INFO - Epoch 1 Step 1620 (Global: 1620): loss=1.7267, ppl=5.62, grad_norm=1.52, lr=9.91e-05 | |
| 2025-11-13 03:51:18,288 - INFO - Epoch 1 Step 1630 (Global: 1630): loss=1.6193, ppl=5.05, grad_norm=3.55, lr=9.90e-05 | |
| 2025-11-13 03:53:10,982 - INFO - Epoch 1 Step 1640 (Global: 1640): loss=1.4589, ppl=4.30, grad_norm=2.95, lr=9.90e-05 | |
| 2025-11-13 03:55:03,766 - INFO - Epoch 1 Step 1650 (Global: 1650): loss=1.3245, ppl=3.76, grad_norm=2.45, lr=9.90e-05 | |
| 2025-11-13 03:57:06,300 - INFO - Epoch 1 Step 1660 (Global: 1660): loss=1.1793, ppl=3.25, grad_norm=2.86, lr=9.89e-05 | |
| 2025-11-13 03:58:59,544 - INFO - Epoch 1 Step 1670 (Global: 1670): loss=0.9392, ppl=2.56, grad_norm=2.69, lr=9.89e-05 | |
| 2025-11-13 04:00:52,416 - INFO - Epoch 1 Step 1680 (Global: 1680): loss=0.9722, ppl=2.64, grad_norm=2.39, lr=9.89e-05 | |
| 2025-11-13 04:02:45,031 - INFO - Epoch 1 Step 1690 (Global: 1690): loss=0.8659, ppl=2.38, grad_norm=2.42, lr=9.88e-05 | |
| 2025-11-13 04:04:47,945 - INFO - Epoch 1 Step 1700 (Global: 1700): loss=0.8113, ppl=2.25, grad_norm=2.12, lr=9.88e-05 | |
| 2025-11-13 04:06:39,980 - INFO - Epoch 1 Step 1710 (Global: 1710): loss=0.8610, ppl=2.37, grad_norm=2.39, lr=9.87e-05 | |
| 2025-11-13 04:08:32,214 - INFO - Epoch 1 Step 1720 (Global: 1720): loss=0.7373, ppl=2.09, grad_norm=2.75, lr=9.87e-05 | |
| 2025-11-13 04:10:24,404 - INFO - Epoch 1 Step 1730 (Global: 1730): loss=0.7497, ppl=2.12, grad_norm=2.41, lr=9.87e-05 | |
| 2025-11-13 04:12:25,989 - INFO - Epoch 1 Step 1740 (Global: 1740): loss=0.7545, ppl=2.13, grad_norm=2.36, lr=9.86e-05 | |
| 2025-11-13 04:14:16,824 - INFO - Epoch 1 Step 1750 (Global: 1750): loss=0.6459, ppl=1.91, grad_norm=2.83, lr=9.86e-05 | |
| 2025-11-13 04:16:08,421 - INFO - Epoch 1 Step 1760 (Global: 1760): loss=0.6606, ppl=1.94, grad_norm=2.58, lr=9.86e-05 | |
| 2025-11-13 04:17:59,643 - INFO - Epoch 1 Step 1770 (Global: 1770): loss=0.6479, ppl=1.91, grad_norm=2.27, lr=9.85e-05 | |
| 2025-11-13 04:20:01,564 - INFO - Epoch 1 Step 1780 (Global: 1780): loss=0.6964, ppl=2.01, grad_norm=2.47, lr=9.85e-05 | |
| 2025-11-13 04:21:53,053 - INFO - Epoch 1 Step 1790 (Global: 1790): loss=0.5613, ppl=1.75, grad_norm=2.06, lr=9.84e-05 | |
| 2025-11-13 04:23:44,454 - INFO - Epoch 1 Step 1800 (Global: 1800): loss=0.5216, ppl=1.68, grad_norm=1.98, lr=9.84e-05 | |
| 2025-11-13 04:25:35,251 - INFO - Epoch 1 Step 1810 (Global: 1810): loss=0.6308, ppl=1.88, grad_norm=2.97, lr=9.83e-05 | |
| 2025-11-13 04:27:37,165 - INFO - Epoch 1 Step 1820 (Global: 1820): loss=0.5270, ppl=1.69, grad_norm=2.14, lr=9.83e-05 | |
| 2025-11-13 04:29:27,708 - INFO - Epoch 1 Step 1830 (Global: 1830): loss=0.4641, ppl=1.59, grad_norm=1.90, lr=9.83e-05 | |
| 2025-11-13 04:31:18,592 - INFO - Epoch 1 Step 1840 (Global: 1840): loss=0.4434, ppl=1.56, grad_norm=2.23, lr=9.82e-05 | |
| 2025-11-13 04:33:09,258 - INFO - Epoch 1 Step 1850 (Global: 1850): loss=0.5251, ppl=1.69, grad_norm=2.48, lr=9.82e-05 | |
| 2025-11-13 04:35:10,411 - INFO - Epoch 1 Step 1860 (Global: 1860): loss=0.4967, ppl=1.64, grad_norm=2.34, lr=9.81e-05 | |
| 2025-11-13 04:37:01,530 - INFO - Epoch 1 Step 1870 (Global: 1870): loss=0.4584, ppl=1.58, grad_norm=2.39, lr=9.81e-05 | |
| 2025-11-13 04:38:52,387 - INFO - Epoch 1 Step 1880 (Global: 1880): loss=0.4406, ppl=1.55, grad_norm=2.20, lr=9.80e-05 | |
| 2025-11-13 04:40:43,119 - INFO - Epoch 1 Step 1890 (Global: 1890): loss=0.4368, ppl=1.55, grad_norm=2.31, lr=9.80e-05 | |
| 2025-11-13 04:42:43,876 - INFO - Epoch 1 Step 1900 (Global: 1900): loss=0.4448, ppl=1.56, grad_norm=2.30, lr=9.79e-05 | |
| 2025-11-13 04:44:34,608 - INFO - Epoch 1 Step 1910 (Global: 1910): loss=0.4510, ppl=1.57, grad_norm=2.20, lr=9.79e-05 | |
| 2025-11-13 04:46:25,643 - INFO - Epoch 1 Step 1920 (Global: 1920): loss=0.4009, ppl=1.49, grad_norm=2.08, lr=9.78e-05 | |
| 2025-11-13 04:48:16,437 - INFO - Epoch 1 Step 1930 (Global: 1930): loss=0.3991, ppl=1.49, grad_norm=2.14, lr=9.78e-05 | |
| 2025-11-13 04:50:17,530 - INFO - Epoch 1 Step 1940 (Global: 1940): loss=0.3340, ppl=1.40, grad_norm=2.02, lr=9.77e-05 | |
| 2025-11-13 04:52:09,084 - INFO - Epoch 1 Step 1950 (Global: 1950): loss=0.4044, ppl=1.50, grad_norm=2.28, lr=9.77e-05 | |
| 2025-11-13 04:54:00,549 - INFO - Epoch 1 Step 1960 (Global: 1960): loss=0.3297, ppl=1.39, grad_norm=1.84, lr=9.76e-05 | |
| 2025-11-13 04:55:52,109 - INFO - Epoch 1 Step 1970 (Global: 1970): loss=0.3689, ppl=1.45, grad_norm=1.87, lr=9.76e-05 | |
| 2025-11-13 04:57:53,870 - INFO - Epoch 1 Step 1980 (Global: 1980): loss=0.3269, ppl=1.39, grad_norm=2.41, lr=9.75e-05 | |
| 2025-11-13 04:59:45,330 - INFO - Epoch 1 Step 1990 (Global: 1990): loss=0.3356, ppl=1.40, grad_norm=1.85, lr=9.75e-05 | |
| 2025-11-13 05:01:36,481 - INFO - Epoch 1 Step 2000 (Global: 2000): loss=0.3161, ppl=1.37, grad_norm=1.93, lr=9.74e-05 | |
| 2025-11-13 05:01:36,483 - INFO - | |
| Running validation at step 2000... | |
| 2025-11-13 05:07:11,528 - INFO - Validation loss: 0.3194, perplexity: 1.38 | |
| 2025-11-13 05:07:11,529 - INFO - Qualitative metrics (n=5): | |
| 2025-11-13 05:07:11,529 - INFO - BLEU: 0.2332 | |
| 2025-11-13 05:07:11,529 - INFO - METEOR: 0.4504 | |
| 2025-11-13 05:07:11,529 - INFO - Edit Distance: 0.4980 | |
| 2025-11-13 05:07:11,529 - INFO - F-measure: 0.4699 | |
| 2025-11-13 05:07:11,529 - INFO - | |
| ====================================================================== | |
| 2025-11-13 05:07:11,529 - INFO - Qualitative Evaluation Samples: | |
| 2025-11-13 05:07:11,530 - INFO - ====================================================================== | |
| 2025-11-13 05:07:11,530 - INFO - | |
| Sample 1 (ID: sample_141920_chunk_1): | |
| 2025-11-13 05:07:11,530 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 05:07:11,530 - INFO - Generated: ' four stars out of five gave it that said and "Perhaps on the [unlikely] likelihoods of actually sending these song lilies to their audience would make sense into their thinking\' whereas it\'s not. You...' | |
| 2025-11-13 05:07:11,530 - INFO - Ground Truth: 'Q gave it four stars out of five and said that "Perhaps [the album\'s] seemingly illogical sequencing of songs makes sense if they wish to lure their audience into thinking it\'s as-you-were. But it\'s n...' | |
| 2025-11-13 05:07:11,530 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 05:07:11,530 - INFO - | |
| Sample 2 (ID: sample_170543_chunk_2): | |
| 2025-11-13 05:07:11,530 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 05:07:11,530 - INFO - Generated: "# Saba Neire, US-Chile\nSaba Annette Louise American-Brazilian student member. The other Arab students included in the American Leadership member of the Women's Army; RTC counselor, and the co-head of ..." | |
| 2025-11-13 05:07:11,530 - INFO - Ground Truth: ', was Sirene Abou-Chakra, a Lebanese-American student who led the Arab-American Student Association. Other members included the woman president of the Michigan Student Assembly; the leader of Army ROT...' | |
| 2025-11-13 05:07:11,531 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 05:07:11,531 - INFO - | |
| Sample 3 (ID: sample_107152_chunk_9): | |
| 2025-11-13 05:07:11,531 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 05:07:11,531 - INFO - Generated: ' The meeting headed by Layla. His weapon of choice is an giant, and he has the power to immobilise his opponents when he puts it on their eye. In the fallo games, Obe tells the butt, and keeps beating...' | |
| 2025-11-13 05:07:11,531 - INFO - Ground Truth: ' at the meeting Laymia headed. His weapon of choice is a giant ax, and he has the power to immobilise his opponents if they look at him in the eye. Oga falls for the trick, but Beel stops the ax and b...' | |
| 2025-11-13 05:07:11,531 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 05:07:11,531 - INFO - | |
| Sample 4 (ID: sample_069148_chunk_0): | |
| 2025-11-13 05:07:11,531 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 05:07:11,531 - INFO - Generated: '# Unicode\nUnicode (or Alloyria) is a basic character encoding for the languages of Odonia, Khant and Santali and the state of India in Odonia. In its original incarnation, the code points for Unicode ...' | |
| 2025-11-13 05:07:11,532 - INFO - Ground Truth: '# Oriya (Unicode block)\nOriya is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0B01.....' | |
| 2025-11-13 05:07:11,533 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 05:07:11,533 - INFO - | |
| Sample 5 (ID: sample_103176_chunk_4): | |
| 2025-11-13 05:07:11,533 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 05:07:11,533 - INFO - Generated: ' |\n| 3 | The Sims Generations | May 31, 2011 | Windows | Max Redshouse | N...' | |
| 2025-11-13 05:07:11,533 - INFO - Ground Truth: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Maxis Redwood Shores ...' | |
| 2025-11-13 05:07:11,533 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 05:07:11,534 - INFO - | |
| Qualitative samples saved to: outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/qualitative_step_2000.jsonl | |
| 2025-11-13 05:07:57,088 - INFO - Saved checkpoint to outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/best_checkpoint.pt | |
| 2025-11-13 05:07:57,106 - INFO - New best validation loss: 0.3194, perplexity: 1.38 | |
| 2025-11-13 05:09:50,896 - INFO - Epoch 1 Step 2010 (Global: 2010): loss=0.3496, ppl=1.42, grad_norm=2.19, lr=9.74e-05 | |
| 2025-11-13 05:11:53,836 - INFO - Epoch 1 Step 2020 (Global: 2020): loss=0.2920, ppl=1.34, grad_norm=1.94, lr=9.73e-05 | |
| 2025-11-13 05:13:46,281 - INFO - Epoch 1 Step 2030 (Global: 2030): loss=0.3288, ppl=1.39, grad_norm=2.11, lr=9.73e-05 | |
| 2025-11-13 05:15:38,698 - INFO - Epoch 1 Step 2040 (Global: 2040): loss=0.2937, ppl=1.34, grad_norm=2.08, lr=9.72e-05 | |
| 2025-11-13 05:17:30,537 - INFO - Epoch 1 Step 2050 (Global: 2050): loss=0.3250, ppl=1.38, grad_norm=2.11, lr=9.72e-05 | |
| 2025-11-13 05:19:33,388 - INFO - Epoch 1 Step 2060 (Global: 2060): loss=0.2778, ppl=1.32, grad_norm=1.98, lr=9.71e-05 | |
| 2025-11-13 05:21:25,002 - INFO - Epoch 1 Step 2070 (Global: 2070): loss=0.2912, ppl=1.34, grad_norm=2.03, lr=9.71e-05 | |
| 2025-11-13 05:23:18,139 - INFO - Epoch 1 Step 2080 (Global: 2080): loss=0.2743, ppl=1.32, grad_norm=1.88, lr=9.70e-05 | |
| 2025-11-13 05:25:10,171 - INFO - Epoch 1 Step 2090 (Global: 2090): loss=0.2387, ppl=1.27, grad_norm=1.55, lr=9.69e-05 | |
| 2025-11-13 05:27:12,384 - INFO - Epoch 1 Step 2100 (Global: 2100): loss=0.2367, ppl=1.27, grad_norm=1.80, lr=9.69e-05 | |
| 2025-11-13 05:29:03,935 - INFO - Epoch 1 Step 2110 (Global: 2110): loss=0.2186, ppl=1.24, grad_norm=1.75, lr=9.68e-05 | |
| 2025-11-13 05:30:56,738 - INFO - Epoch 1 Step 2120 (Global: 2120): loss=0.2359, ppl=1.27, grad_norm=1.73, lr=9.68e-05 | |
| 2025-11-13 05:32:49,274 - INFO - Epoch 1 Step 2130 (Global: 2130): loss=0.2063, ppl=1.23, grad_norm=1.66, lr=9.67e-05 | |
| 2025-11-13 05:34:51,515 - INFO - Epoch 1 Step 2140 (Global: 2140): loss=0.2519, ppl=1.29, grad_norm=2.06, lr=9.66e-05 | |
| 2025-11-13 05:36:43,384 - INFO - Epoch 1 Step 2150 (Global: 2150): loss=0.2158, ppl=1.24, grad_norm=1.52, lr=9.66e-05 | |
| 2025-11-13 05:38:35,665 - INFO - Epoch 1 Step 2160 (Global: 2160): loss=0.1978, ppl=1.22, grad_norm=1.73, lr=9.65e-05 | |
| 2025-11-13 05:40:29,944 - INFO - Epoch 1 Step 2170 (Global: 2170): loss=0.2437, ppl=1.28, grad_norm=2.27, lr=9.65e-05 | |
| 2025-11-13 05:42:35,122 - INFO - Epoch 1 Step 2180 (Global: 2180): loss=0.2159, ppl=1.24, grad_norm=1.61, lr=9.64e-05 | |
| 2025-11-13 05:44:29,353 - INFO - Epoch 1 Step 2190 (Global: 2190): loss=0.1784, ppl=1.20, grad_norm=1.44, lr=9.63e-05 | |
| 2025-11-13 05:46:22,703 - INFO - Epoch 1 Step 2200 (Global: 2200): loss=0.2256, ppl=1.25, grad_norm=1.77, lr=9.63e-05 | |
| 2025-11-13 05:48:15,961 - INFO - Epoch 1 Step 2210 (Global: 2210): loss=0.1703, ppl=1.19, grad_norm=1.45, lr=9.62e-05 | |
| 2025-11-13 05:50:20,537 - INFO - Epoch 1 Step 2220 (Global: 2220): loss=0.1729, ppl=1.19, grad_norm=1.47, lr=9.61e-05 | |
| 2025-11-13 05:52:14,168 - INFO - Epoch 1 Step 2230 (Global: 2230): loss=0.1880, ppl=1.21, grad_norm=1.76, lr=9.61e-05 | |
| 2025-11-13 05:54:06,328 - INFO - Epoch 1 Step 2240 (Global: 2240): loss=0.1802, ppl=1.20, grad_norm=1.56, lr=9.60e-05 | |
| 2025-11-13 05:55:57,139 - INFO - Epoch 1 Step 2250 (Global: 2250): loss=0.1719, ppl=1.19, grad_norm=1.76, lr=9.60e-05 | |
| 2025-11-13 05:58:01,503 - INFO - Epoch 1 Step 2260 (Global: 2260): loss=0.1676, ppl=1.18, grad_norm=1.48, lr=9.59e-05 | |
| 2025-11-13 05:59:55,130 - INFO - Epoch 1 Step 2270 (Global: 2270): loss=0.1533, ppl=1.17, grad_norm=2.25, lr=9.58e-05 | |
| 2025-11-13 06:01:48,961 - INFO - Epoch 1 Step 2280 (Global: 2280): loss=0.1740, ppl=1.19, grad_norm=1.67, lr=9.58e-05 | |
| 2025-11-13 06:03:42,159 - INFO - Epoch 1 Step 2290 (Global: 2290): loss=0.1366, ppl=1.15, grad_norm=1.45, lr=9.57e-05 | |
| 2025-11-13 06:05:47,510 - INFO - Epoch 1 Step 2300 (Global: 2300): loss=0.1713, ppl=1.19, grad_norm=1.49, lr=9.56e-05 | |
| 2025-11-13 06:07:41,764 - INFO - Epoch 1 Step 2310 (Global: 2310): loss=0.1505, ppl=1.16, grad_norm=1.52, lr=9.55e-05 | |
| 2025-11-13 06:09:35,747 - INFO - Epoch 1 Step 2320 (Global: 2320): loss=0.1466, ppl=1.16, grad_norm=1.42, lr=9.55e-05 | |
| 2025-11-13 06:11:29,694 - INFO - Epoch 1 Step 2330 (Global: 2330): loss=0.1231, ppl=1.13, grad_norm=1.42, lr=9.54e-05 | |
| 2025-11-13 06:13:32,147 - INFO - Epoch 1 Step 2340 (Global: 2340): loss=0.1376, ppl=1.15, grad_norm=1.49, lr=9.53e-05 | |
| 2025-11-13 06:15:24,200 - INFO - Epoch 1 Step 2350 (Global: 2350): loss=0.1518, ppl=1.16, grad_norm=1.42, lr=9.53e-05 | |
| 2025-11-13 06:17:18,719 - INFO - Epoch 1 Step 2360 (Global: 2360): loss=0.1252, ppl=1.13, grad_norm=1.34, lr=9.52e-05 | |
| 2025-11-13 06:19:12,629 - INFO - Epoch 1 Step 2370 (Global: 2370): loss=0.1442, ppl=1.16, grad_norm=1.53, lr=9.51e-05 | |
| 2025-11-13 06:21:17,492 - INFO - Epoch 1 Step 2380 (Global: 2380): loss=0.1146, ppl=1.12, grad_norm=1.24, lr=9.51e-05 | |
| 2025-11-13 06:23:11,195 - INFO - Epoch 1 Step 2390 (Global: 2390): loss=0.1383, ppl=1.15, grad_norm=1.52, lr=9.50e-05 | |
| 2025-11-13 06:25:05,239 - INFO - Epoch 1 Step 2400 (Global: 2400): loss=0.1150, ppl=1.12, grad_norm=1.49, lr=9.49e-05 | |
| 2025-11-13 06:26:59,684 - INFO - Epoch 1 Step 2410 (Global: 2410): loss=0.1215, ppl=1.13, grad_norm=1.28, lr=9.48e-05 | |
| 2025-11-13 06:29:04,785 - INFO - Epoch 1 Step 2420 (Global: 2420): loss=0.1334, ppl=1.14, grad_norm=1.40, lr=9.48e-05 | |
| 2025-11-13 06:30:56,974 - INFO - Epoch 1 Step 2430 (Global: 2430): loss=0.1049, ppl=1.11, grad_norm=1.30, lr=9.47e-05 | |
| 2025-11-13 06:32:49,432 - INFO - Epoch 1 Step 2440 (Global: 2440): loss=0.1184, ppl=1.13, grad_norm=1.41, lr=9.46e-05 | |
| 2025-11-13 06:34:40,663 - INFO - Epoch 1 Step 2450 (Global: 2450): loss=0.1151, ppl=1.12, grad_norm=1.44, lr=9.45e-05 | |
| 2025-11-13 06:36:42,360 - INFO - Epoch 1 Step 2460 (Global: 2460): loss=0.1252, ppl=1.13, grad_norm=1.29, lr=9.45e-05 | |
| 2025-11-13 06:38:34,513 - INFO - Epoch 1 Step 2470 (Global: 2470): loss=0.1157, ppl=1.12, grad_norm=1.51, lr=9.44e-05 | |
| 2025-11-13 06:40:26,167 - INFO - Epoch 1 Step 2480 (Global: 2480): loss=0.1246, ppl=1.13, grad_norm=1.55, lr=9.43e-05 | |
| 2025-11-13 06:42:17,539 - INFO - Epoch 1 Step 2490 (Global: 2490): loss=0.0961, ppl=1.10, grad_norm=1.26, lr=9.42e-05 | |
| 2025-11-13 06:44:20,486 - INFO - Epoch 1 Step 2500 (Global: 2500): loss=0.1249, ppl=1.13, grad_norm=1.35, lr=9.41e-05 | |
| 2025-11-13 06:44:20,488 - INFO - | |
| Running validation at step 2500... | |
| 2025-11-13 06:49:45,951 - INFO - Validation loss: 0.1113, perplexity: 1.12 | |
| 2025-11-13 06:49:45,951 - INFO - Qualitative metrics (n=5): | |
| 2025-11-13 06:49:45,952 - INFO - BLEU: 0.4019 | |
| 2025-11-13 06:49:45,952 - INFO - METEOR: 0.6089 | |
| 2025-11-13 06:49:45,952 - INFO - Edit Distance: 0.3635 | |
| 2025-11-13 06:49:45,952 - INFO - F-measure: 0.5796 | |
| 2025-11-13 06:49:45,952 - INFO - | |
| ====================================================================== | |
| 2025-11-13 06:49:45,952 - INFO - Qualitative Evaluation Samples: | |
| 2025-11-13 06:49:45,952 - INFO - ====================================================================== | |
| 2025-11-13 06:49:45,952 - INFO - | |
| Sample 1 (ID: sample_141920_chunk_1): | |
| 2025-11-13 06:49:45,952 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 06:49:45,952 - INFO - Generated: ' it gave four stars out of five and said that "Perhaps [it] is the seemingly illogical assortment of seeds [whose words say] if they make it linger into their thoughts like it was...-But you\'re not th...' | |
| 2025-11-13 06:49:45,953 - INFO - Ground Truth: 'Q gave it four stars out of five and said that "Perhaps [the album\'s] seemingly illogical sequencing of songs makes sense if they wish to lure their audience into thinking it\'s as-you-were. But it\'s n...' | |
| 2025-11-13 06:49:45,953 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 06:49:45,953 - INFO - | |
| Sample 2 (ID: sample_170543_chunk_2): | |
| 2025-11-13 06:49:45,953 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 06:49:45,953 - INFO - Generated: ' | Sirene was A-Choubanaba, Lebanese X Street the American Arab student-led Association. Other Americans who participated in the member this mountain school the President: Edward Land of the Rally Cot...' | |
| 2025-11-13 06:49:45,954 - INFO - Ground Truth: ', was Sirene Abou-Chakra, a Lebanese-American student who led the Arab-American Student Association. Other members included the woman president of the Michigan Student Assembly; the leader of Army ROT...' | |
| 2025-11-13 06:49:45,954 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 06:49:45,954 - INFO - | |
| Sample 3 (ID: sample_107152_chunk_9): | |
| 2025-11-13 06:49:45,954 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 06:49:45,954 - INFO - Generated: ' at the meeting Laymia headed. His weapon of choice is a giant ax, and he has the power to immobilise his opponents if he looks at him in the eye. Oga falls for the trick, but Beelz is the fastest and...' | |
| 2025-11-13 06:49:45,954 - INFO - Ground Truth: ' at the meeting Laymia headed. His weapon of choice is a giant ax, and he has the power to immobilise his opponents if they look at him in the eye. Oga falls for the trick, but Beel stops the ax and b...' | |
| 2025-11-13 06:49:45,954 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 06:49:45,954 - INFO - | |
| Sample 4 (ID: sample_069148_chunk_0): | |
| 2025-11-13 06:49:45,954 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 06:49:45,954 - INFO - Generated: '# Oriya (unicode block)\nUnicode is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0010...' | |
| 2025-11-13 06:49:45,955 - INFO - Ground Truth: '# Oriya (Unicode block)\nOriya is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0B01.....' | |
| 2025-11-13 06:49:45,955 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 06:49:45,955 - INFO - | |
| Sample 5 (ID: sample_103176_chunk_4): | |
| 2025-11-13 06:49:45,955 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 06:49:45,955 - INFO - Generated: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Max Redshields Woodia ...' | |
| 2025-11-13 06:49:45,955 - INFO - Ground Truth: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Maxis Redwood Shores ...' | |
| 2025-11-13 06:49:45,955 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 06:49:45,956 - INFO - | |
| Qualitative samples saved to: outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/qualitative_step_2500.jsonl | |
| 2025-11-13 06:50:27,573 - INFO - Saved checkpoint to outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/best_checkpoint.pt | |
| 2025-11-13 06:50:27,587 - INFO - New best validation loss: 0.1113, perplexity: 1.12 | |
| 2025-11-13 06:52:18,463 - INFO - Epoch 1 Step 2510 (Global: 2510): loss=0.1063, ppl=1.11, grad_norm=1.29, lr=9.41e-05 | |
| 2025-11-13 06:54:09,863 - INFO - Epoch 1 Step 2520 (Global: 2520): loss=0.1060, ppl=1.11, grad_norm=1.23, lr=9.40e-05 | |
| 2025-11-13 06:56:10,479 - INFO - Epoch 1 Step 2530 (Global: 2530): loss=0.1053, ppl=1.11, grad_norm=1.24, lr=9.39e-05 | |
| 2025-11-13 06:58:02,144 - INFO - Epoch 1 Step 2540 (Global: 2540): loss=0.1058, ppl=1.11, grad_norm=1.31, lr=9.38e-05 | |
| 2025-11-13 06:59:54,256 - INFO - Epoch 1 Step 2550 (Global: 2550): loss=0.0961, ppl=1.10, grad_norm=1.16, lr=9.37e-05 | |
| 2025-11-13 07:01:45,606 - INFO - Epoch 1 Step 2560 (Global: 2560): loss=0.1059, ppl=1.11, grad_norm=1.33, lr=9.37e-05 | |
| 2025-11-13 07:03:47,296 - INFO - Epoch 1 Step 2570 (Global: 2570): loss=0.1023, ppl=1.11, grad_norm=1.16, lr=9.36e-05 | |
| 2025-11-13 07:05:38,450 - INFO - Epoch 1 Step 2580 (Global: 2580): loss=0.1029, ppl=1.11, grad_norm=1.30, lr=9.35e-05 | |
| 2025-11-13 07:07:30,253 - INFO - Epoch 1 Step 2590 (Global: 2590): loss=0.1043, ppl=1.11, grad_norm=1.18, lr=9.34e-05 | |
| 2025-11-13 07:09:22,231 - INFO - Epoch 1 Step 2600 (Global: 2600): loss=0.1043, ppl=1.11, grad_norm=1.27, lr=9.33e-05 | |
| 2025-11-13 07:11:23,339 - INFO - Epoch 1 Step 2610 (Global: 2610): loss=0.0969, ppl=1.10, grad_norm=1.28, lr=9.32e-05 | |
| 2025-11-13 07:13:14,272 - INFO - Epoch 1 Step 2620 (Global: 2620): loss=0.0985, ppl=1.10, grad_norm=1.15, lr=9.32e-05 | |
| 2025-11-13 07:15:05,128 - INFO - Epoch 1 Step 2630 (Global: 2630): loss=0.0827, ppl=1.09, grad_norm=1.05, lr=9.31e-05 | |
| 2025-11-13 07:16:56,295 - INFO - Epoch 1 Step 2640 (Global: 2640): loss=0.0917, ppl=1.10, grad_norm=1.19, lr=9.30e-05 | |
| 2025-11-13 07:18:56,676 - INFO - Epoch 1 Step 2650 (Global: 2650): loss=0.0855, ppl=1.09, grad_norm=1.15, lr=9.29e-05 | |
| 2025-11-13 07:20:47,120 - INFO - Epoch 1 Step 2660 (Global: 2660): loss=0.0839, ppl=1.09, grad_norm=1.12, lr=9.28e-05 | |
| 2025-11-13 07:22:38,127 - INFO - Epoch 1 Step 2670 (Global: 2670): loss=0.0955, ppl=1.10, grad_norm=1.21, lr=9.27e-05 | |
| 2025-11-13 07:24:29,554 - INFO - Epoch 1 Step 2680 (Global: 2680): loss=0.0910, ppl=1.10, grad_norm=1.18, lr=9.26e-05 | |
| 2025-11-13 07:26:29,522 - INFO - Epoch 1 Step 2690 (Global: 2690): loss=0.0747, ppl=1.08, grad_norm=1.00, lr=9.26e-05 | |
| 2025-11-13 07:28:20,178 - INFO - Epoch 1 Step 2700 (Global: 2700): loss=0.0879, ppl=1.09, grad_norm=1.20, lr=9.25e-05 | |
| 2025-11-13 07:30:11,021 - INFO - Epoch 1 Step 2710 (Global: 2710): loss=0.0823, ppl=1.09, grad_norm=1.16, lr=9.24e-05 | |
| 2025-11-13 07:32:01,793 - INFO - Epoch 1 Step 2720 (Global: 2720): loss=0.0949, ppl=1.10, grad_norm=1.38, lr=9.23e-05 | |
| 2025-11-13 07:34:02,005 - INFO - Epoch 1 Step 2730 (Global: 2730): loss=0.0868, ppl=1.09, grad_norm=1.20, lr=9.22e-05 | |
| 2025-11-13 07:35:52,346 - INFO - Epoch 1 Step 2740 (Global: 2740): loss=0.0941, ppl=1.10, grad_norm=1.14, lr=9.21e-05 | |
| 2025-11-13 07:37:42,915 - INFO - Epoch 1 Step 2750 (Global: 2750): loss=0.0852, ppl=1.09, grad_norm=1.17, lr=9.20e-05 | |
| 2025-11-13 07:39:33,460 - INFO - Epoch 1 Step 2760 (Global: 2760): loss=0.0743, ppl=1.08, grad_norm=1.12, lr=9.19e-05 | |
| 2025-11-13 07:41:33,400 - INFO - Epoch 1 Step 2770 (Global: 2770): loss=0.0880, ppl=1.09, grad_norm=1.22, lr=9.18e-05 | |
| 2025-11-13 07:43:24,298 - INFO - Epoch 1 Step 2780 (Global: 2780): loss=0.0657, ppl=1.07, grad_norm=0.99, lr=9.17e-05 | |
| 2025-11-13 07:45:15,384 - INFO - Epoch 1 Step 2790 (Global: 2790): loss=0.0696, ppl=1.07, grad_norm=1.00, lr=9.17e-05 | |
| 2025-11-13 07:47:05,861 - INFO - Epoch 1 Step 2800 (Global: 2800): loss=0.0667, ppl=1.07, grad_norm=1.05, lr=9.16e-05 | |
| 2025-11-13 07:49:05,645 - INFO - Epoch 1 Step 2810 (Global: 2810): loss=0.0686, ppl=1.07, grad_norm=1.02, lr=9.15e-05 | |
| 2025-11-13 07:50:56,198 - INFO - Epoch 1 Step 2820 (Global: 2820): loss=0.0716, ppl=1.07, grad_norm=0.95, lr=9.14e-05 | |
| 2025-11-13 07:52:46,904 - INFO - Epoch 1 Step 2830 (Global: 2830): loss=0.0680, ppl=1.07, grad_norm=0.98, lr=9.13e-05 | |
| 2025-11-13 07:54:37,326 - INFO - Epoch 1 Step 2840 (Global: 2840): loss=0.0631, ppl=1.07, grad_norm=0.93, lr=9.12e-05 | |
| 2025-11-13 07:56:36,930 - INFO - Epoch 1 Step 2850 (Global: 2850): loss=0.0669, ppl=1.07, grad_norm=0.92, lr=9.11e-05 | |
| 2025-11-13 07:58:27,894 - INFO - Epoch 1 Step 2860 (Global: 2860): loss=0.0655, ppl=1.07, grad_norm=0.93, lr=9.10e-05 | |
| 2025-11-13 08:00:19,716 - INFO - Epoch 1 Step 2870 (Global: 2870): loss=0.0710, ppl=1.07, grad_norm=1.13, lr=9.09e-05 | |
| 2025-11-13 08:02:10,429 - INFO - Epoch 1 Step 2880 (Global: 2880): loss=0.0710, ppl=1.07, grad_norm=1.08, lr=9.08e-05 | |
| 2025-11-13 08:04:11,419 - INFO - Epoch 1 Step 2890 (Global: 2890): loss=0.0657, ppl=1.07, grad_norm=0.97, lr=9.07e-05 | |
| 2025-11-13 08:06:03,474 - INFO - Epoch 1 Step 2900 (Global: 2900): loss=0.0701, ppl=1.07, grad_norm=1.05, lr=9.06e-05 | |
| 2025-11-13 08:07:54,608 - INFO - Epoch 1 Step 2910 (Global: 2910): loss=0.0674, ppl=1.07, grad_norm=1.13, lr=9.05e-05 | |
| 2025-11-13 08:09:46,060 - INFO - Epoch 1 Step 2920 (Global: 2920): loss=0.0554, ppl=1.06, grad_norm=0.91, lr=9.04e-05 | |
| 2025-11-13 08:11:47,099 - INFO - Epoch 1 Step 2930 (Global: 2930): loss=0.0638, ppl=1.07, grad_norm=1.08, lr=9.03e-05 | |
| 2025-11-13 08:13:38,250 - INFO - Epoch 1 Step 2940 (Global: 2940): loss=0.0627, ppl=1.06, grad_norm=1.07, lr=9.02e-05 | |
| 2025-11-13 08:15:29,143 - INFO - Epoch 1 Step 2950 (Global: 2950): loss=0.0599, ppl=1.06, grad_norm=0.97, lr=9.01e-05 | |
| 2025-11-13 08:17:20,262 - INFO - Epoch 1 Step 2960 (Global: 2960): loss=0.0585, ppl=1.06, grad_norm=1.21, lr=9.00e-05 | |
| 2025-11-13 08:19:21,683 - INFO - Epoch 1 Step 2970 (Global: 2970): loss=0.0708, ppl=1.07, grad_norm=0.96, lr=8.99e-05 | |
| 2025-11-13 08:21:12,328 - INFO - Epoch 1 Step 2980 (Global: 2980): loss=0.0522, ppl=1.05, grad_norm=0.82, lr=8.98e-05 | |
| 2025-11-13 08:23:03,172 - INFO - Epoch 1 Step 2990 (Global: 2990): loss=0.0612, ppl=1.06, grad_norm=1.00, lr=8.97e-05 | |
| 2025-11-13 08:24:53,939 - INFO - Epoch 1 Step 3000 (Global: 3000): loss=0.0648, ppl=1.07, grad_norm=0.97, lr=8.96e-05 | |
| 2025-11-13 08:24:53,942 - INFO - | |
| Running validation at step 3000... | |
| 2025-11-13 08:30:23,236 - INFO - Validation loss: 0.0563, perplexity: 1.06 | |
| 2025-11-13 08:30:23,236 - INFO - Qualitative metrics (n=5): | |
| 2025-11-13 08:30:23,236 - INFO - BLEU: 0.6139 | |
| 2025-11-13 08:30:23,237 - INFO - METEOR: 0.7364 | |
| 2025-11-13 08:30:23,237 - INFO - Edit Distance: 0.2262 | |
| 2025-11-13 08:30:23,237 - INFO - F-measure: 0.7281 | |
| 2025-11-13 08:30:23,237 - INFO - | |
| ====================================================================== | |
| 2025-11-13 08:30:23,237 - INFO - Qualitative Evaluation Samples: | |
| 2025-11-13 08:30:23,237 - INFO - ====================================================================== | |
| 2025-11-13 08:30:23,237 - INFO - | |
| Sample 1 (ID: sample_141920_chunk_1): | |
| 2025-11-13 08:30:23,237 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 08:30:23,237 - INFO - Generated: ' it gave Q four stars out of five and said that "Perhaps [the album\'s] seemingly illogical sequences of songs make sense if they wish to lure their audience into thinking it\'s as you-wonder. But it\'s ...' | |
| 2025-11-13 08:30:23,238 - INFO - Ground Truth: 'Q gave it four stars out of five and said that "Perhaps [the album\'s] seemingly illogical sequencing of songs makes sense if they wish to lure their audience into thinking it\'s as-you-were. But it\'s n...' | |
| 2025-11-13 08:30:23,238 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 08:30:23,238 - INFO - | |
| Sample 2 (ID: sample_170543_chunk_2): | |
| 2025-11-13 08:30:23,238 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 08:30:23,238 - INFO - Generated: ', Sire was Aboune-Chakra, a Lebanese-American student who led the Arab-American Student Association. Other members happened in the male student women (the Assembly President; the Army leader of Z.C. a...' | |
| 2025-11-13 08:30:23,238 - INFO - Ground Truth: ', was Sirene Abou-Chakra, a Lebanese-American student who led the Arab-American Student Association. Other members included the woman president of the Michigan Student Assembly; the leader of Army ROT...' | |
| 2025-11-13 08:30:23,238 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 08:30:23,238 - INFO - | |
| Sample 3 (ID: sample_107152_chunk_9): | |
| 2025-11-13 08:30:23,238 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 08:30:23,238 - INFO - Generated: ' meet the Layman headed by Mia. Its shooting hole of a can agent is high, and he has the power vetoises until his opponent he is not in the eye. Oga falls for the hit; Butalek is the both defoss and b...' | |
| 2025-11-13 08:30:23,238 - INFO - Ground Truth: ' at the meeting Laymia headed. His weapon of choice is a giant ax, and he has the power to immobilise his opponents if they look at him in the eye. Oga falls for the trick, but Beel stops the ax and b...' | |
| 2025-11-13 08:30:23,238 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 08:30:23,239 - INFO - | |
| Sample 4 (ID: sample_069148_chunk_0): | |
| 2025-11-13 08:30:23,239 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 08:30:23,239 - INFO - Generated: '# Oriya (Unicode block)\nOriya is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0B01.....' | |
| 2025-11-13 08:30:23,239 - INFO - Ground Truth: '# Oriya (Unicode block)\nOriya is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0B01.....' | |
| 2025-11-13 08:30:23,239 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 08:30:23,239 - INFO - | |
| Sample 5 (ID: sample_103176_chunk_4): | |
| 2025-11-13 08:30:23,239 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 08:30:23,239 - INFO - Generated: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Maxis Redwood Shores ...' | |
| 2025-11-13 08:30:23,239 - INFO - Ground Truth: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Maxis Redwood Shores ...' | |
| 2025-11-13 08:30:23,239 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 08:30:23,240 - INFO - | |
| Qualitative samples saved to: outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/qualitative_step_3000.jsonl | |
| 2025-11-13 08:31:08,259 - INFO - Saved checkpoint to outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/best_checkpoint.pt | |
| 2025-11-13 08:31:08,273 - INFO - New best validation loss: 0.0563, perplexity: 1.06 | |
| 2025-11-13 08:33:08,445 - INFO - Epoch 1 Step 3010 (Global: 3010): loss=0.0517, ppl=1.05, grad_norm=0.83, lr=8.95e-05 | |
| 2025-11-13 08:34:58,845 - INFO - Epoch 1 Step 3020 (Global: 3020): loss=0.0637, ppl=1.07, grad_norm=1.00, lr=8.94e-05 | |
| 2025-11-13 08:36:49,144 - INFO - Epoch 1 Step 3030 (Global: 3030): loss=0.0766, ppl=1.08, grad_norm=1.05, lr=8.93e-05 | |
| 2025-11-13 08:38:40,544 - INFO - Epoch 1 Step 3040 (Global: 3040): loss=0.0680, ppl=1.07, grad_norm=0.95, lr=8.92e-05 | |
| 2025-11-13 08:40:41,328 - INFO - Epoch 1 Step 3050 (Global: 3050): loss=0.0566, ppl=1.06, grad_norm=0.88, lr=8.91e-05 | |
| 2025-11-13 08:42:33,203 - INFO - Epoch 1 Step 3060 (Global: 3060): loss=0.0573, ppl=1.06, grad_norm=0.99, lr=8.90e-05 | |
| 2025-11-13 08:44:24,996 - INFO - Epoch 1 Step 3070 (Global: 3070): loss=0.0586, ppl=1.06, grad_norm=0.97, lr=8.89e-05 | |
| 2025-11-13 08:46:16,056 - INFO - Epoch 1 Step 3080 (Global: 3080): loss=0.0537, ppl=1.06, grad_norm=0.88, lr=8.88e-05 | |
| 2025-11-13 08:48:16,178 - INFO - Epoch 1 Step 3090 (Global: 3090): loss=0.0570, ppl=1.06, grad_norm=1.01, lr=8.87e-05 | |
| 2025-11-13 08:50:06,918 - INFO - Epoch 1 Step 3100 (Global: 3100): loss=0.0556, ppl=1.06, grad_norm=1.02, lr=8.86e-05 | |
| 2025-11-13 08:51:58,265 - INFO - Epoch 1 Step 3110 (Global: 3110): loss=0.0476, ppl=1.05, grad_norm=0.83, lr=8.85e-05 | |
| 2025-11-13 08:53:48,672 - INFO - Epoch 1 Step 3120 (Global: 3120): loss=0.0487, ppl=1.05, grad_norm=0.82, lr=8.84e-05 | |
| 2025-11-13 08:55:48,736 - INFO - Epoch 1 Step 3130 (Global: 3130): loss=0.0508, ppl=1.05, grad_norm=0.91, lr=8.82e-05 | |
| 2025-11-13 08:57:39,604 - INFO - Epoch 1 Step 3140 (Global: 3140): loss=0.0514, ppl=1.05, grad_norm=0.86, lr=8.81e-05 | |
| 2025-11-13 08:59:30,565 - INFO - Epoch 1 Step 3150 (Global: 3150): loss=0.0446, ppl=1.05, grad_norm=0.89, lr=8.80e-05 | |
| 2025-11-13 09:01:21,759 - INFO - Epoch 1 Step 3160 (Global: 3160): loss=0.0375, ppl=1.04, grad_norm=0.72, lr=8.79e-05 | |
| 2025-11-13 09:03:22,290 - INFO - Epoch 1 Step 3170 (Global: 3170): loss=0.0460, ppl=1.05, grad_norm=0.81, lr=8.78e-05 | |
| 2025-11-13 09:05:13,150 - INFO - Epoch 1 Step 3180 (Global: 3180): loss=0.0536, ppl=1.06, grad_norm=0.98, lr=8.77e-05 | |
| 2025-11-13 09:07:03,841 - INFO - Epoch 1 Step 3190 (Global: 3190): loss=0.0489, ppl=1.05, grad_norm=0.89, lr=8.76e-05 | |
| 2025-11-13 09:08:54,331 - INFO - Epoch 1 Step 3200 (Global: 3200): loss=0.0463, ppl=1.05, grad_norm=0.91, lr=8.75e-05 | |
| 2025-11-13 09:10:55,087 - INFO - Epoch 1 Step 3210 (Global: 3210): loss=0.0364, ppl=1.04, grad_norm=0.79, lr=8.74e-05 | |
| 2025-11-13 09:12:46,073 - INFO - Epoch 1 Step 3220 (Global: 3220): loss=0.0480, ppl=1.05, grad_norm=0.80, lr=8.73e-05 | |
| 2025-11-13 09:14:36,896 - INFO - Epoch 1 Step 3230 (Global: 3230): loss=0.0511, ppl=1.05, grad_norm=0.88, lr=8.71e-05 | |
| 2025-11-13 09:16:27,903 - INFO - Epoch 1 Step 3240 (Global: 3240): loss=0.0479, ppl=1.05, grad_norm=1.04, lr=8.70e-05 | |
| 2025-11-13 09:18:27,924 - INFO - Epoch 1 Step 3250 (Global: 3250): loss=0.0382, ppl=1.04, grad_norm=0.73, lr=8.69e-05 | |
| 2025-11-13 09:20:18,949 - INFO - Epoch 1 Step 3260 (Global: 3260): loss=0.0379, ppl=1.04, grad_norm=0.79, lr=8.68e-05 | |
| 2025-11-13 09:22:10,390 - INFO - Epoch 1 Step 3270 (Global: 3270): loss=0.0392, ppl=1.04, grad_norm=0.73, lr=8.67e-05 | |
| 2025-11-13 09:24:00,780 - INFO - Epoch 1 Step 3280 (Global: 3280): loss=0.0368, ppl=1.04, grad_norm=0.73, lr=8.66e-05 | |
| 2025-11-13 09:26:01,392 - INFO - Epoch 1 Step 3290 (Global: 3290): loss=0.0354, ppl=1.04, grad_norm=0.71, lr=8.65e-05 | |
| 2025-11-13 09:27:51,938 - INFO - Epoch 1 Step 3300 (Global: 3300): loss=0.0343, ppl=1.03, grad_norm=0.77, lr=8.63e-05 | |
| 2025-11-13 09:29:42,403 - INFO - Epoch 1 Step 3310 (Global: 3310): loss=0.0366, ppl=1.04, grad_norm=0.71, lr=8.62e-05 | |
| 2025-11-13 09:31:32,758 - INFO - Epoch 1 Step 3320 (Global: 3320): loss=0.0289, ppl=1.03, grad_norm=0.66, lr=8.61e-05 | |
| 2025-11-13 09:33:33,249 - INFO - Epoch 1 Step 3330 (Global: 3330): loss=0.0362, ppl=1.04, grad_norm=0.84, lr=8.60e-05 | |
| 2025-11-13 09:35:24,600 - INFO - Epoch 1 Step 3340 (Global: 3340): loss=0.0336, ppl=1.03, grad_norm=0.66, lr=8.59e-05 | |
| 2025-11-13 09:37:15,189 - INFO - Epoch 1 Step 3350 (Global: 3350): loss=0.0341, ppl=1.03, grad_norm=0.71, lr=8.58e-05 | |
| 2025-11-13 09:39:05,707 - INFO - Epoch 1 Step 3360 (Global: 3360): loss=0.0386, ppl=1.04, grad_norm=0.79, lr=8.57e-05 | |
| 2025-11-13 09:41:06,020 - INFO - Epoch 1 Step 3370 (Global: 3370): loss=0.0354, ppl=1.04, grad_norm=0.70, lr=8.55e-05 | |
| 2025-11-13 09:42:57,464 - INFO - Epoch 1 Step 3380 (Global: 3380): loss=0.0373, ppl=1.04, grad_norm=0.73, lr=8.54e-05 | |
| 2025-11-13 09:44:49,198 - INFO - Epoch 1 Step 3390 (Global: 3390): loss=0.0263, ppl=1.03, grad_norm=0.61, lr=8.53e-05 | |
| 2025-11-13 09:46:40,828 - INFO - Epoch 1 Step 3400 (Global: 3400): loss=0.0331, ppl=1.03, grad_norm=0.80, lr=8.52e-05 | |
| 2025-11-13 09:48:40,897 - INFO - Epoch 1 Step 3410 (Global: 3410): loss=0.0344, ppl=1.03, grad_norm=0.71, lr=8.51e-05 | |
| 2025-11-13 09:50:32,313 - INFO - Epoch 1 Step 3420 (Global: 3420): loss=0.0375, ppl=1.04, grad_norm=0.80, lr=8.49e-05 | |
| 2025-11-13 09:52:23,849 - INFO - Epoch 1 Step 3430 (Global: 3430): loss=0.0364, ppl=1.04, grad_norm=0.79, lr=8.48e-05 | |
| 2025-11-13 09:54:15,173 - INFO - Epoch 1 Step 3440 (Global: 3440): loss=0.0292, ppl=1.03, grad_norm=0.66, lr=8.47e-05 | |
| 2025-11-13 09:56:15,712 - INFO - Epoch 1 Step 3450 (Global: 3450): loss=0.0299, ppl=1.03, grad_norm=0.65, lr=8.46e-05 | |
| 2025-11-13 09:58:06,975 - INFO - Epoch 1 Step 3460 (Global: 3460): loss=0.0332, ppl=1.03, grad_norm=0.79, lr=8.45e-05 | |
| 2025-11-13 09:59:58,107 - INFO - Epoch 1 Step 3470 (Global: 3470): loss=0.0314, ppl=1.03, grad_norm=0.68, lr=8.43e-05 | |
| 2025-11-13 10:01:49,359 - INFO - Epoch 1 Step 3480 (Global: 3480): loss=0.0369, ppl=1.04, grad_norm=0.73, lr=8.42e-05 | |
| 2025-11-13 10:03:49,363 - INFO - Epoch 1 Step 3490 (Global: 3490): loss=0.0332, ppl=1.03, grad_norm=0.76, lr=8.41e-05 | |
| 2025-11-13 10:05:39,973 - INFO - Epoch 1 Step 3500 (Global: 3500): loss=0.0298, ppl=1.03, grad_norm=0.68, lr=8.40e-05 | |
| 2025-11-13 10:05:39,976 - INFO - | |
| Running validation at step 3500... | |
| 2025-11-13 10:11:08,229 - INFO - Validation loss: 0.0327, perplexity: 1.03 | |
| 2025-11-13 10:11:08,229 - INFO - Qualitative metrics (n=5): | |
| 2025-11-13 10:11:08,230 - INFO - BLEU: 0.8587 | |
| 2025-11-13 10:11:08,230 - INFO - METEOR: 0.9436 | |
| 2025-11-13 10:11:08,230 - INFO - Edit Distance: 0.0473 | |
| 2025-11-13 10:11:08,230 - INFO - F-measure: 0.9311 | |
| 2025-11-13 10:11:08,230 - INFO - | |
| ====================================================================== | |
| 2025-11-13 10:11:08,230 - INFO - Qualitative Evaluation Samples: | |
| 2025-11-13 10:11:08,230 - INFO - ====================================================================== | |
| 2025-11-13 10:11:08,230 - INFO - | |
| Sample 1 (ID: sample_141920_chunk_1): | |
| 2025-11-13 10:11:08,230 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 10:11:08,230 - INFO - Generated: ' it gave Q four stars out of five and said that "Perhaps [the album\'s] seemingly illogical sequencing of senses make songs if they wish to lure their audience into thinking it\'s as-you-were. But it\'s ...' | |
| 2025-11-13 10:11:08,230 - INFO - Ground Truth: 'Q gave it four stars out of five and said that "Perhaps [the album\'s] seemingly illogical sequencing of songs makes sense if they wish to lure their audience into thinking it\'s as-you-were. But it\'s n...' | |
| 2025-11-13 10:11:08,231 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 10:11:08,231 - INFO - | |
| Sample 2 (ID: sample_170543_chunk_2): | |
| 2025-11-13 10:11:08,231 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 10:11:08,231 - INFO - Generated: ', Sire was above Deoua-Charke, a Lebanese Muslim student compared the Arab-American Student Association. Other leaders who happened the public member of the Michigan Student Assembly; the leader of Ar...' | |
| 2025-11-13 10:11:08,231 - INFO - Ground Truth: ', was Sirene Abou-Chakra, a Lebanese-American student who led the Arab-American Student Association. Other members included the woman president of the Michigan Student Assembly; the leader of Army ROT...' | |
| 2025-11-13 10:11:08,231 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 10:11:08,231 - INFO - | |
| Sample 3 (ID: sample_107152_chunk_9): | |
| 2025-11-13 10:11:08,231 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 10:11:08,231 - INFO - Generated: ' at the meeting Laymah headed. His weapon of choice is a giant ax, and he has the power to immobilise his opponents if they look at him in the eye. Oga falls for the trick, but Beel stops the ax and b...' | |
| 2025-11-13 10:11:08,232 - INFO - Ground Truth: ' at the meeting Laymia headed. His weapon of choice is a giant ax, and he has the power to immobilise his opponents if they look at him in the eye. Oga falls for the trick, but Beel stops the ax and b...' | |
| 2025-11-13 10:11:08,232 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 10:11:08,232 - INFO - | |
| Sample 4 (ID: sample_069148_chunk_0): | |
| 2025-11-13 10:11:08,232 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 10:11:08,232 - INFO - Generated: '# Oriya (Unicode block)\nOriya is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0B01.....' | |
| 2025-11-13 10:11:08,232 - INFO - Ground Truth: '# Oriya (Unicode block)\nOriya is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0B01.....' | |
| 2025-11-13 10:11:08,232 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 10:11:08,232 - INFO - | |
| Sample 5 (ID: sample_103176_chunk_4): | |
| 2025-11-13 10:11:08,232 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 10:11:08,232 - INFO - Generated: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Maxis Redwood Shores ...' | |
| 2025-11-13 10:11:08,232 - INFO - Ground Truth: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Maxis Redwood Shores ...' | |
| 2025-11-13 10:11:08,233 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 10:11:08,233 - INFO - | |
| Qualitative samples saved to: outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/qualitative_step_3500.jsonl | |
| 2025-11-13 10:11:59,127 - INFO - Saved checkpoint to outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/best_checkpoint.pt | |
| 2025-11-13 10:11:59,142 - INFO - New best validation loss: 0.0327, perplexity: 1.03 | |
| 2025-11-13 10:13:51,474 - INFO - Epoch 1 Step 3510 (Global: 3510): loss=0.0360, ppl=1.04, grad_norm=0.67, lr=8.38e-05 | |
| 2025-11-13 10:15:52,804 - INFO - Epoch 1 Step 3520 (Global: 3520): loss=0.0349, ppl=1.04, grad_norm=0.74, lr=8.37e-05 | |
| 2025-11-13 10:17:44,398 - INFO - Epoch 1 Step 3530 (Global: 3530): loss=0.0273, ppl=1.03, grad_norm=0.65, lr=8.36e-05 | |
| 2025-11-13 10:19:36,337 - INFO - Epoch 1 Step 3540 (Global: 3540): loss=0.0351, ppl=1.04, grad_norm=0.77, lr=8.35e-05 | |
| 2025-11-13 10:21:28,014 - INFO - Epoch 1 Step 3550 (Global: 3550): loss=0.0346, ppl=1.04, grad_norm=0.74, lr=8.33e-05 | |
| 2025-11-13 10:23:28,714 - INFO - Epoch 1 Step 3560 (Global: 3560): loss=0.0277, ppl=1.03, grad_norm=0.70, lr=8.32e-05 | |
| 2025-11-13 10:25:21,937 - INFO - Epoch 1 Step 3570 (Global: 3570): loss=0.0280, ppl=1.03, grad_norm=0.69, lr=8.31e-05 | |
| 2025-11-13 10:27:16,075 - INFO - Epoch 1 Step 3580 (Global: 3580): loss=0.0280, ppl=1.03, grad_norm=0.70, lr=8.30e-05 | |
| 2025-11-13 10:29:08,648 - INFO - Epoch 1 Step 3590 (Global: 3590): loss=0.0346, ppl=1.04, grad_norm=0.73, lr=8.28e-05 | |
| 2025-11-13 10:31:09,262 - INFO - Epoch 1 Step 3600 (Global: 3600): loss=0.0271, ppl=1.03, grad_norm=0.66, lr=8.27e-05 | |
| 2025-11-13 10:32:59,852 - INFO - Epoch 1 Step 3610 (Global: 3610): loss=0.0295, ppl=1.03, grad_norm=0.74, lr=8.26e-05 | |
| 2025-11-13 10:34:50,932 - INFO - Epoch 1 Step 3620 (Global: 3620): loss=0.0314, ppl=1.03, grad_norm=0.77, lr=8.25e-05 | |
| 2025-11-13 10:36:41,500 - INFO - Epoch 1 Step 3630 (Global: 3630): loss=0.0317, ppl=1.03, grad_norm=0.73, lr=8.23e-05 | |
| 2025-11-13 10:38:41,216 - INFO - Epoch 1 Step 3640 (Global: 3640): loss=0.0274, ppl=1.03, grad_norm=0.62, lr=8.22e-05 | |
| 2025-11-13 10:40:32,065 - INFO - Epoch 1 Step 3650 (Global: 3650): loss=0.0245, ppl=1.02, grad_norm=0.67, lr=8.21e-05 | |
| 2025-11-13 10:42:23,271 - INFO - Epoch 1 Step 3660 (Global: 3660): loss=0.0254, ppl=1.03, grad_norm=0.62, lr=8.20e-05 | |
| 2025-11-13 10:44:14,462 - INFO - Epoch 1 Step 3670 (Global: 3670): loss=0.0323, ppl=1.03, grad_norm=0.76, lr=8.18e-05 | |
| 2025-11-13 10:46:14,485 - INFO - Epoch 1 Step 3680 (Global: 3680): loss=0.0253, ppl=1.03, grad_norm=0.60, lr=8.17e-05 | |
| 2025-11-13 10:48:05,396 - INFO - Epoch 1 Step 3690 (Global: 3690): loss=0.0278, ppl=1.03, grad_norm=0.62, lr=8.16e-05 | |
| 2025-11-13 10:49:56,651 - INFO - Epoch 1 Step 3700 (Global: 3700): loss=0.0399, ppl=1.04, grad_norm=0.83, lr=8.14e-05 | |
| 2025-11-13 10:51:47,394 - INFO - Epoch 1 Step 3710 (Global: 3710): loss=0.0319, ppl=1.03, grad_norm=0.85, lr=8.13e-05 | |
| 2025-11-13 10:53:48,154 - INFO - Epoch 1 Step 3720 (Global: 3720): loss=0.0262, ppl=1.03, grad_norm=0.62, lr=8.12e-05 | |
| 2025-11-13 10:55:39,634 - INFO - Epoch 1 Step 3730 (Global: 3730): loss=0.0279, ppl=1.03, grad_norm=0.75, lr=8.10e-05 | |
| 2025-11-13 10:57:31,129 - INFO - Epoch 1 Step 3740 (Global: 3740): loss=0.0239, ppl=1.02, grad_norm=0.61, lr=8.09e-05 | |
| 2025-11-13 10:59:22,105 - INFO - Epoch 1 Step 3750 (Global: 3750): loss=0.0242, ppl=1.02, grad_norm=0.57, lr=8.08e-05 | |
| 2025-11-13 11:01:22,821 - INFO - Epoch 1 Step 3760 (Global: 3760): loss=0.0307, ppl=1.03, grad_norm=0.70, lr=8.06e-05 | |
| 2025-11-13 11:03:15,289 - INFO - Epoch 1 Step 3770 (Global: 3770): loss=0.0231, ppl=1.02, grad_norm=0.61, lr=8.05e-05 | |
| 2025-11-13 11:05:07,246 - INFO - Epoch 1 Step 3780 (Global: 3780): loss=0.0232, ppl=1.02, grad_norm=0.59, lr=8.04e-05 | |
| 2025-11-13 11:06:58,740 - INFO - Epoch 1 Step 3790 (Global: 3790): loss=0.0225, ppl=1.02, grad_norm=0.57, lr=8.02e-05 | |
| 2025-11-13 11:08:58,974 - INFO - Epoch 1 Step 3800 (Global: 3800): loss=0.0235, ppl=1.02, grad_norm=0.64, lr=8.01e-05 | |
| 2025-11-13 11:10:49,877 - INFO - Epoch 1 Step 3810 (Global: 3810): loss=0.0266, ppl=1.03, grad_norm=0.68, lr=8.00e-05 | |
| 2025-11-13 11:12:41,186 - INFO - Epoch 1 Step 3820 (Global: 3820): loss=0.0231, ppl=1.02, grad_norm=0.69, lr=7.98e-05 | |
| 2025-11-13 11:14:32,247 - INFO - Epoch 1 Step 3830 (Global: 3830): loss=0.0200, ppl=1.02, grad_norm=0.55, lr=7.97e-05 | |
| 2025-11-13 11:16:32,491 - INFO - Epoch 1 Step 3840 (Global: 3840): loss=0.0223, ppl=1.02, grad_norm=0.58, lr=7.96e-05 | |
| 2025-11-13 11:18:24,282 - INFO - Epoch 1 Step 3850 (Global: 3850): loss=0.0248, ppl=1.03, grad_norm=0.65, lr=7.94e-05 | |
| 2025-11-13 11:20:16,981 - INFO - Epoch 1 Step 3860 (Global: 3860): loss=0.0157, ppl=1.02, grad_norm=0.49, lr=7.93e-05 | |
| 2025-11-13 11:22:08,796 - INFO - Epoch 1 Step 3870 (Global: 3870): loss=0.0233, ppl=1.02, grad_norm=0.61, lr=7.92e-05 | |
| 2025-11-13 11:24:00,986 - INFO - Epoch 1 Step 3880 (Global: 3880): loss=0.0244, ppl=1.02, grad_norm=0.63, lr=7.90e-05 | |
| 2025-11-13 11:26:02,962 - INFO - Epoch 1 Step 3890 (Global: 3890): loss=0.0217, ppl=1.02, grad_norm=0.58, lr=7.89e-05 | |
| 2025-11-13 11:27:54,088 - INFO - Epoch 1 Step 3900 (Global: 3900): loss=0.0277, ppl=1.03, grad_norm=0.65, lr=7.88e-05 | |
| 2025-11-13 11:29:44,796 - INFO - Epoch 1 Step 3910 (Global: 3910): loss=0.0278, ppl=1.03, grad_norm=0.68, lr=7.86e-05 | |
| 2025-11-13 11:31:45,189 - INFO - Epoch 1 Step 3920 (Global: 3920): loss=0.0197, ppl=1.02, grad_norm=0.56, lr=7.85e-05 | |
| 2025-11-13 11:33:35,457 - INFO - Epoch 1 Step 3930 (Global: 3930): loss=0.0225, ppl=1.02, grad_norm=0.58, lr=7.83e-05 | |
| 2025-11-13 11:35:26,696 - INFO - Epoch 1 Step 3940 (Global: 3940): loss=0.0244, ppl=1.02, grad_norm=0.64, lr=7.82e-05 | |
| 2025-11-13 11:37:17,485 - INFO - Epoch 1 Step 3950 (Global: 3950): loss=0.0213, ppl=1.02, grad_norm=0.61, lr=7.81e-05 | |
| 2025-11-13 11:39:18,468 - INFO - Epoch 1 Step 3960 (Global: 3960): loss=0.0169, ppl=1.02, grad_norm=0.50, lr=7.79e-05 | |
| 2025-11-13 11:41:08,984 - INFO - Epoch 1 Step 3970 (Global: 3970): loss=0.0232, ppl=1.02, grad_norm=0.60, lr=7.78e-05 | |
| 2025-11-13 11:42:59,992 - INFO - Epoch 1 Step 3980 (Global: 3980): loss=0.0199, ppl=1.02, grad_norm=0.60, lr=7.77e-05 | |
| 2025-11-13 11:44:50,896 - INFO - Epoch 1 Step 3990 (Global: 3990): loss=0.0170, ppl=1.02, grad_norm=0.54, lr=7.75e-05 | |
| 2025-11-13 11:46:50,716 - INFO - Epoch 1 Step 4000 (Global: 4000): loss=0.0181, ppl=1.02, grad_norm=0.53, lr=7.74e-05 | |
| 2025-11-13 11:46:50,717 - INFO - | |
| Running validation at step 4000... | |
| 2025-11-13 11:52:16,661 - INFO - Validation loss: 0.0220, perplexity: 1.02 | |
| 2025-11-13 11:52:16,661 - INFO - Qualitative metrics (n=5): | |
| 2025-11-13 11:52:16,662 - INFO - BLEU: 0.7634 | |
| 2025-11-13 11:52:16,662 - INFO - METEOR: 0.8473 | |
| 2025-11-13 11:52:16,662 - INFO - Edit Distance: 0.1306 | |
| 2025-11-13 11:52:16,662 - INFO - F-measure: 0.8402 | |
| 2025-11-13 11:52:16,662 - INFO - | |
| ====================================================================== | |
| 2025-11-13 11:52:16,663 - INFO - Qualitative Evaluation Samples: | |
| 2025-11-13 11:52:16,663 - INFO - ====================================================================== | |
| 2025-11-13 11:52:16,663 - INFO - | |
| Sample 1 (ID: sample_141920_chunk_1): | |
| 2025-11-13 11:52:16,663 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 11:52:16,663 - INFO - Generated: ' gave it Q4 stars out of five and said that "Perhaps [the album\'s] seemingly illogical sequencing of songs makes sense if they wish to lure their audience into thinking it\'s as-you-were. But it\'s not:...' | |
| 2025-11-13 11:52:16,663 - INFO - Ground Truth: 'Q gave it four stars out of five and said that "Perhaps [the album\'s] seemingly illogical sequencing of songs makes sense if they wish to lure their audience into thinking it\'s as-you-were. But it\'s n...' | |
| 2025-11-13 11:52:16,663 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 11:52:16,663 - INFO - | |
| Sample 2 (ID: sample_170543_chunk_2): | |
| 2025-11-13 11:52:16,663 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 11:52:16,664 - INFO - Generated: ', was Sirene Abou-Chakra, a Lebanese-American student who led the Arab-American Student Association. Other members included the woman president of the Michigan Student Assembly; the leader of Army ROT...' | |
| 2025-11-13 11:52:16,664 - INFO - Ground Truth: ', was Sirene Abou-Chakra, a Lebanese-American student who led the Arab-American Student Association. Other members included the woman president of the Michigan Student Assembly; the leader of Army ROT...' | |
| 2025-11-13 11:52:16,664 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 11:52:16,664 - INFO - | |
| Sample 3 (ID: sample_107152_chunk_9): | |
| 2025-11-13 11:52:16,664 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 11:52:16,664 - INFO - Generated: " at the meeting Laymah headed. His vision of agent is a giant choix, and he can have the lip positions to immobilise his opponents if they get in the eye. Ola falls for the phrase, but Bell can't keep..." | |
| 2025-11-13 11:52:16,664 - INFO - Ground Truth: ' at the meeting Laymia headed. His weapon of choice is a giant ax, and he has the power to immobilise his opponents if they look at him in the eye. Oga falls for the trick, but Beel stops the ax and b...' | |
| 2025-11-13 11:52:16,664 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 11:52:16,665 - INFO - | |
| Sample 4 (ID: sample_069148_chunk_0): | |
| 2025-11-13 11:52:16,665 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 11:52:16,665 - INFO - Generated: '# Oriya (Unicode block)\nOriya is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0B01.....' | |
| 2025-11-13 11:52:16,666 - INFO - Ground Truth: '# Oriya (Unicode block)\nOriya is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0B01.....' | |
| 2025-11-13 11:52:16,666 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 11:52:16,666 - INFO - | |
| Sample 5 (ID: sample_103176_chunk_4): | |
| 2025-11-13 11:52:16,666 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 11:52:16,667 - INFO - Generated: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Maxis Redwood Shores ...' | |
| 2025-11-13 11:52:16,667 - INFO - Ground Truth: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Maxis Redwood Shores ...' | |
| 2025-11-13 11:52:16,667 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 11:52:16,668 - INFO - | |
| Qualitative samples saved to: outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/qualitative_step_4000.jsonl | |
| 2025-11-13 11:53:02,703 - INFO - Saved checkpoint to outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/best_checkpoint.pt | |
| 2025-11-13 11:53:02,719 - INFO - New best validation loss: 0.0220, perplexity: 1.02 | |
| 2025-11-13 11:54:53,939 - INFO - Epoch 1 Step 4010 (Global: 4010): loss=0.0218, ppl=1.02, grad_norm=0.60, lr=7.72e-05 | |
| 2025-11-13 11:56:44,877 - INFO - Epoch 1 Step 4020 (Global: 4020): loss=0.0177, ppl=1.02, grad_norm=0.52, lr=7.71e-05 | |
| 2025-11-13 11:58:35,666 - INFO - Epoch 1 Step 4030 (Global: 4030): loss=0.0196, ppl=1.02, grad_norm=0.57, lr=7.70e-05 | |
| 2025-11-13 12:00:36,336 - INFO - Epoch 1 Step 4040 (Global: 4040): loss=0.0193, ppl=1.02, grad_norm=0.62, lr=7.68e-05 | |
| 2025-11-13 12:02:27,382 - INFO - Epoch 1 Step 4050 (Global: 4050): loss=0.0184, ppl=1.02, grad_norm=0.55, lr=7.67e-05 | |
| 2025-11-13 12:04:18,495 - INFO - Epoch 1 Step 4060 (Global: 4060): loss=0.0209, ppl=1.02, grad_norm=0.58, lr=7.65e-05 | |
| 2025-11-13 12:06:09,244 - INFO - Epoch 1 Step 4070 (Global: 4070): loss=0.0176, ppl=1.02, grad_norm=0.50, lr=7.64e-05 | |
| 2025-11-13 12:08:12,022 - INFO - Epoch 1 Step 4080 (Global: 4080): loss=0.0170, ppl=1.02, grad_norm=0.56, lr=7.62e-05 | |
| 2025-11-13 12:10:04,509 - INFO - Epoch 1 Step 4090 (Global: 4090): loss=0.0216, ppl=1.02, grad_norm=0.65, lr=7.61e-05 | |
| 2025-11-13 12:11:56,280 - INFO - Epoch 1 Step 4100 (Global: 4100): loss=0.0173, ppl=1.02, grad_norm=0.54, lr=7.60e-05 | |
| 2025-11-13 12:13:47,700 - INFO - Epoch 1 Step 4110 (Global: 4110): loss=0.0178, ppl=1.02, grad_norm=0.63, lr=7.58e-05 | |
| 2025-11-13 12:15:48,262 - INFO - Epoch 1 Step 4120 (Global: 4120): loss=0.0184, ppl=1.02, grad_norm=0.57, lr=7.57e-05 | |
| 2025-11-13 12:17:39,300 - INFO - Epoch 1 Step 4130 (Global: 4130): loss=0.0158, ppl=1.02, grad_norm=0.51, lr=7.55e-05 | |
| 2025-11-13 12:19:30,118 - INFO - Epoch 1 Step 4140 (Global: 4140): loss=0.0147, ppl=1.01, grad_norm=0.49, lr=7.54e-05 | |
| 2025-11-13 12:21:21,190 - INFO - Epoch 1 Step 4150 (Global: 4150): loss=0.0199, ppl=1.02, grad_norm=0.58, lr=7.52e-05 | |
| 2025-11-13 12:23:21,655 - INFO - Epoch 1 Step 4160 (Global: 4160): loss=0.0215, ppl=1.02, grad_norm=0.62, lr=7.51e-05 | |
| 2025-11-13 12:25:12,811 - INFO - Epoch 1 Step 4170 (Global: 4170): loss=0.0222, ppl=1.02, grad_norm=0.62, lr=7.49e-05 | |
| 2025-11-13 12:27:06,194 - INFO - Epoch 1 Step 4180 (Global: 4180): loss=0.0185, ppl=1.02, grad_norm=0.59, lr=7.48e-05 | |
| 2025-11-13 12:28:57,646 - INFO - Epoch 1 Step 4190 (Global: 4190): loss=0.0168, ppl=1.02, grad_norm=0.54, lr=7.47e-05 | |
| 2025-11-13 12:30:58,210 - INFO - Epoch 1 Step 4200 (Global: 4200): loss=0.0157, ppl=1.02, grad_norm=0.50, lr=7.45e-05 | |
| 2025-11-13 12:32:51,101 - INFO - Epoch 1 Step 4210 (Global: 4210): loss=0.0195, ppl=1.02, grad_norm=0.61, lr=7.44e-05 | |
| 2025-11-13 12:34:43,396 - INFO - Epoch 1 Step 4220 (Global: 4220): loss=0.0267, ppl=1.03, grad_norm=0.68, lr=7.42e-05 | |
| 2025-11-13 12:36:36,001 - INFO - Epoch 1 Step 4230 (Global: 4230): loss=0.0172, ppl=1.02, grad_norm=0.56, lr=7.41e-05 | |
| 2025-11-13 12:38:37,792 - INFO - Epoch 1 Step 4240 (Global: 4240): loss=0.0213, ppl=1.02, grad_norm=0.59, lr=7.39e-05 | |
| 2025-11-13 12:40:29,666 - INFO - Epoch 1 Step 4250 (Global: 4250): loss=0.0202, ppl=1.02, grad_norm=0.59, lr=7.38e-05 | |
| 2025-11-13 12:42:20,979 - INFO - Epoch 1 Step 4260 (Global: 4260): loss=0.0148, ppl=1.01, grad_norm=0.49, lr=7.36e-05 | |
| 2025-11-13 12:44:13,238 - INFO - Epoch 1 Step 4270 (Global: 4270): loss=0.0175, ppl=1.02, grad_norm=0.54, lr=7.35e-05 | |
| 2025-11-13 12:46:13,646 - INFO - Epoch 1 Step 4280 (Global: 4280): loss=0.0177, ppl=1.02, grad_norm=0.54, lr=7.33e-05 | |
| 2025-11-13 12:48:04,113 - INFO - Epoch 1 Step 4290 (Global: 4290): loss=0.0198, ppl=1.02, grad_norm=0.59, lr=7.32e-05 | |
| 2025-11-13 12:49:55,328 - INFO - Epoch 1 Step 4300 (Global: 4300): loss=0.0160, ppl=1.02, grad_norm=0.50, lr=7.30e-05 | |
| 2025-11-13 12:51:46,652 - INFO - Epoch 1 Step 4310 (Global: 4310): loss=0.0181, ppl=1.02, grad_norm=0.60, lr=7.29e-05 | |
| 2025-11-13 12:53:47,417 - INFO - Epoch 1 Step 4320 (Global: 4320): loss=0.0207, ppl=1.02, grad_norm=0.61, lr=7.27e-05 | |
| 2025-11-13 12:55:41,111 - INFO - Epoch 1 Step 4330 (Global: 4330): loss=0.0144, ppl=1.01, grad_norm=0.47, lr=7.26e-05 | |
| 2025-11-13 12:57:33,052 - INFO - Epoch 1 Step 4340 (Global: 4340): loss=0.0196, ppl=1.02, grad_norm=0.57, lr=7.24e-05 | |
| 2025-11-13 12:59:26,063 - INFO - Epoch 1 Step 4350 (Global: 4350): loss=0.0161, ppl=1.02, grad_norm=0.51, lr=7.23e-05 | |
| 2025-11-13 13:01:29,483 - INFO - Epoch 1 Step 4360 (Global: 4360): loss=0.0183, ppl=1.02, grad_norm=0.55, lr=7.21e-05 | |
| 2025-11-13 13:03:22,708 - INFO - Epoch 1 Step 4370 (Global: 4370): loss=0.0191, ppl=1.02, grad_norm=0.57, lr=7.20e-05 | |
| 2025-11-13 13:05:14,780 - INFO - Epoch 1 Step 4380 (Global: 4380): loss=0.0223, ppl=1.02, grad_norm=0.57, lr=7.18e-05 | |
| 2025-11-13 13:07:06,122 - INFO - Epoch 1 Step 4390 (Global: 4390): loss=0.0187, ppl=1.02, grad_norm=0.55, lr=7.17e-05 | |
| 2025-11-13 13:09:06,501 - INFO - Epoch 1 Step 4400 (Global: 4400): loss=0.0209, ppl=1.02, grad_norm=0.55, lr=7.15e-05 | |
| 2025-11-13 13:11:00,878 - INFO - Epoch 1 Step 4410 (Global: 4410): loss=0.0188, ppl=1.02, grad_norm=0.56, lr=7.14e-05 | |
| 2025-11-13 13:12:54,601 - INFO - Epoch 1 Step 4420 (Global: 4420): loss=0.0195, ppl=1.02, grad_norm=0.54, lr=7.12e-05 | |
| 2025-11-13 13:14:47,566 - INFO - Epoch 1 Step 4430 (Global: 4430): loss=0.0173, ppl=1.02, grad_norm=0.53, lr=7.11e-05 | |
| 2025-11-13 13:16:50,071 - INFO - Epoch 1 Step 4440 (Global: 4440): loss=0.0207, ppl=1.02, grad_norm=0.62, lr=7.09e-05 | |
| 2025-11-13 13:18:42,269 - INFO - Epoch 1 Step 4450 (Global: 4450): loss=0.0163, ppl=1.02, grad_norm=0.54, lr=7.08e-05 | |
| 2025-11-13 13:20:34,594 - INFO - Epoch 1 Step 4460 (Global: 4460): loss=0.0189, ppl=1.02, grad_norm=0.57, lr=7.06e-05 | |
| 2025-11-13 13:22:27,079 - INFO - Epoch 1 Step 4470 (Global: 4470): loss=0.0149, ppl=1.02, grad_norm=0.58, lr=7.05e-05 | |
| 2025-11-13 13:24:28,767 - INFO - Epoch 1 Step 4480 (Global: 4480): loss=0.0185, ppl=1.02, grad_norm=0.61, lr=7.03e-05 | |
| 2025-11-13 13:26:21,490 - INFO - Epoch 1 Step 4490 (Global: 4490): loss=0.0181, ppl=1.02, grad_norm=0.57, lr=7.02e-05 | |
| 2025-11-13 13:28:14,334 - INFO - Epoch 1 Step 4500 (Global: 4500): loss=0.0160, ppl=1.02, grad_norm=0.53, lr=7.00e-05 | |
| 2025-11-13 13:28:14,339 - INFO - | |
| Running validation at step 4500... | |
| 2025-11-13 13:33:49,060 - INFO - Validation loss: 0.0173, perplexity: 1.02 | |
| 2025-11-13 13:33:49,060 - INFO - Qualitative metrics (n=5): | |
| 2025-11-13 13:33:49,061 - INFO - BLEU: 0.8182 | |
| 2025-11-13 13:33:49,061 - INFO - METEOR: 0.8510 | |
| 2025-11-13 13:33:49,061 - INFO - Edit Distance: 0.1193 | |
| 2025-11-13 13:33:49,061 - INFO - F-measure: 0.8703 | |
| 2025-11-13 13:33:49,061 - INFO - | |
| ====================================================================== | |
| 2025-11-13 13:33:49,061 - INFO - Qualitative Evaluation Samples: | |
| 2025-11-13 13:33:49,061 - INFO - ====================================================================== | |
| 2025-11-13 13:33:49,061 - INFO - | |
| Sample 1 (ID: sample_141920_chunk_1): | |
| 2025-11-13 13:33:49,061 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 13:33:49,061 - INFO - Generated: ' gave Q it four stars out of five and said that "Perhaps [the album\'s] seemingly illogical sequencing of songs means keep its desire to wish you that queer imagery into their it\'s becomes like-Water!....' | |
| 2025-11-13 13:33:49,062 - INFO - Ground Truth: 'Q gave it four stars out of five and said that "Perhaps [the album\'s] seemingly illogical sequencing of songs makes sense if they wish to lure their audience into thinking it\'s as-you-were. But it\'s n...' | |
| 2025-11-13 13:33:49,062 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 13:33:49,062 - INFO - | |
| Sample 2 (ID: sample_170543_chunk_2): | |
| 2025-11-13 13:33:49,062 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 13:33:49,062 - INFO - Generated: ', Sire was Aboune-Chakra, a Lebanese-American student who led the Arab-American Student Association. Other members included the woman president of the Michigan Student Assembly; the leader of Army ROT...' | |
| 2025-11-13 13:33:49,063 - INFO - Ground Truth: ', was Sirene Abou-Chakra, a Lebanese-American student who led the Arab-American Student Association. Other members included the woman president of the Michigan Student Assembly; the leader of Army ROT...' | |
| 2025-11-13 13:33:49,063 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 13:33:49,063 - INFO - | |
| Sample 3 (ID: sample_107152_chunk_9): | |
| 2025-11-13 13:33:49,063 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 13:33:49,063 - INFO - Generated: ' at the meeting Laymah headed. His investigator of way a giant is cast, he and a hand is the prosthetic power to immobilises him at what they look in. Ethe oil ga foronda, take the Bell but gets cri a...' | |
| 2025-11-13 13:33:49,063 - INFO - Ground Truth: ' at the meeting Laymia headed. His weapon of choice is a giant ax, and he has the power to immobilise his opponents if they look at him in the eye. Oga falls for the trick, but Beel stops the ax and b...' | |
| 2025-11-13 13:33:49,064 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 13:33:49,064 - INFO - | |
| Sample 4 (ID: sample_069148_chunk_0): | |
| 2025-11-13 13:33:49,064 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 13:33:49,064 - INFO - Generated: '# Oriya (Unicode block)\nOriya is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0B01.....' | |
| 2025-11-13 13:33:49,064 - INFO - Ground Truth: '# Oriya (Unicode block)\nOriya is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0B01.....' | |
| 2025-11-13 13:33:49,064 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 13:33:49,065 - INFO - | |
| Sample 5 (ID: sample_103176_chunk_4): | |
| 2025-11-13 13:33:49,065 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 13:33:49,065 - INFO - Generated: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Maxis Redwood Shores ...' | |
| 2025-11-13 13:33:49,066 - INFO - Ground Truth: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Maxis Redwood Shores ...' | |
| 2025-11-13 13:33:49,066 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 13:33:49,066 - INFO - | |
| Qualitative samples saved to: outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/qualitative_step_4500.jsonl | |
| 2025-11-13 13:34:38,303 - INFO - Saved checkpoint to outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/best_checkpoint.pt | |
| 2025-11-13 13:34:38,318 - INFO - New best validation loss: 0.0173, perplexity: 1.02 | |
| 2025-11-13 13:36:31,735 - INFO - Epoch 1 Step 4510 (Global: 4510): loss=0.0147, ppl=1.01, grad_norm=0.50, lr=6.99e-05 | |
| 2025-11-13 13:38:35,183 - INFO - Epoch 1 Step 4520 (Global: 4520): loss=0.0131, ppl=1.01, grad_norm=0.48, lr=6.97e-05 | |
| 2025-11-13 13:40:27,099 - INFO - Epoch 1 Step 4530 (Global: 4530): loss=0.0173, ppl=1.02, grad_norm=0.55, lr=6.96e-05 | |
| 2025-11-13 13:42:19,661 - INFO - Epoch 1 Step 4540 (Global: 4540): loss=0.0169, ppl=1.02, grad_norm=0.54, lr=6.94e-05 | |
| 2025-11-13 13:44:12,405 - INFO - Epoch 1 Step 4550 (Global: 4550): loss=0.0131, ppl=1.01, grad_norm=0.47, lr=6.92e-05 | |
| 2025-11-13 13:46:14,349 - INFO - Epoch 1 Step 4560 (Global: 4560): loss=0.0188, ppl=1.02, grad_norm=0.69, lr=6.91e-05 | |
| 2025-11-13 13:48:09,508 - INFO - Epoch 1 Step 4570 (Global: 4570): loss=0.0174, ppl=1.02, grad_norm=0.54, lr=6.89e-05 | |
| 2025-11-13 13:50:02,108 - INFO - Epoch 1 Step 4580 (Global: 4580): loss=0.0145, ppl=1.01, grad_norm=0.49, lr=6.88e-05 | |
| 2025-11-13 13:51:56,663 - INFO - Epoch 1 Step 4590 (Global: 4590): loss=0.0139, ppl=1.01, grad_norm=0.48, lr=6.86e-05 | |
| 2025-11-13 13:54:00,352 - INFO - Epoch 1 Step 4600 (Global: 4600): loss=0.0139, ppl=1.01, grad_norm=0.49, lr=6.85e-05 | |
| 2025-11-13 13:55:58,544 - INFO - Epoch 1 Step 4610 (Global: 4610): loss=0.0198, ppl=1.02, grad_norm=0.59, lr=6.83e-05 | |
| 2025-11-13 13:57:53,194 - INFO - Epoch 1 Step 4620 (Global: 4620): loss=0.0156, ppl=1.02, grad_norm=0.55, lr=6.82e-05 | |
| 2025-11-13 13:59:47,293 - INFO - Epoch 1 Step 4630 (Global: 4630): loss=0.0124, ppl=1.01, grad_norm=0.47, lr=6.80e-05 | |
| 2025-11-13 14:01:50,314 - INFO - Epoch 1 Step 4640 (Global: 4640): loss=0.0173, ppl=1.02, grad_norm=0.76, lr=6.78e-05 | |
| 2025-11-13 14:03:44,005 - INFO - Epoch 1 Step 4650 (Global: 4650): loss=0.0196, ppl=1.02, grad_norm=0.63, lr=6.77e-05 | |
| 2025-11-13 14:05:37,635 - INFO - Epoch 1 Step 4660 (Global: 4660): loss=0.0204, ppl=1.02, grad_norm=0.62, lr=6.75e-05 | |
| 2025-11-13 14:07:31,148 - INFO - Epoch 1 Step 4670 (Global: 4670): loss=0.0171, ppl=1.02, grad_norm=0.55, lr=6.74e-05 | |
| 2025-11-13 14:09:33,669 - INFO - Epoch 1 Step 4680 (Global: 4680): loss=0.0145, ppl=1.01, grad_norm=0.51, lr=6.72e-05 | |
| 2025-11-13 14:11:26,625 - INFO - Epoch 1 Step 4690 (Global: 4690): loss=0.0166, ppl=1.02, grad_norm=0.53, lr=6.71e-05 | |
| 2025-11-13 14:13:20,264 - INFO - Epoch 1 Step 4700 (Global: 4700): loss=0.0186, ppl=1.02, grad_norm=0.56, lr=6.69e-05 | |
| 2025-11-13 14:15:14,980 - INFO - Epoch 1 Step 4710 (Global: 4710): loss=0.0182, ppl=1.02, grad_norm=0.52, lr=6.67e-05 | |
| 2025-11-13 14:17:18,945 - INFO - Epoch 1 Step 4720 (Global: 4720): loss=0.0127, ppl=1.01, grad_norm=0.44, lr=6.66e-05 | |
| 2025-11-13 14:19:17,210 - INFO - Epoch 1 Step 4730 (Global: 4730): loss=0.0182, ppl=1.02, grad_norm=0.56, lr=6.64e-05 | |
| 2025-11-13 14:21:09,430 - INFO - Epoch 1 Step 4740 (Global: 4740): loss=0.0160, ppl=1.02, grad_norm=0.58, lr=6.63e-05 | |
| 2025-11-13 14:23:02,323 - INFO - Epoch 1 Step 4750 (Global: 4750): loss=0.0126, ppl=1.01, grad_norm=0.48, lr=6.61e-05 | |
| 2025-11-13 14:25:03,331 - INFO - Epoch 1 Step 4760 (Global: 4760): loss=0.0132, ppl=1.01, grad_norm=0.52, lr=6.60e-05 | |
| 2025-11-13 14:26:54,569 - INFO - Epoch 1 Step 4770 (Global: 4770): loss=0.0131, ppl=1.01, grad_norm=0.51, lr=6.58e-05 | |
| 2025-11-13 14:28:49,835 - INFO - Epoch 1 Step 4780 (Global: 4780): loss=0.0154, ppl=1.02, grad_norm=0.50, lr=6.56e-05 | |
| 2025-11-13 14:30:42,703 - INFO - Epoch 1 Step 4790 (Global: 4790): loss=0.0171, ppl=1.02, grad_norm=0.54, lr=6.55e-05 | |
| 2025-11-13 14:32:43,767 - INFO - Epoch 1 Step 4800 (Global: 4800): loss=0.0172, ppl=1.02, grad_norm=0.53, lr=6.53e-05 | |
| 2025-11-13 14:34:34,716 - INFO - Epoch 1 Step 4810 (Global: 4810): loss=0.0153, ppl=1.02, grad_norm=0.52, lr=6.52e-05 | |
| 2025-11-13 14:36:25,566 - INFO - Epoch 1 Step 4820 (Global: 4820): loss=0.0125, ppl=1.01, grad_norm=0.48, lr=6.50e-05 | |
| 2025-11-13 14:38:17,141 - INFO - Epoch 1 Step 4830 (Global: 4830): loss=0.0150, ppl=1.02, grad_norm=0.55, lr=6.48e-05 | |
| 2025-11-13 14:40:18,230 - INFO - Epoch 1 Step 4840 (Global: 4840): loss=0.0122, ppl=1.01, grad_norm=0.51, lr=6.47e-05 | |
| 2025-11-13 14:42:10,486 - INFO - Epoch 1 Step 4850 (Global: 4850): loss=0.0142, ppl=1.01, grad_norm=0.51, lr=6.45e-05 | |
| 2025-11-13 14:44:07,383 - INFO - Epoch 1 Step 4860 (Global: 4860): loss=0.0121, ppl=1.01, grad_norm=0.49, lr=6.44e-05 | |
| 2025-11-13 14:46:05,226 - INFO - Epoch 1 Step 4870 (Global: 4870): loss=0.0143, ppl=1.01, grad_norm=0.49, lr=6.42e-05 | |
| 2025-11-13 14:48:11,803 - INFO - Epoch 1 Step 4880 (Global: 4880): loss=0.0152, ppl=1.02, grad_norm=0.50, lr=6.40e-05 | |
| 2025-11-13 14:50:08,837 - INFO - Epoch 1 Step 4890 (Global: 4890): loss=0.0136, ppl=1.01, grad_norm=0.48, lr=6.39e-05 | |
| 2025-11-13 14:52:08,812 - INFO - Epoch 1 Step 4900 (Global: 4900): loss=0.0106, ppl=1.01, grad_norm=0.43, lr=6.37e-05 | |
| 2025-11-13 14:54:00,332 - INFO - Epoch 1 Step 4910 (Global: 4910): loss=0.0150, ppl=1.02, grad_norm=0.52, lr=6.35e-05 | |
| 2025-11-13 14:56:00,484 - INFO - Epoch 1 Step 4920 (Global: 4920): loss=0.0171, ppl=1.02, grad_norm=0.58, lr=6.34e-05 | |
| 2025-11-13 14:57:51,342 - INFO - Epoch 1 Step 4930 (Global: 4930): loss=0.0132, ppl=1.01, grad_norm=0.45, lr=6.32e-05 | |
| 2025-11-13 14:59:42,613 - INFO - Epoch 1 Step 4940 (Global: 4940): loss=0.0171, ppl=1.02, grad_norm=0.55, lr=6.31e-05 | |
| 2025-11-13 15:01:34,238 - INFO - Epoch 1 Step 4950 (Global: 4950): loss=0.0119, ppl=1.01, grad_norm=0.45, lr=6.29e-05 | |
| 2025-11-13 15:03:35,994 - INFO - Epoch 1 Step 4960 (Global: 4960): loss=0.0118, ppl=1.01, grad_norm=0.49, lr=6.27e-05 | |
| 2025-11-13 15:05:27,273 - INFO - Epoch 1 Step 4970 (Global: 4970): loss=0.0166, ppl=1.02, grad_norm=0.55, lr=6.26e-05 | |
| 2025-11-13 15:07:18,231 - INFO - Epoch 1 Step 4980 (Global: 4980): loss=0.0150, ppl=1.02, grad_norm=0.49, lr=6.24e-05 | |
| 2025-11-13 15:09:09,365 - INFO - Epoch 1 Step 4990 (Global: 4990): loss=0.0129, ppl=1.01, grad_norm=0.47, lr=6.23e-05 | |
| 2025-11-13 15:11:09,259 - INFO - Epoch 1 Step 5000 (Global: 5000): loss=0.0148, ppl=1.01, grad_norm=0.53, lr=6.21e-05 | |
| 2025-11-13 15:11:09,261 - INFO - | |
| Running validation at step 5000... | |
| 2025-11-13 15:16:36,784 - INFO - Validation loss: 0.0137, perplexity: 1.01 | |
| 2025-11-13 15:16:36,784 - INFO - Qualitative metrics (n=5): | |
| 2025-11-13 15:16:36,784 - INFO - BLEU: 0.9890 | |
| 2025-11-13 15:16:36,785 - INFO - METEOR: 0.9949 | |
| 2025-11-13 15:16:36,785 - INFO - Edit Distance: 0.0031 | |
| 2025-11-13 15:16:36,785 - INFO - F-measure: 0.9929 | |
| 2025-11-13 15:16:36,785 - INFO - | |
| ====================================================================== | |
| 2025-11-13 15:16:36,785 - INFO - Qualitative Evaluation Samples: | |
| 2025-11-13 15:16:36,785 - INFO - ====================================================================== | |
| 2025-11-13 15:16:36,785 - INFO - | |
| Sample 1 (ID: sample_141920_chunk_1): | |
| 2025-11-13 15:16:36,785 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 15:16:36,785 - INFO - Generated: ' gave Q it four stars out of five and said that "Perhaps [the album\'s seemingly] illogical sequencing of songs makes sense if they wish to lure their audience into thinking it\'s as-you-were. But it\'s ...' | |
| 2025-11-13 15:16:36,785 - INFO - Ground Truth: 'Q gave it four stars out of five and said that "Perhaps [the album\'s] seemingly illogical sequencing of songs makes sense if they wish to lure their audience into thinking it\'s as-you-were. But it\'s n...' | |
| 2025-11-13 15:16:36,785 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 15:16:36,786 - INFO - | |
| Sample 2 (ID: sample_170543_chunk_2): | |
| 2025-11-13 15:16:36,786 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 15:16:36,786 - INFO - Generated: ', Sire was Aboune-Chakra, a Lebanese-American student who led the Arab-American Student Association. Other members included the woman president of the Michigan Student Assembly; the leader of Army ROT...' | |
| 2025-11-13 15:16:36,786 - INFO - Ground Truth: ', was Sirene Abou-Chakra, a Lebanese-American student who led the Arab-American Student Association. Other members included the woman president of the Michigan Student Assembly; the leader of Army ROT...' | |
| 2025-11-13 15:16:36,786 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 15:16:36,786 - INFO - | |
| Sample 3 (ID: sample_107152_chunk_9): | |
| 2025-11-13 15:16:36,786 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 15:16:36,786 - INFO - Generated: ' at the meeting Laymia headed. His weapon of choice is a giant ax, and he has the power to immobilise his opponents if they look at him in the eye. Oga falls for the trick, but Beel stops the ax and b...' | |
| 2025-11-13 15:16:36,786 - INFO - Ground Truth: ' at the meeting Laymia headed. His weapon of choice is a giant ax, and he has the power to immobilise his opponents if they look at him in the eye. Oga falls for the trick, but Beel stops the ax and b...' | |
| 2025-11-13 15:16:36,786 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 15:16:36,787 - INFO - | |
| Sample 4 (ID: sample_069148_chunk_0): | |
| 2025-11-13 15:16:36,787 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 15:16:36,787 - INFO - Generated: '# Oriya (Unicode block)\nOriya is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0B01.....' | |
| 2025-11-13 15:16:36,787 - INFO - Ground Truth: '# Oriya (Unicode block)\nOriya is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0B01.....' | |
| 2025-11-13 15:16:36,787 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 15:16:36,787 - INFO - | |
| Sample 5 (ID: sample_103176_chunk_4): | |
| 2025-11-13 15:16:36,787 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 15:16:36,787 - INFO - Generated: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Maxis Redwood Shores ...' | |
| 2025-11-13 15:16:36,788 - INFO - Ground Truth: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Maxis Redwood Shores ...' | |
| 2025-11-13 15:16:36,788 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 15:16:36,788 - INFO - | |
| Qualitative samples saved to: outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/qualitative_step_5000.jsonl | |
| 2025-11-13 15:17:19,658 - INFO - Saved checkpoint to outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/best_checkpoint.pt | |
| 2025-11-13 15:17:19,673 - INFO - New best validation loss: 0.0137, perplexity: 1.01 | |
| 2025-11-13 15:19:11,213 - INFO - Epoch 1 Step 5010 (Global: 5010): loss=0.0202, ppl=1.02, grad_norm=0.61, lr=6.19e-05 | |
| 2025-11-13 15:21:02,710 - INFO - Epoch 1 Step 5020 (Global: 5020): loss=0.0112, ppl=1.01, grad_norm=0.42, lr=6.18e-05 | |
| 2025-11-13 15:22:53,641 - INFO - Epoch 1 Step 5030 (Global: 5030): loss=0.0131, ppl=1.01, grad_norm=0.46, lr=6.16e-05 | |
| 2025-11-13 15:24:54,159 - INFO - Epoch 1 Step 5040 (Global: 5040): loss=0.0146, ppl=1.01, grad_norm=0.51, lr=6.14e-05 | |
| 2025-11-13 15:26:45,190 - INFO - Epoch 1 Step 5050 (Global: 5050): loss=0.0111, ppl=1.01, grad_norm=0.45, lr=6.13e-05 | |
| 2025-11-13 15:28:36,463 - INFO - Epoch 1 Step 5060 (Global: 5060): loss=0.0115, ppl=1.01, grad_norm=0.44, lr=6.11e-05 | |
| 2025-11-13 15:30:27,461 - INFO - Epoch 1 Step 5070 (Global: 5070): loss=0.0116, ppl=1.01, grad_norm=0.43, lr=6.10e-05 | |
| 2025-11-13 15:32:28,878 - INFO - Epoch 1 Step 5080 (Global: 5080): loss=0.0139, ppl=1.01, grad_norm=0.49, lr=6.08e-05 | |
| 2025-11-13 15:34:19,714 - INFO - Epoch 1 Step 5090 (Global: 5090): loss=0.0105, ppl=1.01, grad_norm=0.41, lr=6.06e-05 | |
| 2025-11-13 15:36:10,436 - INFO - Epoch 1 Step 5100 (Global: 5100): loss=0.0132, ppl=1.01, grad_norm=0.50, lr=6.05e-05 | |
| 2025-11-13 15:38:03,603 - INFO - Epoch 1 Step 5110 (Global: 5110): loss=0.0116, ppl=1.01, grad_norm=0.43, lr=6.03e-05 | |
| 2025-11-13 15:40:08,523 - INFO - Epoch 1 Step 5120 (Global: 5120): loss=0.0162, ppl=1.02, grad_norm=0.53, lr=6.01e-05 | |
| 2025-11-13 15:42:02,882 - INFO - Epoch 1 Step 5130 (Global: 5130): loss=0.0138, ppl=1.01, grad_norm=0.49, lr=6.00e-05 | |
| 2025-11-13 15:43:57,556 - INFO - Epoch 1 Step 5140 (Global: 5140): loss=0.0178, ppl=1.02, grad_norm=0.53, lr=5.98e-05 | |
| 2025-11-13 15:45:53,003 - INFO - Epoch 1 Step 5150 (Global: 5150): loss=0.0124, ppl=1.01, grad_norm=0.48, lr=5.96e-05 | |
| 2025-11-13 15:47:58,101 - INFO - Epoch 1 Step 5160 (Global: 5160): loss=0.0120, ppl=1.01, grad_norm=0.47, lr=5.95e-05 | |
| 2025-11-13 15:49:52,148 - INFO - Epoch 1 Step 5170 (Global: 5170): loss=0.0114, ppl=1.01, grad_norm=0.45, lr=5.93e-05 | |
| 2025-11-13 15:51:46,070 - INFO - Epoch 1 Step 5180 (Global: 5180): loss=0.0194, ppl=1.02, grad_norm=0.84, lr=5.91e-05 | |
| 2025-11-13 15:53:40,284 - INFO - Epoch 1 Step 5190 (Global: 5190): loss=0.0111, ppl=1.01, grad_norm=0.51, lr=5.90e-05 | |
| 2025-11-13 15:55:43,570 - INFO - Epoch 1 Step 5200 (Global: 5200): loss=0.0109, ppl=1.01, grad_norm=0.46, lr=5.88e-05 | |
| 2025-11-13 15:57:38,018 - INFO - Epoch 1 Step 5210 (Global: 5210): loss=0.0104, ppl=1.01, grad_norm=0.45, lr=5.87e-05 | |
| 2025-11-13 15:59:32,200 - INFO - Epoch 1 Step 5220 (Global: 5220): loss=0.0142, ppl=1.01, grad_norm=0.52, lr=5.85e-05 | |
| 2025-11-13 16:01:27,074 - INFO - Epoch 1 Step 5230 (Global: 5230): loss=0.0090, ppl=1.01, grad_norm=0.42, lr=5.83e-05 | |
| 2025-11-13 16:03:31,359 - INFO - Epoch 1 Step 5240 (Global: 5240): loss=0.0126, ppl=1.01, grad_norm=0.47, lr=5.82e-05 | |
| 2025-11-13 16:05:27,572 - INFO - Epoch 1 Step 5250 (Global: 5250): loss=0.0119, ppl=1.01, grad_norm=0.46, lr=5.80e-05 | |
| 2025-11-13 16:07:26,691 - INFO - Epoch 1 Step 5260 (Global: 5260): loss=0.0137, ppl=1.01, grad_norm=0.48, lr=5.78e-05 | |
| 2025-11-13 16:09:25,695 - INFO - Epoch 1 Step 5270 (Global: 5270): loss=0.0100, ppl=1.01, grad_norm=0.45, lr=5.77e-05 | |
| 2025-11-13 16:11:32,950 - INFO - Epoch 1 Step 5280 (Global: 5280): loss=0.0103, ppl=1.01, grad_norm=0.43, lr=5.75e-05 | |
| 2025-11-13 16:13:28,577 - INFO - Epoch 1 Step 5290 (Global: 5290): loss=0.0125, ppl=1.01, grad_norm=0.48, lr=5.73e-05 | |
| 2025-11-13 16:15:23,833 - INFO - Epoch 1 Step 5300 (Global: 5300): loss=0.0101, ppl=1.01, grad_norm=0.43, lr=5.72e-05 | |
| 2025-11-13 16:17:21,906 - INFO - Epoch 1 Step 5310 (Global: 5310): loss=0.0127, ppl=1.01, grad_norm=0.48, lr=5.70e-05 | |
| 2025-11-13 16:19:28,666 - INFO - Epoch 1 Step 5320 (Global: 5320): loss=0.0134, ppl=1.01, grad_norm=0.50, lr=5.68e-05 | |
| 2025-11-13 16:21:24,384 - INFO - Epoch 1 Step 5330 (Global: 5330): loss=0.0115, ppl=1.01, grad_norm=0.45, lr=5.67e-05 | |
| 2025-11-13 16:23:20,056 - INFO - Epoch 1 Step 5340 (Global: 5340): loss=0.0122, ppl=1.01, grad_norm=0.46, lr=5.65e-05 | |
| 2025-11-13 16:25:14,765 - INFO - Epoch 1 Step 5350 (Global: 5350): loss=0.0150, ppl=1.02, grad_norm=0.52, lr=5.63e-05 | |
| 2025-11-13 16:27:19,766 - INFO - Epoch 1 Step 5360 (Global: 5360): loss=0.0110, ppl=1.01, grad_norm=0.46, lr=5.62e-05 | |
| 2025-11-13 16:29:16,695 - INFO - Epoch 1 Step 5370 (Global: 5370): loss=0.0120, ppl=1.01, grad_norm=0.46, lr=5.60e-05 | |
| 2025-11-13 16:31:14,155 - INFO - Epoch 1 Step 5380 (Global: 5380): loss=0.0136, ppl=1.01, grad_norm=0.48, lr=5.58e-05 | |
| 2025-11-13 16:33:10,383 - INFO - Epoch 1 Step 5390 (Global: 5390): loss=0.0104, ppl=1.01, grad_norm=0.45, lr=5.57e-05 | |
| 2025-11-13 16:35:14,600 - INFO - Epoch 1 Step 5400 (Global: 5400): loss=0.0130, ppl=1.01, grad_norm=0.47, lr=5.55e-05 | |
| 2025-11-13 16:37:08,931 - INFO - Epoch 1 Step 5410 (Global: 5410): loss=0.0101, ppl=1.01, grad_norm=0.45, lr=5.53e-05 | |
| 2025-11-13 16:39:03,939 - INFO - Epoch 1 Step 5420 (Global: 5420): loss=0.0140, ppl=1.01, grad_norm=0.45, lr=5.52e-05 | |
| 2025-11-13 16:40:59,001 - INFO - Epoch 1 Step 5430 (Global: 5430): loss=0.0098, ppl=1.01, grad_norm=0.43, lr=5.50e-05 | |
| 2025-11-13 16:43:05,119 - INFO - Epoch 1 Step 5440 (Global: 5440): loss=0.0141, ppl=1.01, grad_norm=0.47, lr=5.48e-05 | |
| 2025-11-13 16:45:03,812 - INFO - Epoch 1 Step 5450 (Global: 5450): loss=0.0109, ppl=1.01, grad_norm=0.47, lr=5.47e-05 | |
| 2025-11-13 16:47:01,848 - INFO - Epoch 1 Step 5460 (Global: 5460): loss=0.0089, ppl=1.01, grad_norm=0.39, lr=5.45e-05 | |
| 2025-11-13 16:48:58,569 - INFO - Epoch 1 Step 5470 (Global: 5470): loss=0.0134, ppl=1.01, grad_norm=0.50, lr=5.43e-05 | |
| 2025-11-13 16:51:07,334 - INFO - Epoch 1 Step 5480 (Global: 5480): loss=0.0100, ppl=1.01, grad_norm=0.42, lr=5.42e-05 | |
| 2025-11-13 16:53:04,257 - INFO - Epoch 1 Step 5490 (Global: 5490): loss=0.0120, ppl=1.01, grad_norm=0.47, lr=5.40e-05 | |
| 2025-11-13 16:54:59,593 - INFO - Epoch 1 Step 5500 (Global: 5500): loss=0.0130, ppl=1.01, grad_norm=0.47, lr=5.38e-05 | |
| 2025-11-13 16:54:59,596 - INFO - | |
| Running validation at step 5500... | |
| 2025-11-13 17:00:43,818 - INFO - Validation loss: 0.0116, perplexity: 1.01 | |
| 2025-11-13 17:00:43,819 - INFO - Qualitative metrics (n=5): | |
| 2025-11-13 17:00:43,819 - INFO - BLEU: 0.8985 | |
| 2025-11-13 17:00:43,819 - INFO - METEOR: 0.9519 | |
| 2025-11-13 17:00:43,819 - INFO - Edit Distance: 0.0504 | |
| 2025-11-13 17:00:43,819 - INFO - F-measure: 0.9473 | |
| 2025-11-13 17:00:43,819 - INFO - | |
| ====================================================================== | |
| 2025-11-13 17:00:43,820 - INFO - Qualitative Evaluation Samples: | |
| 2025-11-13 17:00:43,820 - INFO - ====================================================================== | |
| 2025-11-13 17:00:43,820 - INFO - | |
| Sample 1 (ID: sample_141920_chunk_1): | |
| 2025-11-13 17:00:43,820 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 17:00:43,820 - INFO - Generated: ' gave it Q4 stars out of five and said that "Perhaps [the album\'s] seemingly illogical sequencing of songs makes sense if they wish to lure their audience into thinking it\'s as-you-were. But it\'s not:...' | |
| 2025-11-13 17:00:43,820 - INFO - Ground Truth: 'Q gave it four stars out of five and said that "Perhaps [the album\'s] seemingly illogical sequencing of songs makes sense if they wish to lure their audience into thinking it\'s as-you-were. But it\'s n...' | |
| 2025-11-13 17:00:43,820 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 17:00:43,820 - INFO - | |
| Sample 2 (ID: sample_170543_chunk_2): | |
| 2025-11-13 17:00:43,820 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 17:00:43,821 - INFO - Generated: ', Sire was neou-AchABA, Black-Americans toolute the Korean half a Arab-American Student. Members who associated the incident woman the President of Michigan the Shelby team; and Leader of Army ROTC, t...' | |
| 2025-11-13 17:00:43,821 - INFO - Ground Truth: ', was Sirene Abou-Chakra, a Lebanese-American student who led the Arab-American Student Association. Other members included the woman president of the Michigan Student Assembly; the leader of Army ROT...' | |
| 2025-11-13 17:00:43,821 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 17:00:43,821 - INFO - | |
| Sample 3 (ID: sample_107152_chunk_9): | |
| 2025-11-13 17:00:43,821 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 17:00:43,821 - INFO - Generated: ' at the meeting Laymia headed. His weapon of choice is a giant ax, and he has the power to immobilise his opponents if they look at him in the eye. Oga falls for the trick, but Beel stops the ax and b...' | |
| 2025-11-13 17:00:43,821 - INFO - Ground Truth: ' at the meeting Laymia headed. His weapon of choice is a giant ax, and he has the power to immobilise his opponents if they look at him in the eye. Oga falls for the trick, but Beel stops the ax and b...' | |
| 2025-11-13 17:00:43,821 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 17:00:43,821 - INFO - | |
| Sample 4 (ID: sample_069148_chunk_0): | |
| 2025-11-13 17:00:43,821 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 17:00:43,822 - INFO - Generated: '# Oriya (Unicode block)\nOriya is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0B01.....' | |
| 2025-11-13 17:00:43,822 - INFO - Ground Truth: '# Oriya (Unicode block)\nOriya is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0B01.....' | |
| 2025-11-13 17:00:43,822 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 17:00:43,822 - INFO - | |
| Sample 5 (ID: sample_103176_chunk_4): | |
| 2025-11-13 17:00:43,822 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 17:00:43,822 - INFO - Generated: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Maxis Redwood Shores ...' | |
| 2025-11-13 17:00:43,822 - INFO - Ground Truth: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Maxis Redwood Shores ...' | |
| 2025-11-13 17:00:43,823 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 17:00:43,824 - INFO - | |
| Qualitative samples saved to: outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/qualitative_step_5500.jsonl | |
| 2025-11-13 17:01:29,143 - INFO - Saved checkpoint to outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/best_checkpoint.pt | |
| 2025-11-13 17:01:29,159 - INFO - New best validation loss: 0.0116, perplexity: 1.01 | |
| 2025-11-13 17:03:24,240 - INFO - Epoch 1 Step 5510 (Global: 5510): loss=0.0118, ppl=1.01, grad_norm=0.46, lr=5.37e-05 | |
| 2025-11-13 17:05:29,983 - INFO - Epoch 1 Step 5520 (Global: 5520): loss=0.0111, ppl=1.01, grad_norm=0.48, lr=5.35e-05 | |
| 2025-11-13 17:07:23,019 - INFO - Epoch 1 Step 5530 (Global: 5530): loss=0.0119, ppl=1.01, grad_norm=0.48, lr=5.33e-05 | |
| 2025-11-13 17:09:16,122 - INFO - Epoch 1 Step 5540 (Global: 5540): loss=0.0089, ppl=1.01, grad_norm=0.43, lr=5.32e-05 | |
| 2025-11-13 17:11:20,119 - INFO - Epoch 1 Step 5550 (Global: 5550): loss=0.0116, ppl=1.01, grad_norm=0.46, lr=5.30e-05 | |
| 2025-11-13 17:13:12,764 - INFO - Epoch 1 Step 5560 (Global: 5560): loss=0.0151, ppl=1.02, grad_norm=0.61, lr=5.28e-05 | |
| 2025-11-13 17:15:05,294 - INFO - Epoch 1 Step 5570 (Global: 5570): loss=0.0098, ppl=1.01, grad_norm=0.43, lr=5.27e-05 | |
| 2025-11-13 17:16:57,278 - INFO - Epoch 1 Step 5580 (Global: 5580): loss=0.0109, ppl=1.01, grad_norm=0.43, lr=5.25e-05 | |
| 2025-11-13 17:18:48,170 - INFO - Epoch 1 Step 5590 (Global: 5590): loss=0.0088, ppl=1.01, grad_norm=0.39, lr=5.23e-05 | |
| 2025-11-13 17:20:50,363 - INFO - Epoch 1 Step 5600 (Global: 5600): loss=0.0110, ppl=1.01, grad_norm=0.55, lr=5.22e-05 | |
| 2025-11-13 17:22:42,986 - INFO - Epoch 1 Step 5610 (Global: 5610): loss=0.0098, ppl=1.01, grad_norm=0.44, lr=5.20e-05 | |
| 2025-11-13 17:24:34,823 - INFO - Epoch 1 Step 5620 (Global: 5620): loss=0.0088, ppl=1.01, grad_norm=0.40, lr=5.18e-05 | |
| 2025-11-13 17:26:25,636 - INFO - Epoch 1 Step 5630 (Global: 5630): loss=0.0119, ppl=1.01, grad_norm=0.46, lr=5.17e-05 | |
| 2025-11-13 17:28:26,384 - INFO - Epoch 1 Step 5640 (Global: 5640): loss=0.0099, ppl=1.01, grad_norm=0.51, lr=5.15e-05 | |
| 2025-11-13 17:30:17,328 - INFO - Epoch 1 Step 5650 (Global: 5650): loss=0.0117, ppl=1.01, grad_norm=0.46, lr=5.13e-05 | |
| 2025-11-13 17:32:08,359 - INFO - Epoch 1 Step 5660 (Global: 5660): loss=0.0100, ppl=1.01, grad_norm=0.46, lr=5.12e-05 | |
| 2025-11-13 17:34:00,080 - INFO - Epoch 1 Step 5670 (Global: 5670): loss=0.0112, ppl=1.01, grad_norm=0.44, lr=5.10e-05 | |
| 2025-11-13 17:36:00,811 - INFO - Epoch 1 Step 5680 (Global: 5680): loss=0.0108, ppl=1.01, grad_norm=0.46, lr=5.08e-05 | |
| 2025-11-13 17:37:51,566 - INFO - Epoch 1 Step 5690 (Global: 5690): loss=0.0092, ppl=1.01, grad_norm=0.42, lr=5.07e-05 | |
| 2025-11-13 17:39:44,075 - INFO - Epoch 1 Step 5700 (Global: 5700): loss=0.0096, ppl=1.01, grad_norm=0.43, lr=5.05e-05 | |
| 2025-11-13 17:41:35,978 - INFO - Epoch 1 Step 5710 (Global: 5710): loss=0.0153, ppl=1.02, grad_norm=0.54, lr=5.03e-05 | |
| 2025-11-13 17:43:37,218 - INFO - Epoch 1 Step 5720 (Global: 5720): loss=0.0095, ppl=1.01, grad_norm=0.46, lr=5.02e-05 | |
| 2025-11-13 17:45:29,668 - INFO - Epoch 1 Step 5730 (Global: 5730): loss=0.0086, ppl=1.01, grad_norm=0.39, lr=5.00e-05 | |
| 2025-11-13 17:47:22,192 - INFO - Epoch 1 Step 5740 (Global: 5740): loss=0.0101, ppl=1.01, grad_norm=0.40, lr=4.98e-05 | |
| 2025-11-13 17:49:22,730 - INFO - Epoch 1 Step 5750 (Global: 5750): loss=0.0107, ppl=1.01, grad_norm=0.47, lr=4.96e-05 | |
| 2025-11-13 17:51:14,005 - INFO - Epoch 1 Step 5760 (Global: 5760): loss=0.0097, ppl=1.01, grad_norm=0.44, lr=4.95e-05 | |
| 2025-11-13 17:53:06,373 - INFO - Epoch 1 Step 5770 (Global: 5770): loss=0.0112, ppl=1.01, grad_norm=0.48, lr=4.93e-05 | |
| 2025-11-13 17:54:57,652 - INFO - Epoch 1 Step 5780 (Global: 5780): loss=0.0142, ppl=1.01, grad_norm=0.49, lr=4.91e-05 | |
| 2025-11-13 17:56:49,412 - INFO - Epoch 1 Step 5790 (Global: 5790): loss=0.0124, ppl=1.01, grad_norm=0.46, lr=4.90e-05 | |
| 2025-11-13 17:58:50,388 - INFO - Epoch 1 Step 5800 (Global: 5800): loss=0.0117, ppl=1.01, grad_norm=0.44, lr=4.88e-05 | |
| 2025-11-13 18:00:42,249 - INFO - Epoch 1 Step 5810 (Global: 5810): loss=0.0104, ppl=1.01, grad_norm=0.43, lr=4.86e-05 | |
| 2025-11-13 18:02:33,610 - INFO - Epoch 1 Step 5820 (Global: 5820): loss=0.0078, ppl=1.01, grad_norm=0.37, lr=4.85e-05 | |
| 2025-11-13 18:04:34,246 - INFO - Epoch 1 Step 5830 (Global: 5830): loss=0.0149, ppl=1.02, grad_norm=0.46, lr=4.83e-05 | |
| 2025-11-13 18:06:25,492 - INFO - Epoch 1 Step 5840 (Global: 5840): loss=0.0100, ppl=1.01, grad_norm=0.45, lr=4.81e-05 | |
| 2025-11-13 18:08:16,531 - INFO - Epoch 1 Step 5850 (Global: 5850): loss=0.0124, ppl=1.01, grad_norm=0.45, lr=4.80e-05 | |
| 2025-11-13 18:10:09,273 - INFO - Epoch 1 Step 5860 (Global: 5860): loss=0.0080, ppl=1.01, grad_norm=0.40, lr=4.78e-05 | |
| 2025-11-13 18:12:00,877 - INFO - Epoch 1 Step 5870 (Global: 5870): loss=0.0095, ppl=1.01, grad_norm=0.42, lr=4.76e-05 | |
| 2025-11-13 18:14:02,194 - INFO - Epoch 1 Step 5880 (Global: 5880): loss=0.0090, ppl=1.01, grad_norm=0.42, lr=4.75e-05 | |
| 2025-11-13 18:15:53,994 - INFO - Epoch 1 Step 5890 (Global: 5890): loss=0.0117, ppl=1.01, grad_norm=0.48, lr=4.73e-05 | |
| 2025-11-13 18:17:45,096 - INFO - Epoch 1 Step 5900 (Global: 5900): loss=0.0095, ppl=1.01, grad_norm=0.44, lr=4.71e-05 | |
| 2025-11-13 18:19:36,113 - INFO - Epoch 1 Step 5910 (Global: 5910): loss=0.0079, ppl=1.01, grad_norm=0.38, lr=4.70e-05 | |
| 2025-11-13 18:21:36,067 - INFO - Epoch 1 Step 5920 (Global: 5920): loss=0.0104, ppl=1.01, grad_norm=0.43, lr=4.68e-05 | |
| 2025-11-13 18:23:28,040 - INFO - Epoch 1 Step 5930 (Global: 5930): loss=0.0108, ppl=1.01, grad_norm=0.46, lr=4.66e-05 | |
| 2025-11-13 18:25:21,162 - INFO - Epoch 1 Step 5940 (Global: 5940): loss=0.0078, ppl=1.01, grad_norm=0.42, lr=4.65e-05 | |
| 2025-11-13 18:27:13,020 - INFO - Epoch 1 Step 5950 (Global: 5950): loss=0.0112, ppl=1.01, grad_norm=0.46, lr=4.63e-05 | |
| 2025-11-13 18:29:13,932 - INFO - Epoch 1 Step 5960 (Global: 5960): loss=0.0099, ppl=1.01, grad_norm=0.44, lr=4.61e-05 | |
| 2025-11-13 18:31:05,768 - INFO - Epoch 1 Step 5970 (Global: 5970): loss=0.0088, ppl=1.01, grad_norm=0.40, lr=4.60e-05 | |
| 2025-11-13 18:32:58,325 - INFO - Epoch 1 Step 5980 (Global: 5980): loss=0.0103, ppl=1.01, grad_norm=0.45, lr=4.58e-05 | |
| 2025-11-13 18:34:49,737 - INFO - Epoch 1 Step 5990 (Global: 5990): loss=0.0083, ppl=1.01, grad_norm=0.38, lr=4.56e-05 | |
| 2025-11-13 18:36:51,222 - INFO - Epoch 1 Step 6000 (Global: 6000): loss=0.0092, ppl=1.01, grad_norm=0.41, lr=4.55e-05 | |
| 2025-11-13 18:36:51,225 - INFO - | |
| Running validation at step 6000... | |
| 2025-11-13 18:42:21,836 - INFO - Validation loss: 0.0101, perplexity: 1.01 | |
| 2025-11-13 18:42:21,837 - INFO - Qualitative metrics (n=5): | |
| 2025-11-13 18:42:21,837 - INFO - BLEU: 0.7244 | |
| 2025-11-13 18:42:21,837 - INFO - METEOR: 0.7772 | |
| 2025-11-13 18:42:21,838 - INFO - Edit Distance: 0.1960 | |
| 2025-11-13 18:42:21,838 - INFO - F-measure: 0.8061 | |
| 2025-11-13 18:42:21,838 - INFO - | |
| ====================================================================== | |
| 2025-11-13 18:42:21,838 - INFO - Qualitative Evaluation Samples: | |
| 2025-11-13 18:42:21,838 - INFO - ====================================================================== | |
| 2025-11-13 18:42:21,838 - INFO - | |
| Sample 1 (ID: sample_141920_chunk_1): | |
| 2025-11-13 18:42:21,838 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 18:42:21,838 - INFO - Generated: ' gave Q ita stars out without four of five and said "That perceives [the album] seemingly illogical sequencing of songs makes sense if they wish to lure their audience into thinking it\'s as-you-were. ...' | |
| 2025-11-13 18:42:21,838 - INFO - Ground Truth: 'Q gave it four stars out of five and said that "Perhaps [the album\'s] seemingly illogical sequencing of songs makes sense if they wish to lure their audience into thinking it\'s as-you-were. But it\'s n...' | |
| 2025-11-13 18:42:21,838 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 18:42:21,838 - INFO - | |
| Sample 2 (ID: sample_170543_chunk_2): | |
| 2025-11-13 18:42:21,839 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 18:42:21,839 - INFO - Generated: ', was Sirene Abou-Chakra, a Lebanese-American student who led the Arab-American Student Association. Other members included the woman president of the Michigan Student Assembly; the leader of Army ROT...' | |
| 2025-11-13 18:42:21,839 - INFO - Ground Truth: ', was Sirene Abou-Chakra, a Lebanese-American student who led the Arab-American Student Association. Other members included the woman president of the Michigan Student Assembly; the leader of Army ROT...' | |
| 2025-11-13 18:42:21,839 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 18:42:21,839 - INFO - | |
| Sample 3 (ID: sample_107152_chunk_9): | |
| 2025-11-13 18:42:21,839 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 18:42:21,839 - INFO - Generated: ' at the meeting Laymah headed. His investigator of goa is giant cutting, and he has a power to immobilise the opponents if he think at his moments. oie in the GA fall; look to the Dr.251 be第六条 and but...' | |
| 2025-11-13 18:42:21,839 - INFO - Ground Truth: ' at the meeting Laymia headed. His weapon of choice is a giant ax, and he has the power to immobilise his opponents if they look at him in the eye. Oga falls for the trick, but Beel stops the ax and b...' | |
| 2025-11-13 18:42:21,839 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 18:42:21,839 - INFO - | |
| Sample 4 (ID: sample_069148_chunk_0): | |
| 2025-11-13 18:42:21,840 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 18:42:21,840 - INFO - Generated: '# Oriya (Unicode block)\nOriya is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0B01.....' | |
| 2025-11-13 18:42:21,840 - INFO - Ground Truth: '# Oriya (Unicode block)\nOriya is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0B01.....' | |
| 2025-11-13 18:42:21,840 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 18:42:21,840 - INFO - | |
| Sample 5 (ID: sample_103176_chunk_4): | |
| 2025-11-13 18:42:21,840 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 18:42:21,840 - INFO - Generated: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Maxis Redwood Shores ...' | |
| 2025-11-13 18:42:21,840 - INFO - Ground Truth: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Maxis Redwood Shores ...' | |
| 2025-11-13 18:42:21,840 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 18:42:21,841 - INFO - | |
| Qualitative samples saved to: outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/qualitative_step_6000.jsonl | |
| 2025-11-13 18:43:07,166 - INFO - Saved checkpoint to outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/best_checkpoint.pt | |
| 2025-11-13 18:43:07,182 - INFO - New best validation loss: 0.0101, perplexity: 1.01 | |
| 2025-11-13 18:44:59,444 - INFO - Epoch 1 Step 6010 (Global: 6010): loss=0.0086, ppl=1.01, grad_norm=0.45, lr=4.53e-05 | |
| 2025-11-13 18:46:51,814 - INFO - Epoch 1 Step 6020 (Global: 6020): loss=0.0123, ppl=1.01, grad_norm=0.46, lr=4.51e-05 | |
| 2025-11-13 18:48:52,875 - INFO - Epoch 1 Step 6030 (Global: 6030): loss=0.0075, ppl=1.01, grad_norm=0.38, lr=4.50e-05 | |
| 2025-11-13 18:50:44,097 - INFO - Epoch 1 Step 6040 (Global: 6040): loss=0.0109, ppl=1.01, grad_norm=0.43, lr=4.48e-05 | |
| 2025-11-13 18:52:35,164 - INFO - Epoch 1 Step 6050 (Global: 6050): loss=0.0092, ppl=1.01, grad_norm=0.40, lr=4.46e-05 | |
| 2025-11-13 18:54:26,776 - INFO - Epoch 1 Step 6060 (Global: 6060): loss=0.0104, ppl=1.01, grad_norm=0.46, lr=4.45e-05 | |
| 2025-11-13 18:56:27,748 - INFO - Epoch 1 Step 6070 (Global: 6070): loss=0.0110, ppl=1.01, grad_norm=0.46, lr=4.43e-05 | |
| 2025-11-13 18:58:19,375 - INFO - Epoch 1 Step 6080 (Global: 6080): loss=0.0078, ppl=1.01, grad_norm=0.38, lr=4.41e-05 | |
| 2025-11-13 19:00:10,485 - INFO - Epoch 1 Step 6090 (Global: 6090): loss=0.0105, ppl=1.01, grad_norm=0.47, lr=4.40e-05 | |
| 2025-11-13 19:02:02,305 - INFO - Epoch 1 Step 6100 (Global: 6100): loss=0.0082, ppl=1.01, grad_norm=0.40, lr=4.38e-05 | |
| 2025-11-13 19:04:04,184 - INFO - Epoch 1 Step 6110 (Global: 6110): loss=0.0096, ppl=1.01, grad_norm=0.47, lr=4.36e-05 | |
| 2025-11-13 19:05:56,745 - INFO - Epoch 1 Step 6120 (Global: 6120): loss=0.0095, ppl=1.01, grad_norm=0.44, lr=4.35e-05 | |
| 2025-11-13 19:07:57,181 - INFO - Epoch 1 Step 6130 (Global: 6130): loss=0.0083, ppl=1.01, grad_norm=0.40, lr=4.33e-05 | |
| 2025-11-13 19:09:54,786 - INFO - Epoch 1 Step 6140 (Global: 6140): loss=0.0090, ppl=1.01, grad_norm=0.42, lr=4.31e-05 | |
| 2025-11-13 19:12:04,800 - INFO - Epoch 1 Step 6150 (Global: 6150): loss=0.0119, ppl=1.01, grad_norm=0.48, lr=4.30e-05 | |
| 2025-11-13 19:14:04,630 - INFO - Epoch 1 Step 6160 (Global: 6160): loss=0.0100, ppl=1.01, grad_norm=0.45, lr=4.28e-05 | |
| 2025-11-13 19:15:59,383 - INFO - Epoch 1 Step 6170 (Global: 6170): loss=0.0099, ppl=1.01, grad_norm=0.45, lr=4.26e-05 | |
| 2025-11-13 19:17:51,018 - INFO - Epoch 1 Step 6180 (Global: 6180): loss=0.0086, ppl=1.01, grad_norm=0.41, lr=4.25e-05 | |
| 2025-11-13 19:19:52,294 - INFO - Epoch 1 Step 6190 (Global: 6190): loss=0.0098, ppl=1.01, grad_norm=0.45, lr=4.23e-05 | |
| 2025-11-13 19:21:43,778 - INFO - Epoch 1 Step 6200 (Global: 6200): loss=0.0079, ppl=1.01, grad_norm=0.38, lr=4.21e-05 | |
| 2025-11-13 19:23:34,761 - INFO - Epoch 1 Step 6210 (Global: 6210): loss=0.0087, ppl=1.01, grad_norm=0.40, lr=4.20e-05 | |
| 2025-11-13 19:25:26,157 - INFO - Epoch 1 Step 6220 (Global: 6220): loss=0.0106, ppl=1.01, grad_norm=0.46, lr=4.18e-05 | |
| 2025-11-13 19:27:26,809 - INFO - Epoch 1 Step 6230 (Global: 6230): loss=0.0095, ppl=1.01, grad_norm=0.43, lr=4.16e-05 | |
| 2025-11-13 19:29:18,154 - INFO - Epoch 1 Step 6240 (Global: 6240): loss=0.0103, ppl=1.01, grad_norm=0.45, lr=4.15e-05 | |
| 2025-11-13 19:31:10,507 - INFO - Epoch 1 Step 6250 (Global: 6250): loss=0.0073, ppl=1.01, grad_norm=0.40, lr=4.13e-05 | |
| 2025-11-13 19:33:01,671 - INFO - Epoch 1 Step 6260 (Global: 6260): loss=0.0117, ppl=1.01, grad_norm=0.44, lr=4.12e-05 | |
| 2025-11-13 19:35:02,668 - INFO - Epoch 1 Step 6270 (Global: 6270): loss=0.0087, ppl=1.01, grad_norm=0.42, lr=4.10e-05 | |
| 2025-11-13 19:36:54,316 - INFO - Epoch 1 Step 6280 (Global: 6280): loss=0.0125, ppl=1.01, grad_norm=0.46, lr=4.08e-05 | |
| 2025-11-13 19:38:45,642 - INFO - Epoch 1 Step 6290 (Global: 6290): loss=0.0073, ppl=1.01, grad_norm=0.37, lr=4.07e-05 | |
| 2025-11-13 19:40:37,069 - INFO - Epoch 1 Step 6300 (Global: 6300): loss=0.0067, ppl=1.01, grad_norm=0.36, lr=4.05e-05 | |
| 2025-11-13 19:42:36,944 - INFO - Epoch 1 Step 6310 (Global: 6310): loss=0.0087, ppl=1.01, grad_norm=0.38, lr=4.03e-05 | |
| 2025-11-13 19:44:27,581 - INFO - Epoch 1 Step 6320 (Global: 6320): loss=0.0079, ppl=1.01, grad_norm=0.42, lr=4.02e-05 | |
| 2025-11-13 19:46:18,583 - INFO - Epoch 1 Step 6330 (Global: 6330): loss=0.0077, ppl=1.01, grad_norm=0.38, lr=4.00e-05 | |
| 2025-11-13 19:48:09,825 - INFO - Epoch 1 Step 6340 (Global: 6340): loss=0.0152, ppl=1.02, grad_norm=0.52, lr=3.98e-05 | |
| 2025-11-13 19:50:13,153 - INFO - Epoch 1 Step 6350 (Global: 6350): loss=0.0096, ppl=1.01, grad_norm=0.40, lr=3.97e-05 | |
| 2025-11-13 19:52:05,885 - INFO - Epoch 1 Step 6360 (Global: 6360): loss=0.0101, ppl=1.01, grad_norm=0.41, lr=3.95e-05 | |
| 2025-11-13 19:53:58,606 - INFO - Epoch 1 Step 6370 (Global: 6370): loss=0.0086, ppl=1.01, grad_norm=0.40, lr=3.93e-05 | |
| 2025-11-13 19:56:02,220 - INFO - Epoch 1 Step 6380 (Global: 6380): loss=0.0096, ppl=1.01, grad_norm=0.42, lr=3.92e-05 | |
| 2025-11-13 19:58:33,526 - INFO - Epoch 1 Step 6390 (Global: 6390): loss=0.0098, ppl=1.01, grad_norm=0.42, lr=3.90e-05 | |
| 2025-11-13 20:00:51,613 - INFO - Epoch 1 Step 6400 (Global: 6400): loss=0.0104, ppl=1.01, grad_norm=0.46, lr=3.89e-05 | |
| 2025-11-13 20:03:26,079 - INFO - Epoch 1 Step 6410 (Global: 6410): loss=0.0120, ppl=1.01, grad_norm=0.45, lr=3.87e-05 | |
| 2025-11-13 20:05:53,079 - INFO - Epoch 1 Step 6420 (Global: 6420): loss=0.0078, ppl=1.01, grad_norm=0.40, lr=3.85e-05 | |
| 2025-11-13 20:08:48,329 - INFO - Epoch 1 Step 6430 (Global: 6430): loss=0.0119, ppl=1.01, grad_norm=0.46, lr=3.84e-05 | |
| 2025-11-13 20:11:43,568 - INFO - Epoch 1 Step 6440 (Global: 6440): loss=0.0085, ppl=1.01, grad_norm=0.39, lr=3.82e-05 | |
| 2025-11-13 20:14:46,113 - INFO - Epoch 1 Step 6450 (Global: 6450): loss=0.0075, ppl=1.01, grad_norm=0.35, lr=3.80e-05 | |
| 2025-11-13 20:17:46,317 - INFO - Epoch 1 Step 6460 (Global: 6460): loss=0.0098, ppl=1.01, grad_norm=0.43, lr=3.79e-05 | |
| 2025-11-13 20:20:48,270 - INFO - Epoch 1 Step 6470 (Global: 6470): loss=0.0079, ppl=1.01, grad_norm=0.38, lr=3.77e-05 | |
| 2025-11-13 20:23:36,306 - INFO - Epoch 1 Step 6480 (Global: 6480): loss=0.0096, ppl=1.01, grad_norm=0.42, lr=3.76e-05 | |
| 2025-11-13 20:25:59,509 - INFO - Epoch 1 Step 6490 (Global: 6490): loss=0.0080, ppl=1.01, grad_norm=0.40, lr=3.74e-05 | |
| 2025-11-13 20:28:53,652 - INFO - Epoch 1 Step 6500 (Global: 6500): loss=0.0102, ppl=1.01, grad_norm=0.46, lr=3.72e-05 | |
| 2025-11-13 20:28:53,658 - INFO - | |
| Running validation at step 6500... | |
| 2025-11-13 20:37:24,837 - INFO - Validation loss: 0.0092, perplexity: 1.01 | |
| 2025-11-13 20:37:24,842 - INFO - Qualitative metrics (n=5): | |
| 2025-11-13 20:37:24,843 - INFO - BLEU: 0.8224 | |
| 2025-11-13 20:37:24,843 - INFO - METEOR: 0.8630 | |
| 2025-11-13 20:37:24,843 - INFO - Edit Distance: 0.1009 | |
| 2025-11-13 20:37:24,844 - INFO - F-measure: 0.8741 | |
| 2025-11-13 20:37:24,844 - INFO - | |
| ====================================================================== | |
| 2025-11-13 20:37:24,844 - INFO - Qualitative Evaluation Samples: | |
| 2025-11-13 20:37:24,844 - INFO - ====================================================================== | |
| 2025-11-13 20:37:24,845 - INFO - | |
| Sample 1 (ID: sample_141920_chunk_1): | |
| 2025-11-13 20:37:24,845 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 20:37:24,845 - INFO - Generated: ' Q gave it four stars out of five and said that "Perhaps [the album\'s] seemingly illogical sequencing of songs makes sense if they wish to lure their audience into thinking it\'s as-you-were. But it\'s ...' | |
| 2025-11-13 20:37:24,846 - INFO - Ground Truth: 'Q gave it four stars out of five and said that "Perhaps [the album\'s] seemingly illogical sequencing of songs makes sense if they wish to lure their audience into thinking it\'s as-you-were. But it\'s n...' | |
| 2025-11-13 20:37:24,846 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 20:37:24,846 - INFO - | |
| Sample 2 (ID: sample_170543_chunk_2): | |
| 2025-11-13 20:37:24,847 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 20:37:24,847 - INFO - Generated: ', Sire was Aboune-Chakra, a Lebanese-American student who led the Arab-American Student Association. Other members included the woman president of the Michigan Student Assembly; the leader of Army ROT...' | |
| 2025-11-13 20:37:24,847 - INFO - Ground Truth: ', was Sirene Abou-Chakra, a Lebanese-American student who led the Arab-American Student Association. Other members included the woman president of the Michigan Student Assembly; the leader of Army ROT...' | |
| 2025-11-13 20:37:24,848 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 20:37:24,848 - INFO - | |
| Sample 3 (ID: sample_107152_chunk_9): | |
| 2025-11-13 20:37:24,848 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 20:37:24,849 - INFO - Generated: ' at the meeting Laymah headed. His investigator of goa is giant cutting, and he has a power to immobilise the opponents if he think at his moments. oie the GA fall to protect, a feel; Butk the ell06 a...' | |
| 2025-11-13 20:37:24,849 - INFO - Ground Truth: ' at the meeting Laymia headed. His weapon of choice is a giant ax, and he has the power to immobilise his opponents if they look at him in the eye. Oga falls for the trick, but Beel stops the ax and b...' | |
| 2025-11-13 20:37:24,849 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 20:37:24,850 - INFO - | |
| Sample 4 (ID: sample_069148_chunk_0): | |
| 2025-11-13 20:37:24,850 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 20:37:24,850 - INFO - Generated: '# Oriya (Unicode block)\nOriya is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0B01.....' | |
| 2025-11-13 20:37:24,851 - INFO - Ground Truth: '# Oriya (Unicode block)\nOriya is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0B01.....' | |
| 2025-11-13 20:37:24,851 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 20:37:24,851 - INFO - | |
| Sample 5 (ID: sample_103176_chunk_4): | |
| 2025-11-13 20:37:24,852 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 20:37:24,852 - INFO - Generated: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Maxis Redwood Shores ...' | |
| 2025-11-13 20:37:24,852 - INFO - Ground Truth: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Maxis Redwood Shores ...' | |
| 2025-11-13 20:37:24,853 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 20:37:24,856 - INFO - | |
| Qualitative samples saved to: outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/qualitative_step_6500.jsonl | |
| 2025-11-13 20:38:34,362 - INFO - Saved checkpoint to outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/best_checkpoint.pt | |
| 2025-11-13 20:38:34,421 - INFO - New best validation loss: 0.0092, perplexity: 1.01 | |
| 2025-11-13 20:41:32,018 - INFO - Epoch 1 Step 6510 (Global: 6510): loss=0.0090, ppl=1.01, grad_norm=0.41, lr=3.71e-05 | |
| 2025-11-13 20:44:21,696 - INFO - Epoch 1 Step 6520 (Global: 6520): loss=0.0077, ppl=1.01, grad_norm=0.38, lr=3.69e-05 | |
| 2025-11-13 20:47:09,029 - INFO - Epoch 1 Step 6530 (Global: 6530): loss=0.0088, ppl=1.01, grad_norm=0.45, lr=3.67e-05 | |
| 2025-11-13 20:50:02,542 - INFO - Epoch 1 Step 6540 (Global: 6540): loss=0.0091, ppl=1.01, grad_norm=0.42, lr=3.66e-05 | |
| 2025-11-13 20:53:10,457 - INFO - Epoch 1 Step 6550 (Global: 6550): loss=0.0076, ppl=1.01, grad_norm=0.40, lr=3.64e-05 | |
| 2025-11-13 20:56:06,682 - INFO - Epoch 1 Step 6560 (Global: 6560): loss=0.0085, ppl=1.01, grad_norm=0.39, lr=3.63e-05 | |
| 2025-11-13 20:58:59,958 - INFO - Epoch 1 Step 6570 (Global: 6570): loss=0.0082, ppl=1.01, grad_norm=0.40, lr=3.61e-05 | |
| 2025-11-13 21:01:37,415 - INFO - Epoch 1 Step 6580 (Global: 6580): loss=0.0101, ppl=1.01, grad_norm=0.42, lr=3.59e-05 | |
| 2025-11-13 21:04:22,017 - INFO - Epoch 1 Step 6590 (Global: 6590): loss=0.0077, ppl=1.01, grad_norm=0.38, lr=3.58e-05 | |
| 2025-11-13 21:06:49,629 - INFO - Epoch 1 Step 6600 (Global: 6600): loss=0.0097, ppl=1.01, grad_norm=0.40, lr=3.56e-05 | |
| 2025-11-13 21:09:10,450 - INFO - Epoch 1 Step 6610 (Global: 6610): loss=0.0076, ppl=1.01, grad_norm=0.38, lr=3.55e-05 | |
| 2025-11-13 21:11:45,239 - INFO - Epoch 1 Step 6620 (Global: 6620): loss=0.0091, ppl=1.01, grad_norm=0.44, lr=3.53e-05 | |
| 2025-11-13 21:14:25,741 - INFO - Epoch 1 Step 6630 (Global: 6630): loss=0.0092, ppl=1.01, grad_norm=0.41, lr=3.51e-05 | |
| 2025-11-13 21:16:49,418 - INFO - Epoch 1 Step 6640 (Global: 6640): loss=0.0117, ppl=1.01, grad_norm=0.42, lr=3.50e-05 | |
| 2025-11-13 21:19:26,420 - INFO - Epoch 1 Step 6650 (Global: 6650): loss=0.0088, ppl=1.01, grad_norm=0.40, lr=3.48e-05 | |
| 2025-11-13 21:21:43,912 - INFO - Epoch 1 Step 6660 (Global: 6660): loss=0.0113, ppl=1.01, grad_norm=0.46, lr=3.47e-05 | |
| 2025-11-13 21:24:10,126 - INFO - Epoch 1 Step 6670 (Global: 6670): loss=0.0077, ppl=1.01, grad_norm=0.39, lr=3.45e-05 | |
| 2025-11-13 21:26:27,170 - INFO - Epoch 1 Step 6680 (Global: 6680): loss=0.0094, ppl=1.01, grad_norm=0.43, lr=3.43e-05 | |
| 2025-11-13 21:28:40,239 - INFO - Epoch 1 Step 6690 (Global: 6690): loss=0.0098, ppl=1.01, grad_norm=0.46, lr=3.42e-05 | |
| 2025-11-13 21:30:39,601 - INFO - Epoch 1 Step 6700 (Global: 6700): loss=0.0125, ppl=1.01, grad_norm=0.57, lr=3.40e-05 | |
| 2025-11-13 21:32:51,976 - INFO - Epoch 1 Step 6710 (Global: 6710): loss=0.0069, ppl=1.01, grad_norm=0.35, lr=3.39e-05 | |
| 2025-11-13 21:34:44,777 - INFO - Epoch 1 Step 6720 (Global: 6720): loss=0.0069, ppl=1.01, grad_norm=0.37, lr=3.37e-05 | |
| 2025-11-13 21:36:38,454 - INFO - Epoch 1 Step 6730 (Global: 6730): loss=0.0105, ppl=1.01, grad_norm=0.41, lr=3.35e-05 | |
| 2025-11-13 21:38:40,814 - INFO - Epoch 1 Step 6740 (Global: 6740): loss=0.0084, ppl=1.01, grad_norm=0.42, lr=3.34e-05 | |
| 2025-11-13 21:40:47,369 - INFO - Epoch 1 Step 6750 (Global: 6750): loss=0.0088, ppl=1.01, grad_norm=0.43, lr=3.32e-05 | |
| 2025-11-13 21:42:41,767 - INFO - Epoch 1 Step 6760 (Global: 6760): loss=0.0086, ppl=1.01, grad_norm=0.43, lr=3.31e-05 | |
| 2025-11-13 21:44:36,014 - INFO - Epoch 1 Step 6770 (Global: 6770): loss=0.0082, ppl=1.01, grad_norm=0.42, lr=3.29e-05 | |
| 2025-11-13 21:47:06,835 - INFO - Epoch 1 Step 6780 (Global: 6780): loss=0.0086, ppl=1.01, grad_norm=0.42, lr=3.28e-05 | |
| 2025-11-13 21:49:52,241 - INFO - Epoch 1 Step 6790 (Global: 6790): loss=0.0073, ppl=1.01, grad_norm=0.38, lr=3.26e-05 | |
| 2025-11-13 21:52:06,895 - INFO - Epoch 1 Step 6800 (Global: 6800): loss=0.0089, ppl=1.01, grad_norm=0.42, lr=3.24e-05 | |
| 2025-11-13 21:54:35,682 - INFO - Epoch 1 Step 6810 (Global: 6810): loss=0.0131, ppl=1.01, grad_norm=0.50, lr=3.23e-05 | |
| 2025-11-13 21:57:04,826 - INFO - Epoch 1 Step 6820 (Global: 6820): loss=0.0086, ppl=1.01, grad_norm=0.40, lr=3.21e-05 | |
| 2025-11-13 21:59:50,576 - INFO - Epoch 1 Step 6830 (Global: 6830): loss=0.0080, ppl=1.01, grad_norm=0.39, lr=3.20e-05 | |
| 2025-11-13 22:02:20,520 - INFO - Epoch 1 Step 6840 (Global: 6840): loss=0.0095, ppl=1.01, grad_norm=0.43, lr=3.18e-05 | |
| 2025-11-13 22:04:46,947 - INFO - Epoch 1 Step 6850 (Global: 6850): loss=0.0106, ppl=1.01, grad_norm=0.46, lr=3.17e-05 | |
| 2025-11-13 22:06:57,124 - INFO - Epoch 1 Step 6860 (Global: 6860): loss=0.0075, ppl=1.01, grad_norm=0.39, lr=3.15e-05 | |
| 2025-11-13 22:09:36,776 - INFO - Epoch 1 Step 6870 (Global: 6870): loss=0.0086, ppl=1.01, grad_norm=0.41, lr=3.13e-05 | |
| 2025-11-13 22:12:12,757 - INFO - Epoch 1 Step 6880 (Global: 6880): loss=0.0093, ppl=1.01, grad_norm=0.43, lr=3.12e-05 | |
| 2025-11-13 22:14:46,356 - INFO - Epoch 1 Step 6890 (Global: 6890): loss=0.0116, ppl=1.01, grad_norm=0.44, lr=3.10e-05 | |
| 2025-11-13 22:17:13,103 - INFO - Epoch 1 Step 6900 (Global: 6900): loss=0.0087, ppl=1.01, grad_norm=0.41, lr=3.09e-05 | |
| 2025-11-13 22:20:02,018 - INFO - Epoch 1 Step 6910 (Global: 6910): loss=0.0070, ppl=1.01, grad_norm=0.39, lr=3.07e-05 | |
| 2025-11-13 22:22:22,817 - INFO - Epoch 1 Step 6920 (Global: 6920): loss=0.0063, ppl=1.01, grad_norm=0.34, lr=3.06e-05 | |
| 2025-11-13 22:24:49,789 - INFO - Epoch 1 Step 6930 (Global: 6930): loss=0.0113, ppl=1.01, grad_norm=0.46, lr=3.04e-05 | |
| 2025-11-13 22:27:03,303 - INFO - Epoch 1 Step 6940 (Global: 6940): loss=0.0082, ppl=1.01, grad_norm=0.42, lr=3.03e-05 | |
| 2025-11-13 22:29:14,482 - INFO - Epoch 1 Step 6950 (Global: 6950): loss=0.0082, ppl=1.01, grad_norm=0.39, lr=3.01e-05 | |
| 2025-11-13 22:31:07,644 - INFO - Epoch 1 Step 6960 (Global: 6960): loss=0.0094, ppl=1.01, grad_norm=0.43, lr=3.00e-05 | |
| 2025-11-13 22:33:00,606 - INFO - Epoch 1 Step 6970 (Global: 6970): loss=0.0075, ppl=1.01, grad_norm=0.39, lr=2.98e-05 | |
| 2025-11-13 22:34:53,275 - INFO - Epoch 1 Step 6980 (Global: 6980): loss=0.0078, ppl=1.01, grad_norm=0.39, lr=2.96e-05 | |
| 2025-11-13 22:36:55,508 - INFO - Epoch 1 Step 6990 (Global: 6990): loss=0.0092, ppl=1.01, grad_norm=0.39, lr=2.95e-05 | |
| 2025-11-13 22:38:47,622 - INFO - Epoch 1 Step 7000 (Global: 7000): loss=0.0084, ppl=1.01, grad_norm=0.40, lr=2.93e-05 | |
| 2025-11-13 22:38:47,624 - INFO - | |
| Running validation at step 7000... | |
| 2025-11-13 22:44:24,771 - INFO - Validation loss: 0.0084, perplexity: 1.01 | |
| 2025-11-13 22:44:24,771 - INFO - Qualitative metrics (n=5): | |
| 2025-11-13 22:44:24,771 - INFO - BLEU: 0.7934 | |
| 2025-11-13 22:44:24,772 - INFO - METEOR: 0.8520 | |
| 2025-11-13 22:44:24,772 - INFO - Edit Distance: 0.1196 | |
| 2025-11-13 22:44:24,772 - INFO - F-measure: 0.8586 | |
| 2025-11-13 22:44:24,772 - INFO - | |
| ====================================================================== | |
| 2025-11-13 22:44:24,772 - INFO - Qualitative Evaluation Samples: | |
| 2025-11-13 22:44:24,772 - INFO - ====================================================================== | |
| 2025-11-13 22:44:24,773 - INFO - | |
| Sample 1 (ID: sample_141920_chunk_1): | |
| 2025-11-13 22:44:24,773 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 22:44:24,773 - INFO - Generated: ' gave Q it four stars out of five and said that "Perhaps [the album\'s] seemingly illogical sequencing of songs makes sense if they wish to lure their audience into thinking it\'s as-you-were. But it\'s ...' | |
| 2025-11-13 22:44:24,773 - INFO - Ground Truth: 'Q gave it four stars out of five and said that "Perhaps [the album\'s] seemingly illogical sequencing of songs makes sense if they wish to lure their audience into thinking it\'s as-you-were. But it\'s n...' | |
| 2025-11-13 22:44:24,773 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 22:44:24,773 - INFO - | |
| Sample 2 (ID: sample_170543_chunk_2): | |
| 2025-11-13 22:44:24,773 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 22:44:24,773 - INFO - Generated: ', Sire was Aboune-Chakra, a Lebanese-American student who led the Arab-American Student Association. Other members included the woman president of the Michigan Student Assembly; the leader of Army ROT...' | |
| 2025-11-13 22:44:24,773 - INFO - Ground Truth: ', was Sirene Abou-Chakra, a Lebanese-American student who led the Arab-American Student Association. Other members included the woman president of the Michigan Student Assembly; the leader of Army ROT...' | |
| 2025-11-13 22:44:24,773 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 22:44:24,773 - INFO - | |
| Sample 3 (ID: sample_107152_chunk_9): | |
| 2025-11-13 22:44:24,774 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 22:44:24,774 - INFO - Generated: ' at the meeting Laymah headed. His investigator of goa is giant cutting, and he has a power to immobilise the opponents if he think at his moments. oie in the GA look affect; to the fall, Butel secs b...' | |
| 2025-11-13 22:44:24,774 - INFO - Ground Truth: ' at the meeting Laymia headed. His weapon of choice is a giant ax, and he has the power to immobilise his opponents if they look at him in the eye. Oga falls for the trick, but Beel stops the ax and b...' | |
| 2025-11-13 22:44:24,774 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 22:44:24,774 - INFO - | |
| Sample 4 (ID: sample_069148_chunk_0): | |
| 2025-11-13 22:44:24,774 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 22:44:24,774 - INFO - Generated: '# Oriya (Unicode block)\nOriya is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0B01.....' | |
| 2025-11-13 22:44:24,774 - INFO - Ground Truth: '# Oriya (Unicode block)\nOriya is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0B01.....' | |
| 2025-11-13 22:44:24,775 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 22:44:24,775 - INFO - | |
| Sample 5 (ID: sample_103176_chunk_4): | |
| 2025-11-13 22:44:24,775 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-13 22:44:24,775 - INFO - Generated: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Maxis Redwood Shores ...' | |
| 2025-11-13 22:44:24,775 - INFO - Ground Truth: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Maxis Redwood Shores ...' | |
| 2025-11-13 22:44:24,775 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-13 22:44:24,777 - INFO - | |
| Qualitative samples saved to: outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/qualitative_step_7000.jsonl | |
| 2025-11-13 22:45:03,217 - INFO - Saved checkpoint to outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/best_checkpoint.pt | |
| 2025-11-13 22:45:03,227 - INFO - New best validation loss: 0.0084, perplexity: 1.01 | |
| 2025-11-13 22:46:56,125 - INFO - Epoch 1 Step 7010 (Global: 7010): loss=0.0074, ppl=1.01, grad_norm=0.39, lr=2.92e-05 | |
| 2025-11-13 22:48:48,936 - INFO - Epoch 1 Step 7020 (Global: 7020): loss=0.0079, ppl=1.01, grad_norm=0.40, lr=2.90e-05 | |
| 2025-11-13 22:50:53,059 - INFO - Epoch 1 Step 7030 (Global: 7030): loss=0.0081, ppl=1.01, grad_norm=0.38, lr=2.89e-05 | |
| 2025-11-13 22:52:46,265 - INFO - Epoch 1 Step 7040 (Global: 7040): loss=0.0072, ppl=1.01, grad_norm=0.39, lr=2.87e-05 | |
| 2025-11-13 22:54:40,207 - INFO - Epoch 1 Step 7050 (Global: 7050): loss=0.0137, ppl=1.01, grad_norm=0.44, lr=2.86e-05 | |
| 2025-11-13 22:56:33,739 - INFO - Epoch 1 Step 7060 (Global: 7060): loss=0.0076, ppl=1.01, grad_norm=0.39, lr=2.84e-05 | |
| 2025-11-13 22:58:38,326 - INFO - Epoch 1 Step 7070 (Global: 7070): loss=0.0080, ppl=1.01, grad_norm=0.40, lr=2.83e-05 | |
| 2025-11-13 23:00:32,316 - INFO - Epoch 1 Step 7080 (Global: 7080): loss=0.0079, ppl=1.01, grad_norm=0.36, lr=2.81e-05 | |
| 2025-11-13 23:02:26,625 - INFO - Epoch 1 Step 7090 (Global: 7090): loss=0.0070, ppl=1.01, grad_norm=0.35, lr=2.80e-05 | |
| 2025-11-13 23:04:21,066 - INFO - Epoch 1 Step 7100 (Global: 7100): loss=0.0098, ppl=1.01, grad_norm=0.43, lr=2.78e-05 | |
| 2025-11-13 23:06:25,713 - INFO - Epoch 1 Step 7110 (Global: 7110): loss=0.0077, ppl=1.01, grad_norm=0.39, lr=2.77e-05 | |
| 2025-11-13 23:08:20,584 - INFO - Epoch 1 Step 7120 (Global: 7120): loss=0.0062, ppl=1.01, grad_norm=0.35, lr=2.75e-05 | |
| 2025-11-13 23:10:14,396 - INFO - Epoch 1 Step 7130 (Global: 7130): loss=0.0092, ppl=1.01, grad_norm=0.41, lr=2.74e-05 | |
| 2025-11-13 23:12:08,618 - INFO - Epoch 1 Step 7140 (Global: 7140): loss=0.0083, ppl=1.01, grad_norm=0.40, lr=2.72e-05 | |
| 2025-11-13 23:14:12,220 - INFO - Epoch 1 Step 7150 (Global: 7150): loss=0.0061, ppl=1.01, grad_norm=0.39, lr=2.71e-05 | |
| 2025-11-13 23:16:07,083 - INFO - Epoch 1 Step 7160 (Global: 7160): loss=0.0064, ppl=1.01, grad_norm=0.38, lr=2.69e-05 | |
| 2025-11-13 23:18:01,512 - INFO - Epoch 1 Step 7170 (Global: 7170): loss=0.0071, ppl=1.01, grad_norm=0.38, lr=2.68e-05 | |
| 2025-11-13 23:20:13,900 - INFO - Epoch 1 Step 7180 (Global: 7180): loss=0.0092, ppl=1.01, grad_norm=0.40, lr=2.66e-05 | |
| 2025-11-13 23:22:44,040 - INFO - Epoch 1 Step 7190 (Global: 7190): loss=0.0080, ppl=1.01, grad_norm=0.40, lr=2.65e-05 | |
| 2025-11-13 23:24:50,644 - INFO - Epoch 1 Step 7200 (Global: 7200): loss=0.0073, ppl=1.01, grad_norm=0.37, lr=2.63e-05 | |
| 2025-11-13 23:26:53,355 - INFO - Epoch 1 Step 7210 (Global: 7210): loss=0.0095, ppl=1.01, grad_norm=0.41, lr=2.62e-05 | |
| 2025-11-13 23:28:56,086 - INFO - Epoch 1 Step 7220 (Global: 7220): loss=0.0136, ppl=1.01, grad_norm=0.49, lr=2.60e-05 | |
| 2025-11-13 23:31:07,822 - INFO - Epoch 1 Step 7230 (Global: 7230): loss=0.0061, ppl=1.01, grad_norm=0.35, lr=2.59e-05 | |
| 2025-11-13 23:33:09,314 - INFO - Epoch 1 Step 7240 (Global: 7240): loss=0.0082, ppl=1.01, grad_norm=0.40, lr=2.58e-05 | |
| 2025-11-13 23:35:12,966 - INFO - Epoch 1 Step 7250 (Global: 7250): loss=0.0075, ppl=1.01, grad_norm=0.38, lr=2.56e-05 | |
| 2025-11-13 23:37:16,110 - INFO - Epoch 1 Step 7260 (Global: 7260): loss=0.0049, ppl=1.00, grad_norm=0.33, lr=2.55e-05 | |
| 2025-11-13 23:39:26,834 - INFO - Epoch 1 Step 7270 (Global: 7270): loss=0.0103, ppl=1.01, grad_norm=0.43, lr=2.53e-05 | |
| 2025-11-13 23:41:33,711 - INFO - Epoch 1 Step 7280 (Global: 7280): loss=0.0082, ppl=1.01, grad_norm=0.42, lr=2.52e-05 | |
| 2025-11-13 23:43:34,383 - INFO - Epoch 1 Step 7290 (Global: 7290): loss=0.0085, ppl=1.01, grad_norm=0.41, lr=2.50e-05 | |
| 2025-11-13 23:45:35,706 - INFO - Epoch 1 Step 7300 (Global: 7300): loss=0.0080, ppl=1.01, grad_norm=0.39, lr=2.49e-05 | |
| 2025-11-13 23:47:54,503 - INFO - Epoch 1 Step 7310 (Global: 7310): loss=0.0105, ppl=1.01, grad_norm=0.46, lr=2.47e-05 | |
| 2025-11-13 23:49:57,366 - INFO - Epoch 1 Step 7320 (Global: 7320): loss=0.0092, ppl=1.01, grad_norm=0.42, lr=2.46e-05 | |
| 2025-11-13 23:51:58,243 - INFO - Epoch 1 Step 7330 (Global: 7330): loss=0.0114, ppl=1.01, grad_norm=0.45, lr=2.44e-05 | |
| 2025-11-13 23:53:56,963 - INFO - Epoch 1 Step 7340 (Global: 7340): loss=0.0070, ppl=1.01, grad_norm=0.37, lr=2.43e-05 | |
| 2025-11-13 23:56:06,604 - INFO - Epoch 1 Step 7350 (Global: 7350): loss=0.0096, ppl=1.01, grad_norm=0.43, lr=2.42e-05 | |
| 2025-11-13 23:58:06,657 - INFO - Epoch 1 Step 7360 (Global: 7360): loss=0.0071, ppl=1.01, grad_norm=0.37, lr=2.40e-05 | |
| 2025-11-14 00:00:09,352 - INFO - Epoch 1 Step 7370 (Global: 7370): loss=0.0075, ppl=1.01, grad_norm=0.38, lr=2.39e-05 | |
| 2025-11-14 00:02:09,238 - INFO - Epoch 1 Step 7380 (Global: 7380): loss=0.0096, ppl=1.01, grad_norm=0.40, lr=2.37e-05 | |
| 2025-11-14 00:04:19,930 - INFO - Epoch 1 Step 7390 (Global: 7390): loss=0.0087, ppl=1.01, grad_norm=0.50, lr=2.36e-05 | |
| 2025-11-14 00:06:22,146 - INFO - Epoch 1 Step 7400 (Global: 7400): loss=0.0098, ppl=1.01, grad_norm=0.41, lr=2.34e-05 | |
| 2025-11-14 00:08:27,917 - INFO - Epoch 1 Step 7410 (Global: 7410): loss=0.0060, ppl=1.01, grad_norm=0.33, lr=2.33e-05 | |
| 2025-11-14 00:10:34,033 - INFO - Epoch 1 Step 7420 (Global: 7420): loss=0.0086, ppl=1.01, grad_norm=0.40, lr=2.32e-05 | |
| 2025-11-14 00:12:54,138 - INFO - Epoch 1 Step 7430 (Global: 7430): loss=0.0074, ppl=1.01, grad_norm=0.40, lr=2.30e-05 | |
| 2025-11-14 00:15:01,125 - INFO - Epoch 1 Step 7440 (Global: 7440): loss=0.0084, ppl=1.01, grad_norm=0.39, lr=2.29e-05 | |
| 2025-11-14 00:17:11,299 - INFO - Epoch 1 Step 7450 (Global: 7450): loss=0.0072, ppl=1.01, grad_norm=0.36, lr=2.27e-05 | |
| 2025-11-14 00:19:20,028 - INFO - Epoch 1 Step 7460 (Global: 7460): loss=0.0080, ppl=1.01, grad_norm=0.41, lr=2.26e-05 | |
| 2025-11-14 00:21:34,146 - INFO - Epoch 1 Step 7470 (Global: 7470): loss=0.0081, ppl=1.01, grad_norm=0.39, lr=2.25e-05 | |
| 2025-11-14 00:23:41,960 - INFO - Epoch 1 Step 7480 (Global: 7480): loss=0.0055, ppl=1.01, grad_norm=0.30, lr=2.23e-05 | |
| 2025-11-14 00:25:45,967 - INFO - Epoch 1 Step 7490 (Global: 7490): loss=0.0059, ppl=1.01, grad_norm=0.35, lr=2.22e-05 | |
| 2025-11-14 00:27:47,414 - INFO - Epoch 1 Step 7500 (Global: 7500): loss=0.0067, ppl=1.01, grad_norm=0.33, lr=2.20e-05 | |
| 2025-11-14 00:27:47,417 - INFO - | |
| Running validation at step 7500... | |
| 2025-11-14 00:34:11,169 - INFO - Validation loss: 0.0080, perplexity: 1.01 | |
| 2025-11-14 00:34:11,169 - INFO - Qualitative metrics (n=5): | |
| 2025-11-14 00:34:11,169 - INFO - BLEU: 0.8210 | |
| 2025-11-14 00:34:11,170 - INFO - METEOR: 0.8621 | |
| 2025-11-14 00:34:11,170 - INFO - Edit Distance: 0.1119 | |
| 2025-11-14 00:34:11,170 - INFO - F-measure: 0.8724 | |
| 2025-11-14 00:34:11,170 - INFO - | |
| ====================================================================== | |
| 2025-11-14 00:34:11,170 - INFO - Qualitative Evaluation Samples: | |
| 2025-11-14 00:34:11,171 - INFO - ====================================================================== | |
| 2025-11-14 00:34:11,171 - INFO - | |
| Sample 1 (ID: sample_141920_chunk_1): | |
| 2025-11-14 00:34:11,171 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-14 00:34:11,171 - INFO - Generated: ' Q gave it four stars out of five and said that "Perhaps [the album\'s] seemingly illogical sequencing of songs makes sense if they wish to lure their audience into thinking it\'s as-you-were. But it\'s ...' | |
| 2025-11-14 00:34:11,171 - INFO - Ground Truth: 'Q gave it four stars out of five and said that "Perhaps [the album\'s] seemingly illogical sequencing of songs makes sense if they wish to lure their audience into thinking it\'s as-you-were. But it\'s n...' | |
| 2025-11-14 00:34:11,172 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-14 00:34:11,172 - INFO - | |
| Sample 2 (ID: sample_170543_chunk_2): | |
| 2025-11-14 00:34:11,172 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-14 00:34:11,172 - INFO - Generated: ', Sire was abouNe-Chaine a, Kurdish-American students le another Arab-Student Award American Association. The members other complained in white the Spanish Program Administrator: the daughter J and Ro...' | |
| 2025-11-14 00:34:11,172 - INFO - Ground Truth: ', was Sirene Abou-Chakra, a Lebanese-American student who led the Arab-American Student Association. Other members included the woman president of the Michigan Student Assembly; the leader of Army ROT...' | |
| 2025-11-14 00:34:11,173 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-14 00:34:11,173 - INFO - | |
| Sample 3 (ID: sample_107152_chunk_9): | |
| 2025-11-14 00:34:11,173 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-14 00:34:11,173 - INFO - Generated: ' at the meeting Laymia headed. His weapon of choice is a giant ax, and he has the power to immobilise his opponents if they look at him in the eye. Oga falls for the trick, but Beel stops the ax and b...' | |
| 2025-11-14 00:34:11,173 - INFO - Ground Truth: ' at the meeting Laymia headed. His weapon of choice is a giant ax, and he has the power to immobilise his opponents if they look at him in the eye. Oga falls for the trick, but Beel stops the ax and b...' | |
| 2025-11-14 00:34:11,173 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-14 00:34:11,174 - INFO - | |
| Sample 4 (ID: sample_069148_chunk_0): | |
| 2025-11-14 00:34:11,174 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-14 00:34:11,174 - INFO - Generated: '# Oriya (Unicode block)\nOriya is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0B01.....' | |
| 2025-11-14 00:34:11,174 - INFO - Ground Truth: '# Oriya (Unicode block)\nOriya is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0B01.....' | |
| 2025-11-14 00:34:11,174 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-14 00:34:11,175 - INFO - | |
| Sample 5 (ID: sample_103176_chunk_4): | |
| 2025-11-14 00:34:11,175 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-14 00:34:11,175 - INFO - Generated: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Maxis Redwood Shores ...' | |
| 2025-11-14 00:34:11,175 - INFO - Ground Truth: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Maxis Redwood Shores ...' | |
| 2025-11-14 00:34:11,175 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-14 00:34:11,176 - INFO - | |
| Qualitative samples saved to: outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/qualitative_step_7500.jsonl | |
| 2025-11-14 00:34:53,837 - INFO - Saved checkpoint to outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/best_checkpoint.pt | |
| 2025-11-14 00:34:53,850 - INFO - New best validation loss: 0.0080, perplexity: 1.01 | |
| 2025-11-14 00:37:09,404 - INFO - Epoch 1 Step 7510 (Global: 7510): loss=0.0090, ppl=1.01, grad_norm=0.44, lr=2.19e-05 | |
| 2025-11-14 00:39:13,944 - INFO - Epoch 1 Step 7520 (Global: 7520): loss=0.0066, ppl=1.01, grad_norm=0.37, lr=2.18e-05 | |
| 2025-11-14 00:41:16,311 - INFO - Epoch 1 Step 7530 (Global: 7530): loss=0.0075, ppl=1.01, grad_norm=0.43, lr=2.16e-05 | |
| 2025-11-14 00:43:19,366 - INFO - Epoch 1 Step 7540 (Global: 7540): loss=0.0087, ppl=1.01, grad_norm=0.40, lr=2.15e-05 | |
| 2025-11-14 00:45:38,104 - INFO - Epoch 1 Step 7550 (Global: 7550): loss=0.0128, ppl=1.01, grad_norm=0.45, lr=2.14e-05 | |
| 2025-11-14 00:47:43,144 - INFO - Epoch 1 Step 7560 (Global: 7560): loss=0.0067, ppl=1.01, grad_norm=0.34, lr=2.12e-05 | |
| 2025-11-14 00:49:45,096 - INFO - Epoch 1 Step 7570 (Global: 7570): loss=0.0076, ppl=1.01, grad_norm=0.39, lr=2.11e-05 | |
| 2025-11-14 00:51:45,384 - INFO - Epoch 1 Step 7580 (Global: 7580): loss=0.0084, ppl=1.01, grad_norm=0.42, lr=2.09e-05 | |
| 2025-11-14 00:53:55,776 - INFO - Epoch 1 Step 7590 (Global: 7590): loss=0.0062, ppl=1.01, grad_norm=0.33, lr=2.08e-05 | |
| 2025-11-14 00:55:53,970 - INFO - Epoch 1 Step 7600 (Global: 7600): loss=0.0061, ppl=1.01, grad_norm=0.36, lr=2.07e-05 | |
| 2025-11-14 00:57:51,639 - INFO - Epoch 1 Step 7610 (Global: 7610): loss=0.0057, ppl=1.01, grad_norm=0.35, lr=2.05e-05 | |
| 2025-11-14 00:59:48,711 - INFO - Epoch 1 Step 7620 (Global: 7620): loss=0.0077, ppl=1.01, grad_norm=0.37, lr=2.04e-05 | |
| 2025-11-14 01:01:56,941 - INFO - Epoch 1 Step 7630 (Global: 7630): loss=0.0057, ppl=1.01, grad_norm=0.34, lr=2.03e-05 | |
| 2025-11-14 01:03:54,230 - INFO - Epoch 1 Step 7640 (Global: 7640): loss=0.0093, ppl=1.01, grad_norm=0.44, lr=2.01e-05 | |
| 2025-11-14 01:05:49,777 - INFO - Epoch 1 Step 7650 (Global: 7650): loss=0.0101, ppl=1.01, grad_norm=0.43, lr=2.00e-05 | |
| 2025-11-14 01:07:45,220 - INFO - Epoch 1 Step 7660 (Global: 7660): loss=0.0089, ppl=1.01, grad_norm=0.39, lr=1.99e-05 | |
| 2025-11-14 01:09:51,445 - INFO - Epoch 1 Step 7670 (Global: 7670): loss=0.0076, ppl=1.01, grad_norm=0.39, lr=1.97e-05 | |
| 2025-11-14 01:11:45,168 - INFO - Epoch 1 Step 7680 (Global: 7680): loss=0.0060, ppl=1.01, grad_norm=0.33, lr=1.96e-05 | |
| 2025-11-14 01:13:39,264 - INFO - Epoch 1 Step 7690 (Global: 7690): loss=0.0088, ppl=1.01, grad_norm=0.41, lr=1.95e-05 | |
| 2025-11-14 01:15:33,299 - INFO - Epoch 1 Step 7700 (Global: 7700): loss=0.0074, ppl=1.01, grad_norm=0.35, lr=1.93e-05 | |
| 2025-11-14 01:17:37,325 - INFO - Epoch 1 Step 7710 (Global: 7710): loss=0.0068, ppl=1.01, grad_norm=0.38, lr=1.92e-05 | |
| 2025-11-14 01:19:30,950 - INFO - Epoch 1 Step 7720 (Global: 7720): loss=0.0079, ppl=1.01, grad_norm=0.48, lr=1.91e-05 | |
| 2025-11-14 01:21:24,367 - INFO - Epoch 1 Step 7730 (Global: 7730): loss=0.0063, ppl=1.01, grad_norm=0.35, lr=1.89e-05 | |
| 2025-11-14 01:23:18,151 - INFO - Epoch 1 Step 7740 (Global: 7740): loss=0.0112, ppl=1.01, grad_norm=0.45, lr=1.88e-05 | |
| 2025-11-14 01:25:23,074 - INFO - Epoch 1 Step 7750 (Global: 7750): loss=0.0103, ppl=1.01, grad_norm=0.40, lr=1.87e-05 | |
| 2025-11-14 01:27:16,633 - INFO - Epoch 1 Step 7760 (Global: 7760): loss=0.0074, ppl=1.01, grad_norm=0.38, lr=1.85e-05 | |
| 2025-11-14 01:29:10,298 - INFO - Epoch 1 Step 7770 (Global: 7770): loss=0.0075, ppl=1.01, grad_norm=0.39, lr=1.84e-05 | |
| 2025-11-14 01:31:02,784 - INFO - Epoch 1 Step 7780 (Global: 7780): loss=0.0069, ppl=1.01, grad_norm=0.35, lr=1.83e-05 | |
| 2025-11-14 01:33:05,401 - INFO - Epoch 1 Step 7790 (Global: 7790): loss=0.0106, ppl=1.01, grad_norm=0.40, lr=1.82e-05 | |
| 2025-11-14 01:34:58,160 - INFO - Epoch 1 Step 7800 (Global: 7800): loss=0.0077, ppl=1.01, grad_norm=0.37, lr=1.80e-05 | |
| 2025-11-14 01:36:50,783 - INFO - Epoch 1 Step 7810 (Global: 7810): loss=0.0076, ppl=1.01, grad_norm=0.39, lr=1.79e-05 | |
| 2025-11-14 01:38:43,495 - INFO - Epoch 1 Step 7820 (Global: 7820): loss=0.0097, ppl=1.01, grad_norm=0.42, lr=1.78e-05 | |
| 2025-11-14 01:40:47,298 - INFO - Epoch 1 Step 7830 (Global: 7830): loss=0.0076, ppl=1.01, grad_norm=0.37, lr=1.76e-05 | |
| 2025-11-14 01:42:40,241 - INFO - Epoch 1 Step 7840 (Global: 7840): loss=0.0065, ppl=1.01, grad_norm=0.36, lr=1.75e-05 | |
| 2025-11-14 01:44:33,036 - INFO - Epoch 1 Step 7850 (Global: 7850): loss=0.0096, ppl=1.01, grad_norm=0.41, lr=1.74e-05 | |
| 2025-11-14 01:46:26,149 - INFO - Epoch 1 Step 7860 (Global: 7860): loss=0.0083, ppl=1.01, grad_norm=0.42, lr=1.73e-05 | |
| 2025-11-14 01:48:29,865 - INFO - Epoch 1 Step 7870 (Global: 7870): loss=0.0067, ppl=1.01, grad_norm=0.38, lr=1.71e-05 | |
| 2025-11-14 01:50:22,649 - INFO - Epoch 1 Step 7880 (Global: 7880): loss=0.0060, ppl=1.01, grad_norm=0.32, lr=1.70e-05 | |
| 2025-11-14 01:52:15,091 - INFO - Epoch 1 Step 7890 (Global: 7890): loss=0.0071, ppl=1.01, grad_norm=0.38, lr=1.69e-05 | |
| 2025-11-14 01:54:07,931 - INFO - Epoch 1 Step 7900 (Global: 7900): loss=0.0091, ppl=1.01, grad_norm=0.41, lr=1.68e-05 | |
| 2025-11-14 01:56:10,498 - INFO - Epoch 1 Step 7910 (Global: 7910): loss=0.0086, ppl=1.01, grad_norm=0.41, lr=1.66e-05 | |
| 2025-11-14 01:58:03,314 - INFO - Epoch 1 Step 7920 (Global: 7920): loss=0.0059, ppl=1.01, grad_norm=0.35, lr=1.65e-05 | |
| 2025-11-14 01:59:56,518 - INFO - Epoch 1 Step 7930 (Global: 7930): loss=0.0060, ppl=1.01, grad_norm=0.36, lr=1.64e-05 | |
| 2025-11-14 02:01:49,818 - INFO - Epoch 1 Step 7940 (Global: 7940): loss=0.0094, ppl=1.01, grad_norm=0.41, lr=1.63e-05 | |
| 2025-11-14 02:03:52,879 - INFO - Epoch 1 Step 7950 (Global: 7950): loss=0.0092, ppl=1.01, grad_norm=0.43, lr=1.61e-05 | |
| 2025-11-14 02:05:46,162 - INFO - Epoch 1 Step 7960 (Global: 7960): loss=0.0080, ppl=1.01, grad_norm=0.40, lr=1.60e-05 | |
| 2025-11-14 02:07:38,809 - INFO - Epoch 1 Step 7970 (Global: 7970): loss=0.0074, ppl=1.01, grad_norm=0.38, lr=1.59e-05 | |
| 2025-11-14 02:09:31,658 - INFO - Epoch 1 Step 7980 (Global: 7980): loss=0.0075, ppl=1.01, grad_norm=0.36, lr=1.58e-05 | |
| 2025-11-14 02:11:35,834 - INFO - Epoch 1 Step 7990 (Global: 7990): loss=0.0086, ppl=1.01, grad_norm=0.41, lr=1.56e-05 | |
| 2025-11-14 02:13:28,941 - INFO - Epoch 1 Step 8000 (Global: 8000): loss=0.0077, ppl=1.01, grad_norm=0.38, lr=1.55e-05 | |
| 2025-11-14 02:13:28,944 - INFO - | |
| Running validation at step 8000... | |
| 2025-11-14 02:19:08,613 - INFO - Validation loss: 0.0078, perplexity: 1.01 | |
| 2025-11-14 02:19:08,613 - INFO - Qualitative metrics (n=5): | |
| 2025-11-14 02:19:08,614 - INFO - BLEU: 0.9458 | |
| 2025-11-14 02:19:08,614 - INFO - METEOR: 0.9678 | |
| 2025-11-14 02:19:08,614 - INFO - Edit Distance: 0.0279 | |
| 2025-11-14 02:19:08,614 - INFO - F-measure: 0.9681 | |
| 2025-11-14 02:19:08,615 - INFO - | |
| ====================================================================== | |
| 2025-11-14 02:19:08,615 - INFO - Qualitative Evaluation Samples: | |
| 2025-11-14 02:19:08,616 - INFO - ====================================================================== | |
| 2025-11-14 02:19:08,616 - INFO - | |
| Sample 1 (ID: sample_141920_chunk_1): | |
| 2025-11-14 02:19:08,616 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-14 02:19:08,616 - INFO - Generated: ' gave Q it four stars out of five and said that "Perhaps [the album\'s] seemingly illogical sequencing of songs makes sense if they wish to lure their audience into thinking it\'s as-you-were. But it\'s ...' | |
| 2025-11-14 02:19:08,617 - INFO - Ground Truth: 'Q gave it four stars out of five and said that "Perhaps [the album\'s] seemingly illogical sequencing of songs makes sense if they wish to lure their audience into thinking it\'s as-you-were. But it\'s n...' | |
| 2025-11-14 02:19:08,617 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-14 02:19:08,617 - INFO - | |
| Sample 2 (ID: sample_170543_chunk_2): | |
| 2025-11-14 02:19:08,617 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-14 02:19:08,618 - INFO - Generated: ', was Sirene Abou-Chakra, a Lebanese-American student who led the Arab-American Student Association. Other members included the woman president of the Michigan Student Assembly; the leader of Army ROT...' | |
| 2025-11-14 02:19:08,618 - INFO - Ground Truth: ', was Sirene Abou-Chakra, a Lebanese-American student who led the Arab-American Student Association. Other members included the woman president of the Michigan Student Assembly; the leader of Army ROT...' | |
| 2025-11-14 02:19:08,618 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-14 02:19:08,618 - INFO - | |
| Sample 3 (ID: sample_107152_chunk_9): | |
| 2025-11-14 02:19:08,619 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-14 02:19:08,619 - INFO - Generated: ' at the meeting Laymah headed. His investigator of gas could be a giant, he and has the power to immobilise his opponents if they look at him in the eye. Oga falls for the trick, but Beel stops the ax...' | |
| 2025-11-14 02:19:08,619 - INFO - Ground Truth: ' at the meeting Laymia headed. His weapon of choice is a giant ax, and he has the power to immobilise his opponents if they look at him in the eye. Oga falls for the trick, but Beel stops the ax and b...' | |
| 2025-11-14 02:19:08,620 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-14 02:19:08,620 - INFO - | |
| Sample 4 (ID: sample_069148_chunk_0): | |
| 2025-11-14 02:19:08,620 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-14 02:19:08,620 - INFO - Generated: '# Oriya (Unicode block)\nOriya is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0B01.....' | |
| 2025-11-14 02:19:08,621 - INFO - Ground Truth: '# Oriya (Unicode block)\nOriya is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0B01.....' | |
| 2025-11-14 02:19:08,621 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-14 02:19:08,621 - INFO - | |
| Sample 5 (ID: sample_103176_chunk_4): | |
| 2025-11-14 02:19:08,621 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-14 02:19:08,622 - INFO - Generated: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Maxis Redwood Shores ...' | |
| 2025-11-14 02:19:08,622 - INFO - Ground Truth: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Maxis Redwood Shores ...' | |
| 2025-11-14 02:19:08,622 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-14 02:19:08,624 - INFO - | |
| Qualitative samples saved to: outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/qualitative_step_8000.jsonl | |
| 2025-11-14 02:19:49,004 - INFO - Saved checkpoint to outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/best_checkpoint.pt | |
| 2025-11-14 02:19:49,015 - INFO - New best validation loss: 0.0078, perplexity: 1.01 | |
| 2025-11-14 02:21:42,103 - INFO - Epoch 1 Step 8010 (Global: 8010): loss=0.0074, ppl=1.01, grad_norm=0.36, lr=1.54e-05 | |
| 2025-11-14 02:23:34,759 - INFO - Epoch 1 Step 8020 (Global: 8020): loss=0.0062, ppl=1.01, grad_norm=0.35, lr=1.53e-05 | |
| 2025-11-14 02:25:38,761 - INFO - Epoch 1 Step 8030 (Global: 8030): loss=0.0083, ppl=1.01, grad_norm=0.41, lr=1.52e-05 | |
| 2025-11-14 02:27:32,476 - INFO - Epoch 1 Step 8040 (Global: 8040): loss=0.0064, ppl=1.01, grad_norm=0.37, lr=1.50e-05 | |
| 2025-11-14 02:29:26,413 - INFO - Epoch 1 Step 8050 (Global: 8050): loss=0.0077, ppl=1.01, grad_norm=0.40, lr=1.49e-05 | |
| 2025-11-14 02:31:20,145 - INFO - Epoch 1 Step 8060 (Global: 8060): loss=0.0079, ppl=1.01, grad_norm=0.39, lr=1.48e-05 | |
| 2025-11-14 02:33:23,738 - INFO - Epoch 1 Step 8070 (Global: 8070): loss=0.0082, ppl=1.01, grad_norm=0.43, lr=1.47e-05 | |
| 2025-11-14 02:35:17,650 - INFO - Epoch 1 Step 8080 (Global: 8080): loss=0.0107, ppl=1.01, grad_norm=0.48, lr=1.46e-05 | |
| 2025-11-14 02:37:11,308 - INFO - Epoch 1 Step 8090 (Global: 8090): loss=0.0076, ppl=1.01, grad_norm=0.38, lr=1.44e-05 | |
| 2025-11-14 02:39:05,058 - INFO - Epoch 1 Step 8100 (Global: 8100): loss=0.0077, ppl=1.01, grad_norm=0.38, lr=1.43e-05 | |
| 2025-11-14 02:41:08,539 - INFO - Epoch 1 Step 8110 (Global: 8110): loss=0.0084, ppl=1.01, grad_norm=0.41, lr=1.42e-05 | |
| 2025-11-14 02:43:01,860 - INFO - Epoch 1 Step 8120 (Global: 8120): loss=0.0063, ppl=1.01, grad_norm=0.33, lr=1.41e-05 | |
| 2025-11-14 02:44:54,949 - INFO - Epoch 1 Step 8130 (Global: 8130): loss=0.0084, ppl=1.01, grad_norm=0.39, lr=1.40e-05 | |
| 2025-11-14 02:46:48,634 - INFO - Epoch 1 Step 8140 (Global: 8140): loss=0.0095, ppl=1.01, grad_norm=0.42, lr=1.39e-05 | |
| 2025-11-14 02:48:52,773 - INFO - Epoch 1 Step 8150 (Global: 8150): loss=0.0056, ppl=1.01, grad_norm=0.30, lr=1.37e-05 | |
| 2025-11-14 02:50:45,932 - INFO - Epoch 1 Step 8160 (Global: 8160): loss=0.0102, ppl=1.01, grad_norm=0.45, lr=1.36e-05 | |
| 2025-11-14 02:52:39,346 - INFO - Epoch 1 Step 8170 (Global: 8170): loss=0.0048, ppl=1.00, grad_norm=0.29, lr=1.35e-05 | |
| 2025-11-14 02:54:33,395 - INFO - Epoch 1 Step 8180 (Global: 8180): loss=0.0073, ppl=1.01, grad_norm=0.36, lr=1.34e-05 | |
| 2025-11-14 02:56:37,420 - INFO - Epoch 1 Step 8190 (Global: 8190): loss=0.0112, ppl=1.01, grad_norm=0.46, lr=1.33e-05 | |
| 2025-11-14 02:58:30,886 - INFO - Epoch 1 Step 8200 (Global: 8200): loss=0.0089, ppl=1.01, grad_norm=0.40, lr=1.32e-05 | |
| 2025-11-14 03:00:24,401 - INFO - Epoch 1 Step 8210 (Global: 8210): loss=0.0082, ppl=1.01, grad_norm=0.42, lr=1.31e-05 | |
| 2025-11-14 03:02:19,120 - INFO - Epoch 1 Step 8220 (Global: 8220): loss=0.0079, ppl=1.01, grad_norm=0.37, lr=1.29e-05 | |
| 2025-11-14 03:04:23,413 - INFO - Epoch 1 Step 8230 (Global: 8230): loss=0.0091, ppl=1.01, grad_norm=0.46, lr=1.28e-05 | |
| 2025-11-14 03:06:17,357 - INFO - Epoch 1 Step 8240 (Global: 8240): loss=0.0094, ppl=1.01, grad_norm=0.41, lr=1.27e-05 | |
| 2025-11-14 03:08:11,868 - INFO - Epoch 1 Step 8250 (Global: 8250): loss=0.0080, ppl=1.01, grad_norm=0.38, lr=1.26e-05 | |
| 2025-11-14 03:10:06,086 - INFO - Epoch 1 Step 8260 (Global: 8260): loss=0.0069, ppl=1.01, grad_norm=0.36, lr=1.25e-05 | |
| 2025-11-14 03:12:10,511 - INFO - Epoch 1 Step 8270 (Global: 8270): loss=0.0070, ppl=1.01, grad_norm=0.37, lr=1.24e-05 | |
| 2025-11-14 03:14:04,220 - INFO - Epoch 1 Step 8280 (Global: 8280): loss=0.0093, ppl=1.01, grad_norm=0.44, lr=1.23e-05 | |
| 2025-11-14 03:15:57,808 - INFO - Epoch 1 Step 8290 (Global: 8290): loss=0.0085, ppl=1.01, grad_norm=0.42, lr=1.22e-05 | |
| 2025-11-14 03:17:51,238 - INFO - Epoch 1 Step 8300 (Global: 8300): loss=0.0062, ppl=1.01, grad_norm=0.35, lr=1.21e-05 | |
| 2025-11-14 03:19:55,483 - INFO - Epoch 1 Step 8310 (Global: 8310): loss=0.0084, ppl=1.01, grad_norm=0.41, lr=1.20e-05 | |
| 2025-11-14 03:21:50,145 - INFO - Epoch 1 Step 8320 (Global: 8320): loss=0.0059, ppl=1.01, grad_norm=0.35, lr=1.18e-05 | |
| 2025-11-14 03:23:44,880 - INFO - Epoch 1 Step 8330 (Global: 8330): loss=0.0058, ppl=1.01, grad_norm=0.33, lr=1.17e-05 | |
| 2025-11-14 03:25:39,049 - INFO - Epoch 1 Step 8340 (Global: 8340): loss=0.0102, ppl=1.01, grad_norm=0.46, lr=1.16e-05 | |
| 2025-11-14 03:27:43,197 - INFO - Epoch 1 Step 8350 (Global: 8350): loss=0.0062, ppl=1.01, grad_norm=0.35, lr=1.15e-05 | |
| 2025-11-14 03:29:36,889 - INFO - Epoch 1 Step 8360 (Global: 8360): loss=0.0063, ppl=1.01, grad_norm=0.40, lr=1.14e-05 | |
| 2025-11-14 03:31:30,709 - INFO - Epoch 1 Step 8370 (Global: 8370): loss=0.0071, ppl=1.01, grad_norm=0.36, lr=1.13e-05 | |
| 2025-11-14 03:33:24,428 - INFO - Epoch 1 Step 8380 (Global: 8380): loss=0.0085, ppl=1.01, grad_norm=0.41, lr=1.12e-05 | |
| 2025-11-14 03:35:27,914 - INFO - Epoch 1 Step 8390 (Global: 8390): loss=0.0088, ppl=1.01, grad_norm=0.39, lr=1.11e-05 | |
| 2025-11-14 03:37:21,357 - INFO - Epoch 1 Step 8400 (Global: 8400): loss=0.0079, ppl=1.01, grad_norm=0.40, lr=1.10e-05 | |
| 2025-11-14 03:39:14,945 - INFO - Epoch 1 Step 8410 (Global: 8410): loss=0.0066, ppl=1.01, grad_norm=0.34, lr=1.09e-05 | |
| 2025-11-14 03:41:08,647 - INFO - Epoch 1 Step 8420 (Global: 8420): loss=0.0095, ppl=1.01, grad_norm=0.48, lr=1.08e-05 | |
| 2025-11-14 03:43:12,664 - INFO - Epoch 1 Step 8430 (Global: 8430): loss=0.0055, ppl=1.01, grad_norm=0.33, lr=1.07e-05 | |
| 2025-11-14 03:45:06,438 - INFO - Epoch 1 Step 8440 (Global: 8440): loss=0.0069, ppl=1.01, grad_norm=0.37, lr=1.06e-05 | |
| 2025-11-14 03:47:00,003 - INFO - Epoch 1 Step 8450 (Global: 8450): loss=0.0070, ppl=1.01, grad_norm=0.36, lr=1.05e-05 | |
| 2025-11-14 03:48:53,810 - INFO - Epoch 1 Step 8460 (Global: 8460): loss=0.0073, ppl=1.01, grad_norm=0.38, lr=1.04e-05 | |
| 2025-11-14 03:50:57,621 - INFO - Epoch 1 Step 8470 (Global: 8470): loss=0.0075, ppl=1.01, grad_norm=0.37, lr=1.03e-05 | |
| 2025-11-14 03:52:51,039 - INFO - Epoch 1 Step 8480 (Global: 8480): loss=0.0062, ppl=1.01, grad_norm=0.36, lr=1.02e-05 | |
| 2025-11-14 03:54:45,147 - INFO - Epoch 1 Step 8490 (Global: 8490): loss=0.0059, ppl=1.01, grad_norm=0.33, lr=1.01e-05 | |
| 2025-11-14 03:56:39,081 - INFO - Epoch 1 Step 8500 (Global: 8500): loss=0.0087, ppl=1.01, grad_norm=0.40, lr=9.96e-06 | |
| 2025-11-14 03:56:39,085 - INFO - | |
| Running validation at step 8500... | |
| 2025-11-14 04:02:18,409 - INFO - Validation loss: 0.0077, perplexity: 1.01 | |
| 2025-11-14 04:02:18,410 - INFO - Qualitative metrics (n=5): | |
| 2025-11-14 04:02:18,410 - INFO - BLEU: 0.8139 | |
| 2025-11-14 04:02:18,410 - INFO - METEOR: 0.8771 | |
| 2025-11-14 04:02:18,410 - INFO - Edit Distance: 0.1035 | |
| 2025-11-14 04:02:18,410 - INFO - F-measure: 0.8821 | |
| 2025-11-14 04:02:18,411 - INFO - | |
| ====================================================================== | |
| 2025-11-14 04:02:18,411 - INFO - Qualitative Evaluation Samples: | |
| 2025-11-14 04:02:18,411 - INFO - ====================================================================== | |
| 2025-11-14 04:02:18,411 - INFO - | |
| Sample 1 (ID: sample_141920_chunk_1): | |
| 2025-11-14 04:02:18,411 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-14 04:02:18,412 - INFO - Generated: ' gave Q it four stars out of five and said that "Perhaps [the album\'s] seemingly illogical sequencing of songs makes sense if they wish to lure their audience into thinking it\'s as-you-were. But it\'s ...' | |
| 2025-11-14 04:02:18,412 - INFO - Ground Truth: 'Q gave it four stars out of five and said that "Perhaps [the album\'s] seemingly illogical sequencing of songs makes sense if they wish to lure their audience into thinking it\'s as-you-were. But it\'s n...' | |
| 2025-11-14 04:02:18,412 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-14 04:02:18,412 - INFO - | |
| Sample 2 (ID: sample_170543_chunk_2): | |
| 2025-11-14 04:02:18,412 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-14 04:02:18,412 - INFO - Generated: ', was Sirene Abou-Chakra, a Lebanese-American student who led the Arab-American Student Association. Other members included the woman president of the Michigan Student Assembly; the leader of Army ROT...' | |
| 2025-11-14 04:02:18,412 - INFO - Ground Truth: ', was Sirene Abou-Chakra, a Lebanese-American student who led the Arab-American Student Association. Other members included the woman president of the Michigan Student Assembly; the leader of Army ROT...' | |
| 2025-11-14 04:02:18,413 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-14 04:02:18,413 - INFO - | |
| Sample 3 (ID: sample_107152_chunk_9): | |
| 2025-11-14 04:02:18,413 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-14 04:02:18,413 - INFO - Generated: ' at the meeting Laymah headed. His investigator of goa is giant cutting, and he has a power to immobilise the opponents if his mind at they look in. the eye Oga falls for, the trick but Beel stops and...' | |
| 2025-11-14 04:02:18,414 - INFO - Ground Truth: ' at the meeting Laymia headed. His weapon of choice is a giant ax, and he has the power to immobilise his opponents if they look at him in the eye. Oga falls for the trick, but Beel stops the ax and b...' | |
| 2025-11-14 04:02:18,414 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-14 04:02:18,414 - INFO - | |
| Sample 4 (ID: sample_069148_chunk_0): | |
| 2025-11-14 04:02:18,414 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-14 04:02:18,414 - INFO - Generated: '# Oriya (Unicode block)\nOriya is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0B01.....' | |
| 2025-11-14 04:02:18,414 - INFO - Ground Truth: '# Oriya (Unicode block)\nOriya is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0B01.....' | |
| 2025-11-14 04:02:18,415 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-14 04:02:18,415 - INFO - | |
| Sample 5 (ID: sample_103176_chunk_4): | |
| 2025-11-14 04:02:18,415 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-14 04:02:18,415 - INFO - Generated: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Maxis Redwood Shores ...' | |
| 2025-11-14 04:02:18,415 - INFO - Ground Truth: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Maxis Redwood Shores ...' | |
| 2025-11-14 04:02:18,415 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-14 04:02:18,416 - INFO - | |
| Qualitative samples saved to: outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/qualitative_step_8500.jsonl | |
| 2025-11-14 04:02:57,558 - INFO - Saved checkpoint to outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/best_checkpoint.pt | |
| 2025-11-14 04:02:57,572 - INFO - New best validation loss: 0.0077, perplexity: 1.01 | |
| 2025-11-14 04:05:02,217 - INFO - Epoch 1 Step 8510 (Global: 8510): loss=0.0064, ppl=1.01, grad_norm=0.37, lr=9.86e-06 | |
| 2025-11-14 04:06:56,018 - INFO - Epoch 1 Step 8520 (Global: 8520): loss=0.0071, ppl=1.01, grad_norm=0.37, lr=9.76e-06 | |
| 2025-11-14 04:08:50,260 - INFO - Epoch 1 Step 8530 (Global: 8530): loss=0.0060, ppl=1.01, grad_norm=0.38, lr=9.67e-06 | |
| 2025-11-14 04:10:44,716 - INFO - Epoch 1 Step 8540 (Global: 8540): loss=0.0072, ppl=1.01, grad_norm=0.38, lr=9.57e-06 | |
| 2025-11-14 04:12:49,147 - INFO - Epoch 1 Step 8550 (Global: 8550): loss=0.0129, ppl=1.01, grad_norm=0.48, lr=9.47e-06 | |
| 2025-11-14 04:14:42,785 - INFO - Epoch 1 Step 8560 (Global: 8560): loss=0.0102, ppl=1.01, grad_norm=0.43, lr=9.37e-06 | |
| 2025-11-14 04:16:36,679 - INFO - Epoch 1 Step 8570 (Global: 8570): loss=0.0083, ppl=1.01, grad_norm=0.40, lr=9.27e-06 | |
| 2025-11-14 04:18:30,454 - INFO - Epoch 1 Step 8580 (Global: 8580): loss=0.0075, ppl=1.01, grad_norm=0.39, lr=9.18e-06 | |
| 2025-11-14 04:20:34,980 - INFO - Epoch 1 Step 8590 (Global: 8590): loss=0.0073, ppl=1.01, grad_norm=0.38, lr=9.08e-06 | |
| 2025-11-14 04:22:29,433 - INFO - Epoch 1 Step 8600 (Global: 8600): loss=0.0080, ppl=1.01, grad_norm=0.47, lr=8.98e-06 | |
| 2025-11-14 04:24:23,771 - INFO - Epoch 1 Step 8610 (Global: 8610): loss=0.0085, ppl=1.01, grad_norm=0.37, lr=8.89e-06 | |
| 2025-11-14 04:26:17,081 - INFO - Epoch 1 Step 8620 (Global: 8620): loss=0.0067, ppl=1.01, grad_norm=0.38, lr=8.79e-06 | |
| 2025-11-14 04:28:30,110 - INFO - Epoch 1 Step 8630 (Global: 8630): loss=0.0083, ppl=1.01, grad_norm=0.42, lr=8.70e-06 | |
| 2025-11-14 04:30:24,041 - INFO - Epoch 1 Step 8640 (Global: 8640): loss=0.0119, ppl=1.01, grad_norm=0.45, lr=8.60e-06 | |
| 2025-11-14 04:32:17,072 - INFO - Epoch 1 Step 8650 (Global: 8650): loss=0.0067, ppl=1.01, grad_norm=0.37, lr=8.51e-06 | |
| 2025-11-14 04:34:10,757 - INFO - Epoch 1 Step 8660 (Global: 8660): loss=0.0069, ppl=1.01, grad_norm=0.43, lr=8.42e-06 | |
| 2025-11-14 04:36:14,104 - INFO - Epoch 1 Step 8670 (Global: 8670): loss=0.0079, ppl=1.01, grad_norm=0.36, lr=8.32e-06 | |
| 2025-11-14 04:38:07,907 - INFO - Epoch 1 Step 8680 (Global: 8680): loss=0.0073, ppl=1.01, grad_norm=0.37, lr=8.23e-06 | |
| 2025-11-14 04:40:00,861 - INFO - Epoch 1 Step 8690 (Global: 8690): loss=0.0061, ppl=1.01, grad_norm=0.34, lr=8.14e-06 | |
| 2025-11-14 04:41:54,297 - INFO - Epoch 1 Step 8700 (Global: 8700): loss=0.0078, ppl=1.01, grad_norm=0.39, lr=8.05e-06 | |
| 2025-11-14 04:43:58,015 - INFO - Epoch 1 Step 8710 (Global: 8710): loss=0.0060, ppl=1.01, grad_norm=0.35, lr=7.96e-06 | |
| 2025-11-14 04:45:51,622 - INFO - Epoch 1 Step 8720 (Global: 8720): loss=0.0094, ppl=1.01, grad_norm=0.51, lr=7.87e-06 | |
| 2025-11-14 04:47:45,153 - INFO - Epoch 1 Step 8730 (Global: 8730): loss=0.0059, ppl=1.01, grad_norm=0.31, lr=7.78e-06 | |
| 2025-11-14 04:49:39,146 - INFO - Epoch 1 Step 8740 (Global: 8740): loss=0.0087, ppl=1.01, grad_norm=0.40, lr=7.69e-06 | |
| 2025-11-14 04:51:43,288 - INFO - Epoch 1 Step 8750 (Global: 8750): loss=0.0061, ppl=1.01, grad_norm=0.36, lr=7.60e-06 | |
| 2025-11-14 04:53:36,993 - INFO - Epoch 1 Step 8760 (Global: 8760): loss=0.0082, ppl=1.01, grad_norm=0.41, lr=7.51e-06 | |
| 2025-11-14 04:55:30,358 - INFO - Epoch 1 Step 8770 (Global: 8770): loss=0.0065, ppl=1.01, grad_norm=0.35, lr=7.42e-06 | |
| 2025-11-14 04:57:23,797 - INFO - Epoch 1 Step 8780 (Global: 8780): loss=0.0089, ppl=1.01, grad_norm=0.41, lr=7.33e-06 | |
| 2025-11-14 04:59:26,684 - INFO - Epoch 1 Step 8790 (Global: 8790): loss=0.0069, ppl=1.01, grad_norm=0.36, lr=7.25e-06 | |
| 2025-11-14 05:01:20,204 - INFO - Epoch 1 Step 8800 (Global: 8800): loss=0.0077, ppl=1.01, grad_norm=0.39, lr=7.16e-06 | |
| 2025-11-14 05:03:13,690 - INFO - Epoch 1 Step 8810 (Global: 8810): loss=0.0080, ppl=1.01, grad_norm=0.41, lr=7.07e-06 | |
| 2025-11-14 05:05:06,761 - INFO - Epoch 1 Step 8820 (Global: 8820): loss=0.0102, ppl=1.01, grad_norm=0.38, lr=6.99e-06 | |
| 2025-11-14 05:07:10,169 - INFO - Epoch 1 Step 8830 (Global: 8830): loss=0.0092, ppl=1.01, grad_norm=0.41, lr=6.90e-06 | |
| 2025-11-14 05:09:03,396 - INFO - Epoch 1 Step 8840 (Global: 8840): loss=0.0072, ppl=1.01, grad_norm=0.39, lr=6.82e-06 | |
| 2025-11-14 05:10:56,820 - INFO - Epoch 1 Step 8850 (Global: 8850): loss=0.0050, ppl=1.00, grad_norm=0.31, lr=6.74e-06 | |
| 2025-11-14 05:12:50,227 - INFO - Epoch 1 Step 8860 (Global: 8860): loss=0.0056, ppl=1.01, grad_norm=0.34, lr=6.65e-06 | |
| 2025-11-14 05:14:54,406 - INFO - Epoch 1 Step 8870 (Global: 8870): loss=0.0063, ppl=1.01, grad_norm=0.35, lr=6.57e-06 | |
| 2025-11-14 05:16:47,665 - INFO - Epoch 1 Step 8880 (Global: 8880): loss=0.0079, ppl=1.01, grad_norm=0.36, lr=6.49e-06 | |
| 2025-11-14 05:18:40,766 - INFO - Epoch 1 Step 8890 (Global: 8890): loss=0.0086, ppl=1.01, grad_norm=0.38, lr=6.40e-06 | |
| 2025-11-14 05:20:34,419 - INFO - Epoch 1 Step 8900 (Global: 8900): loss=0.0077, ppl=1.01, grad_norm=0.37, lr=6.32e-06 | |
| 2025-11-14 05:22:38,026 - INFO - Epoch 1 Step 8910 (Global: 8910): loss=0.0069, ppl=1.01, grad_norm=0.40, lr=6.24e-06 | |
| 2025-11-14 05:24:30,998 - INFO - Epoch 1 Step 8920 (Global: 8920): loss=0.0067, ppl=1.01, grad_norm=0.36, lr=6.16e-06 | |
| 2025-11-14 05:26:25,038 - INFO - Epoch 1 Step 8930 (Global: 8930): loss=0.0067, ppl=1.01, grad_norm=0.37, lr=6.08e-06 | |
| 2025-11-14 05:28:18,283 - INFO - Epoch 1 Step 8940 (Global: 8940): loss=0.0059, ppl=1.01, grad_norm=0.33, lr=6.00e-06 | |
| 2025-11-14 05:30:22,271 - INFO - Epoch 1 Step 8950 (Global: 8950): loss=0.0067, ppl=1.01, grad_norm=0.35, lr=5.92e-06 | |
| 2025-11-14 05:32:15,322 - INFO - Epoch 1 Step 8960 (Global: 8960): loss=0.0068, ppl=1.01, grad_norm=0.37, lr=5.84e-06 | |
| 2025-11-14 05:34:08,957 - INFO - Epoch 1 Step 8970 (Global: 8970): loss=0.0082, ppl=1.01, grad_norm=0.41, lr=5.76e-06 | |
| 2025-11-14 05:36:01,729 - INFO - Epoch 1 Step 8980 (Global: 8980): loss=0.0078, ppl=1.01, grad_norm=0.39, lr=5.68e-06 | |
| 2025-11-14 05:38:04,596 - INFO - Epoch 1 Step 8990 (Global: 8990): loss=0.0058, ppl=1.01, grad_norm=0.34, lr=5.61e-06 | |
| 2025-11-14 05:39:57,691 - INFO - Epoch 1 Step 9000 (Global: 9000): loss=0.0083, ppl=1.01, grad_norm=0.42, lr=5.53e-06 | |
| 2025-11-14 05:39:57,694 - INFO - | |
| Running validation at step 9000... | |
| 2025-11-14 05:45:34,528 - INFO - Validation loss: 0.0076, perplexity: 1.01 | |
| 2025-11-14 05:45:34,528 - INFO - Qualitative metrics (n=5): | |
| 2025-11-14 05:45:34,529 - INFO - BLEU: 0.8950 | |
| 2025-11-14 05:45:34,529 - INFO - METEOR: 0.8834 | |
| 2025-11-14 05:45:34,529 - INFO - Edit Distance: 0.0850 | |
| 2025-11-14 05:45:34,529 - INFO - F-measure: 0.9017 | |
| 2025-11-14 05:45:34,529 - INFO - | |
| ====================================================================== | |
| 2025-11-14 05:45:34,530 - INFO - Qualitative Evaluation Samples: | |
| 2025-11-14 05:45:34,530 - INFO - ====================================================================== | |
| 2025-11-14 05:45:34,530 - INFO - | |
| Sample 1 (ID: sample_141920_chunk_1): | |
| 2025-11-14 05:45:34,530 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-14 05:45:34,530 - INFO - Generated: ' gave Q it four stars out of five and said that "Perhaps [the album\'s] seemingly illogical sequencing of songs makes sense if they wish to lure their audience into thinking it\'s as-you-were. But it\'s ...' | |
| 2025-11-14 05:45:34,530 - INFO - Ground Truth: 'Q gave it four stars out of five and said that "Perhaps [the album\'s] seemingly illogical sequencing of songs makes sense if they wish to lure their audience into thinking it\'s as-you-were. But it\'s n...' | |
| 2025-11-14 05:45:34,531 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-14 05:45:34,531 - INFO - | |
| Sample 2 (ID: sample_170543_chunk_2): | |
| 2025-11-14 05:45:34,531 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-14 05:45:34,531 - INFO - Generated: ', was Sirene Abou-Chakra, a Lebanese-American student who led the Arab-American Student Association. Other members included the woman president of the Michigan Student Assembly; the leader of Army ROT...' | |
| 2025-11-14 05:45:34,531 - INFO - Ground Truth: ', was Sirene Abou-Chakra, a Lebanese-American student who led the Arab-American Student Association. Other members included the woman president of the Michigan Student Assembly; the leader of Army ROT...' | |
| 2025-11-14 05:45:34,531 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-14 05:45:34,532 - INFO - | |
| Sample 3 (ID: sample_107152_chunk_9): | |
| 2025-11-14 05:45:34,532 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-14 05:45:34,532 - INFO - Generated: ' at the meeting Laymah headed. His investigator of goa is giant cutting, and he has a power to immobilise the opponents if he mind at his looks. Oin the eyeога = fall, but the Jeffs stop are Nel and t...' | |
| 2025-11-14 05:45:34,532 - INFO - Ground Truth: ' at the meeting Laymia headed. His weapon of choice is a giant ax, and he has the power to immobilise his opponents if they look at him in the eye. Oga falls for the trick, but Beel stops the ax and b...' | |
| 2025-11-14 05:45:34,532 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-14 05:45:34,532 - INFO - | |
| Sample 4 (ID: sample_069148_chunk_0): | |
| 2025-11-14 05:45:34,533 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-14 05:45:34,533 - INFO - Generated: '# Oriya (Unicode block)\nOriya is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0B01.....' | |
| 2025-11-14 05:45:34,533 - INFO - Ground Truth: '# Oriya (Unicode block)\nOriya is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0B01.....' | |
| 2025-11-14 05:45:34,533 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-14 05:45:34,533 - INFO - | |
| Sample 5 (ID: sample_103176_chunk_4): | |
| 2025-11-14 05:45:34,533 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-14 05:45:34,533 - INFO - Generated: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Maxis Redwood Shores ...' | |
| 2025-11-14 05:45:34,534 - INFO - Ground Truth: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Maxis Redwood Shores ...' | |
| 2025-11-14 05:45:34,534 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-14 05:45:34,724 - INFO - | |
| Qualitative samples saved to: outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/qualitative_step_9000.jsonl | |
| 2025-11-14 05:46:16,600 - INFO - Saved checkpoint to outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/best_checkpoint.pt | |
| 2025-11-14 05:46:16,614 - INFO - New best validation loss: 0.0076, perplexity: 1.01 | |
| 2025-11-14 05:48:09,584 - INFO - Epoch 1 Step 9010 (Global: 9010): loss=0.0104, ppl=1.01, grad_norm=0.39, lr=5.45e-06 | |
| 2025-11-14 05:50:02,478 - INFO - Epoch 1 Step 9020 (Global: 9020): loss=0.0093, ppl=1.01, grad_norm=0.43, lr=5.38e-06 | |
| 2025-11-14 05:52:06,992 - INFO - Epoch 1 Step 9030 (Global: 9030): loss=0.0069, ppl=1.01, grad_norm=0.35, lr=5.30e-06 | |
| 2025-11-14 05:54:00,483 - INFO - Epoch 1 Step 9040 (Global: 9040): loss=0.0066, ppl=1.01, grad_norm=0.35, lr=5.23e-06 | |
| 2025-11-14 05:55:53,937 - INFO - Epoch 1 Step 9050 (Global: 9050): loss=0.0075, ppl=1.01, grad_norm=0.39, lr=5.15e-06 | |
| 2025-11-14 05:57:47,381 - INFO - Epoch 1 Step 9060 (Global: 9060): loss=0.0083, ppl=1.01, grad_norm=0.43, lr=5.08e-06 | |
| 2025-11-14 05:59:50,902 - INFO - Epoch 1 Step 9070 (Global: 9070): loss=0.0064, ppl=1.01, grad_norm=0.34, lr=5.01e-06 | |
| 2025-11-14 06:01:44,598 - INFO - Epoch 1 Step 9080 (Global: 9080): loss=0.0090, ppl=1.01, grad_norm=0.41, lr=4.93e-06 | |
| 2025-11-14 06:03:38,846 - INFO - Epoch 1 Step 9090 (Global: 9090): loss=0.0079, ppl=1.01, grad_norm=0.37, lr=4.86e-06 | |
| 2025-11-14 06:05:33,556 - INFO - Epoch 1 Step 9100 (Global: 9100): loss=0.0081, ppl=1.01, grad_norm=0.39, lr=4.79e-06 | |
| 2025-11-14 06:07:38,166 - INFO - Epoch 1 Step 9110 (Global: 9110): loss=0.0088, ppl=1.01, grad_norm=0.42, lr=4.72e-06 | |
| 2025-11-14 06:09:32,124 - INFO - Epoch 1 Step 9120 (Global: 9120): loss=0.0087, ppl=1.01, grad_norm=0.39, lr=4.65e-06 | |
| 2025-11-14 06:11:25,749 - INFO - Epoch 1 Step 9130 (Global: 9130): loss=0.0077, ppl=1.01, grad_norm=0.42, lr=4.58e-06 | |
| 2025-11-14 06:13:20,295 - INFO - Epoch 1 Step 9140 (Global: 9140): loss=0.0097, ppl=1.01, grad_norm=0.45, lr=4.51e-06 | |
| 2025-11-14 06:15:24,894 - INFO - Epoch 1 Step 9150 (Global: 9150): loss=0.0089, ppl=1.01, grad_norm=0.39, lr=4.44e-06 | |
| 2025-11-14 06:17:19,218 - INFO - Epoch 1 Step 9160 (Global: 9160): loss=0.0085, ppl=1.01, grad_norm=0.43, lr=4.37e-06 | |
| 2025-11-14 06:19:12,800 - INFO - Epoch 1 Step 9170 (Global: 9170): loss=0.0117, ppl=1.01, grad_norm=0.43, lr=4.30e-06 | |
| 2025-11-14 06:21:06,966 - INFO - Epoch 1 Step 9180 (Global: 9180): loss=0.0065, ppl=1.01, grad_norm=0.35, lr=4.23e-06 | |
| 2025-11-14 06:23:11,382 - INFO - Epoch 1 Step 9190 (Global: 9190): loss=0.0057, ppl=1.01, grad_norm=0.33, lr=4.17e-06 | |
| 2025-11-14 06:25:05,631 - INFO - Epoch 1 Step 9200 (Global: 9200): loss=0.0068, ppl=1.01, grad_norm=0.38, lr=4.10e-06 | |
| 2025-11-14 06:26:59,434 - INFO - Epoch 1 Step 9210 (Global: 9210): loss=0.0057, ppl=1.01, grad_norm=0.36, lr=4.03e-06 | |
| 2025-11-14 06:28:53,188 - INFO - Epoch 1 Step 9220 (Global: 9220): loss=0.0104, ppl=1.01, grad_norm=0.44, lr=3.97e-06 | |
| 2025-11-14 06:30:56,805 - INFO - Epoch 1 Step 9230 (Global: 9230): loss=0.0078, ppl=1.01, grad_norm=0.40, lr=3.90e-06 | |
| 2025-11-14 06:32:49,558 - INFO - Epoch 1 Step 9240 (Global: 9240): loss=0.0076, ppl=1.01, grad_norm=0.42, lr=3.84e-06 | |
| 2025-11-14 06:34:43,316 - INFO - Epoch 1 Step 9250 (Global: 9250): loss=0.0094, ppl=1.01, grad_norm=0.42, lr=3.77e-06 | |
| 2025-11-14 06:36:37,011 - INFO - Epoch 1 Step 9260 (Global: 9260): loss=0.0056, ppl=1.01, grad_norm=0.35, lr=3.71e-06 | |
| 2025-11-14 06:38:41,110 - INFO - Epoch 1 Step 9270 (Global: 9270): loss=0.0062, ppl=1.01, grad_norm=0.34, lr=3.65e-06 | |
| 2025-11-14 06:40:34,564 - INFO - Epoch 1 Step 9280 (Global: 9280): loss=0.0075, ppl=1.01, grad_norm=0.39, lr=3.58e-06 | |
| 2025-11-14 06:42:28,067 - INFO - Epoch 1 Step 9290 (Global: 9290): loss=0.0090, ppl=1.01, grad_norm=0.37, lr=3.52e-06 | |
| 2025-11-14 06:44:21,541 - INFO - Epoch 1 Step 9300 (Global: 9300): loss=0.0070, ppl=1.01, grad_norm=0.38, lr=3.46e-06 | |
| 2025-11-14 06:46:25,091 - INFO - Epoch 1 Step 9310 (Global: 9310): loss=0.0122, ppl=1.01, grad_norm=0.45, lr=3.40e-06 | |
| 2025-11-14 06:48:18,814 - INFO - Epoch 1 Step 9320 (Global: 9320): loss=0.0069, ppl=1.01, grad_norm=0.37, lr=3.34e-06 | |
| 2025-11-14 06:50:12,466 - INFO - Epoch 1 Step 9330 (Global: 9330): loss=0.0067, ppl=1.01, grad_norm=0.35, lr=3.28e-06 | |
| 2025-11-14 06:52:06,235 - INFO - Epoch 1 Step 9340 (Global: 9340): loss=0.0075, ppl=1.01, grad_norm=0.44, lr=3.22e-06 | |
| 2025-11-14 06:54:10,367 - INFO - Epoch 1 Step 9350 (Global: 9350): loss=0.0067, ppl=1.01, grad_norm=0.36, lr=3.16e-06 | |
| 2025-11-14 06:56:03,195 - INFO - Epoch 1 Step 9360 (Global: 9360): loss=0.0078, ppl=1.01, grad_norm=0.44, lr=3.10e-06 | |
| 2025-11-14 06:57:56,502 - INFO - Epoch 1 Step 9370 (Global: 9370): loss=0.0071, ppl=1.01, grad_norm=0.37, lr=3.05e-06 | |
| 2025-11-14 06:59:49,510 - INFO - Epoch 1 Step 9380 (Global: 9380): loss=0.0075, ppl=1.01, grad_norm=0.37, lr=2.99e-06 | |
| 2025-11-14 07:01:53,187 - INFO - Epoch 1 Step 9390 (Global: 9390): loss=0.0082, ppl=1.01, grad_norm=0.38, lr=2.93e-06 | |
| 2025-11-14 07:03:46,809 - INFO - Epoch 1 Step 9400 (Global: 9400): loss=0.0089, ppl=1.01, grad_norm=0.42, lr=2.88e-06 | |
| 2025-11-14 07:05:40,758 - INFO - Epoch 1 Step 9410 (Global: 9410): loss=0.0042, ppl=1.00, grad_norm=0.29, lr=2.82e-06 | |
| 2025-11-14 07:07:34,125 - INFO - Epoch 1 Step 9420 (Global: 9420): loss=0.0075, ppl=1.01, grad_norm=0.40, lr=2.76e-06 | |
| 2025-11-14 07:09:37,889 - INFO - Epoch 1 Step 9430 (Global: 9430): loss=0.0089, ppl=1.01, grad_norm=0.38, lr=2.71e-06 | |
| 2025-11-14 07:11:30,856 - INFO - Epoch 1 Step 9440 (Global: 9440): loss=0.0076, ppl=1.01, grad_norm=0.38, lr=2.66e-06 | |
| 2025-11-14 07:13:23,794 - INFO - Epoch 1 Step 9450 (Global: 9450): loss=0.0082, ppl=1.01, grad_norm=0.37, lr=2.60e-06 | |
| 2025-11-14 07:15:17,031 - INFO - Epoch 1 Step 9460 (Global: 9460): loss=0.0078, ppl=1.01, grad_norm=0.35, lr=2.55e-06 | |
| 2025-11-14 07:17:21,115 - INFO - Epoch 1 Step 9470 (Global: 9470): loss=0.0061, ppl=1.01, grad_norm=0.34, lr=2.50e-06 | |
| 2025-11-14 07:19:14,577 - INFO - Epoch 1 Step 9480 (Global: 9480): loss=0.0066, ppl=1.01, grad_norm=0.36, lr=2.44e-06 | |
| 2025-11-14 07:21:07,791 - INFO - Epoch 1 Step 9490 (Global: 9490): loss=0.0081, ppl=1.01, grad_norm=0.37, lr=2.39e-06 | |
| 2025-11-14 07:23:01,272 - INFO - Epoch 1 Step 9500 (Global: 9500): loss=0.0065, ppl=1.01, grad_norm=0.37, lr=2.34e-06 | |
| 2025-11-14 07:23:01,275 - INFO - | |
| Running validation at step 9500... | |
| 2025-11-14 07:28:37,104 - INFO - Validation loss: 0.0076, perplexity: 1.01 | |
| 2025-11-14 07:28:37,105 - INFO - Qualitative metrics (n=5): | |
| 2025-11-14 07:28:37,105 - INFO - BLEU: 0.8560 | |
| 2025-11-14 07:28:37,105 - INFO - METEOR: 0.8765 | |
| 2025-11-14 07:28:37,105 - INFO - Edit Distance: 0.1150 | |
| 2025-11-14 07:28:37,105 - INFO - F-measure: 0.8790 | |
| 2025-11-14 07:28:37,106 - INFO - | |
| ====================================================================== | |
| 2025-11-14 07:28:37,106 - INFO - Qualitative Evaluation Samples: | |
| 2025-11-14 07:28:37,106 - INFO - ====================================================================== | |
| 2025-11-14 07:28:37,106 - INFO - | |
| Sample 1 (ID: sample_141920_chunk_1): | |
| 2025-11-14 07:28:37,106 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-14 07:28:37,106 - INFO - Generated: ' gave Q it four stars out of five and said that "Perhaps [the album\'s] seemingly illogical sequencing of songs makes sense if they wish to lure their audience into thinking it\'s as-you-were. But it\'s ...' | |
| 2025-11-14 07:28:37,106 - INFO - Ground Truth: 'Q gave it four stars out of five and said that "Perhaps [the album\'s] seemingly illogical sequencing of songs makes sense if they wish to lure their audience into thinking it\'s as-you-were. But it\'s n...' | |
| 2025-11-14 07:28:37,106 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-14 07:28:37,106 - INFO - | |
| Sample 2 (ID: sample_170543_chunk_2): | |
| 2025-11-14 07:28:37,106 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-14 07:28:37,106 - INFO - Generated: ', was Sirene Abou-Chakra, a Lebanese-American student who led the Arab-American Student Association. Other members included the woman president of the Michigan Student Assembly; the leader of Army ROT...' | |
| 2025-11-14 07:28:37,107 - INFO - Ground Truth: ', was Sirene Abou-Chakra, a Lebanese-American student who led the Arab-American Student Association. Other members included the woman president of the Michigan Student Assembly; the leader of Army ROT...' | |
| 2025-11-14 07:28:37,107 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-14 07:28:37,107 - INFO - | |
| Sample 3 (ID: sample_107152_chunk_9): | |
| 2025-11-14 07:28:37,107 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-14 07:28:37,107 - INFO - Generated: ' at the meeting Laymah headed. His investigator of goa is giant cutting, and he has a power to immobilise the opponents if he mind at his looks ino. The eye ga forfall are cut, Douel hets as both底的 b...' | |
| 2025-11-14 07:28:37,107 - INFO - Ground Truth: ' at the meeting Laymia headed. His weapon of choice is a giant ax, and he has the power to immobilise his opponents if they look at him in the eye. Oga falls for the trick, but Beel stops the ax and b...' | |
| 2025-11-14 07:28:37,107 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-14 07:28:37,107 - INFO - | |
| Sample 4 (ID: sample_069148_chunk_0): | |
| 2025-11-14 07:28:37,107 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-14 07:28:37,107 - INFO - Generated: '# Oriya (Unicode block)\nOriya is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0B01.....' | |
| 2025-11-14 07:28:37,107 - INFO - Ground Truth: '# Oriya (Unicode block)\nOriya is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0B01.....' | |
| 2025-11-14 07:28:37,108 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-14 07:28:37,108 - INFO - | |
| Sample 5 (ID: sample_103176_chunk_4): | |
| 2025-11-14 07:28:37,108 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-14 07:28:37,108 - INFO - Generated: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Maxis Redwood Shores ...' | |
| 2025-11-14 07:28:37,108 - INFO - Ground Truth: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Maxis Redwood Shores ...' | |
| 2025-11-14 07:28:37,108 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-14 07:28:37,109 - INFO - | |
| Qualitative samples saved to: outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/qualitative_step_9500.jsonl | |
| 2025-11-14 07:29:15,728 - INFO - Saved checkpoint to outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/best_checkpoint.pt | |
| 2025-11-14 07:29:15,738 - INFO - New best validation loss: 0.0076, perplexity: 1.01 | |
| 2025-11-14 07:31:19,145 - INFO - Epoch 1 Step 9510 (Global: 9510): loss=0.0101, ppl=1.01, grad_norm=0.42, lr=2.29e-06 | |
| 2025-11-14 07:33:12,958 - INFO - Epoch 1 Step 9520 (Global: 9520): loss=0.0073, ppl=1.01, grad_norm=0.37, lr=2.24e-06 | |
| 2025-11-14 07:35:05,350 - INFO - Epoch 1 Step 9530 (Global: 9530): loss=0.0075, ppl=1.01, grad_norm=0.37, lr=2.19e-06 | |
| 2025-11-14 07:36:57,575 - INFO - Epoch 1 Step 9540 (Global: 9540): loss=0.0093, ppl=1.01, grad_norm=0.40, lr=2.14e-06 | |
| 2025-11-14 07:39:01,509 - INFO - Epoch 1 Step 9550 (Global: 9550): loss=0.0071, ppl=1.01, grad_norm=0.39, lr=2.10e-06 | |
| 2025-11-14 07:40:54,382 - INFO - Epoch 1 Step 9560 (Global: 9560): loss=0.0069, ppl=1.01, grad_norm=0.38, lr=2.05e-06 | |
| 2025-11-14 07:42:47,688 - INFO - Epoch 1 Step 9570 (Global: 9570): loss=0.0067, ppl=1.01, grad_norm=0.34, lr=2.00e-06 | |
| 2025-11-14 07:44:40,636 - INFO - Epoch 1 Step 9580 (Global: 9580): loss=0.0061, ppl=1.01, grad_norm=0.34, lr=1.95e-06 | |
| 2025-11-14 07:46:44,527 - INFO - Epoch 1 Step 9590 (Global: 9590): loss=0.0052, ppl=1.01, grad_norm=0.33, lr=1.91e-06 | |
| 2025-11-14 07:48:37,575 - INFO - Epoch 1 Step 9600 (Global: 9600): loss=0.0086, ppl=1.01, grad_norm=0.43, lr=1.86e-06 | |
| 2025-11-14 07:50:30,612 - INFO - Epoch 1 Step 9610 (Global: 9610): loss=0.0079, ppl=1.01, grad_norm=0.39, lr=1.82e-06 | |
| 2025-11-14 07:52:23,163 - INFO - Epoch 1 Step 9620 (Global: 9620): loss=0.0055, ppl=1.01, grad_norm=0.34, lr=1.77e-06 | |
| 2025-11-14 07:54:25,718 - INFO - Epoch 1 Step 9630 (Global: 9630): loss=0.0090, ppl=1.01, grad_norm=0.44, lr=1.73e-06 | |
| 2025-11-14 07:56:17,792 - INFO - Epoch 1 Step 9640 (Global: 9640): loss=0.0066, ppl=1.01, grad_norm=0.34, lr=1.68e-06 | |
| 2025-11-14 07:58:10,058 - INFO - Epoch 1 Step 9650 (Global: 9650): loss=0.0074, ppl=1.01, grad_norm=0.43, lr=1.64e-06 | |
| 2025-11-14 08:00:03,243 - INFO - Epoch 1 Step 9660 (Global: 9660): loss=0.0083, ppl=1.01, grad_norm=0.41, lr=1.60e-06 | |
| 2025-11-14 08:02:06,522 - INFO - Epoch 1 Step 9670 (Global: 9670): loss=0.0072, ppl=1.01, grad_norm=0.39, lr=1.56e-06 | |
| 2025-11-14 08:03:59,197 - INFO - Epoch 1 Step 9680 (Global: 9680): loss=0.0101, ppl=1.01, grad_norm=0.42, lr=1.52e-06 | |
| 2025-11-14 08:05:52,085 - INFO - Epoch 1 Step 9690 (Global: 9690): loss=0.0091, ppl=1.01, grad_norm=0.42, lr=1.48e-06 | |
| 2025-11-14 08:07:45,309 - INFO - Epoch 1 Step 9700 (Global: 9700): loss=0.0084, ppl=1.01, grad_norm=0.40, lr=1.44e-06 | |
| 2025-11-14 08:09:48,748 - INFO - Epoch 1 Step 9710 (Global: 9710): loss=0.0061, ppl=1.01, grad_norm=0.34, lr=1.40e-06 | |
| 2025-11-14 08:11:41,639 - INFO - Epoch 1 Step 9720 (Global: 9720): loss=0.0090, ppl=1.01, grad_norm=0.38, lr=1.36e-06 | |
| 2025-11-14 08:13:34,063 - INFO - Epoch 1 Step 9730 (Global: 9730): loss=0.0095, ppl=1.01, grad_norm=0.45, lr=1.32e-06 | |
| 2025-11-14 08:15:27,241 - INFO - Epoch 1 Step 9740 (Global: 9740): loss=0.0078, ppl=1.01, grad_norm=0.40, lr=1.28e-06 | |
| 2025-11-14 08:17:31,391 - INFO - Epoch 1 Step 9750 (Global: 9750): loss=0.0065, ppl=1.01, grad_norm=0.37, lr=1.24e-06 | |
| 2025-11-14 08:19:24,144 - INFO - Epoch 1 Step 9760 (Global: 9760): loss=0.0085, ppl=1.01, grad_norm=0.43, lr=1.21e-06 | |
| 2025-11-14 08:21:15,990 - INFO - Epoch 1 Step 9770 (Global: 9770): loss=0.0083, ppl=1.01, grad_norm=0.40, lr=1.17e-06 | |
| 2025-11-14 08:23:09,330 - INFO - Epoch 1 Step 9780 (Global: 9780): loss=0.0048, ppl=1.00, grad_norm=0.31, lr=1.13e-06 | |
| 2025-11-14 08:25:12,275 - INFO - Epoch 1 Step 9790 (Global: 9790): loss=0.0082, ppl=1.01, grad_norm=0.40, lr=1.10e-06 | |
| 2025-11-14 08:27:04,422 - INFO - Epoch 1 Step 9800 (Global: 9800): loss=0.0064, ppl=1.01, grad_norm=0.36, lr=1.06e-06 | |
| 2025-11-14 08:28:56,721 - INFO - Epoch 1 Step 9810 (Global: 9810): loss=0.0073, ppl=1.01, grad_norm=0.36, lr=1.03e-06 | |
| 2025-11-14 08:30:49,747 - INFO - Epoch 1 Step 9820 (Global: 9820): loss=0.0109, ppl=1.01, grad_norm=0.41, lr=9.97e-07 | |
| 2025-11-14 08:32:52,579 - INFO - Epoch 1 Step 9830 (Global: 9830): loss=0.0080, ppl=1.01, grad_norm=0.41, lr=9.64e-07 | |
| 2025-11-14 08:34:44,414 - INFO - Epoch 1 Step 9840 (Global: 9840): loss=0.0095, ppl=1.01, grad_norm=0.43, lr=9.32e-07 | |
| 2025-11-14 08:36:37,057 - INFO - Epoch 1 Step 9850 (Global: 9850): loss=0.0159, ppl=1.02, grad_norm=0.51, lr=9.00e-07 | |
| 2025-11-14 08:38:30,096 - INFO - Epoch 1 Step 9860 (Global: 9860): loss=0.0075, ppl=1.01, grad_norm=0.42, lr=8.68e-07 | |
| 2025-11-14 08:40:34,617 - INFO - Epoch 1 Step 9870 (Global: 9870): loss=0.0077, ppl=1.01, grad_norm=0.39, lr=8.37e-07 | |
| 2025-11-14 08:42:28,892 - INFO - Epoch 1 Step 9880 (Global: 9880): loss=0.0078, ppl=1.01, grad_norm=0.39, lr=8.07e-07 | |
| 2025-11-14 08:44:22,642 - INFO - Epoch 1 Step 9890 (Global: 9890): loss=0.0124, ppl=1.01, grad_norm=0.52, lr=7.77e-07 | |
| 2025-11-14 08:46:16,405 - INFO - Epoch 1 Step 9900 (Global: 9900): loss=0.0152, ppl=1.02, grad_norm=0.46, lr=7.48e-07 | |
| 2025-11-14 08:48:19,262 - INFO - Epoch 1 Step 9910 (Global: 9910): loss=0.0085, ppl=1.01, grad_norm=0.41, lr=7.20e-07 | |
| 2025-11-14 08:50:12,217 - INFO - Epoch 1 Step 9920 (Global: 9920): loss=0.0066, ppl=1.01, grad_norm=0.36, lr=6.92e-07 | |
| 2025-11-14 08:52:05,850 - INFO - Epoch 1 Step 9930 (Global: 9930): loss=0.0078, ppl=1.01, grad_norm=0.39, lr=6.64e-07 | |
| 2025-11-14 08:53:59,594 - INFO - Epoch 1 Step 9940 (Global: 9940): loss=0.0072, ppl=1.01, grad_norm=0.39, lr=6.37e-07 | |
| 2025-11-14 08:56:05,412 - INFO - Epoch 1 Step 9950 (Global: 9950): loss=0.0072, ppl=1.01, grad_norm=0.37, lr=6.11e-07 | |
| 2025-11-14 08:58:01,033 - INFO - Epoch 1 Step 9960 (Global: 9960): loss=0.0060, ppl=1.01, grad_norm=0.35, lr=5.85e-07 | |
| 2025-11-14 08:59:57,990 - INFO - Epoch 1 Step 9970 (Global: 9970): loss=0.0089, ppl=1.01, grad_norm=0.38, lr=5.60e-07 | |
| 2025-11-14 09:01:55,723 - INFO - Epoch 1 Step 9980 (Global: 9980): loss=0.0052, ppl=1.01, grad_norm=0.30, lr=5.35e-07 | |
| 2025-11-14 09:04:01,932 - INFO - Epoch 1 Step 9990 (Global: 9990): loss=0.0075, ppl=1.01, grad_norm=0.37, lr=5.11e-07 | |
| 2025-11-14 09:05:58,611 - INFO - Epoch 1 Step 10000 (Global: 10000): loss=0.0070, ppl=1.01, grad_norm=0.37, lr=4.87e-07 | |
| 2025-11-14 09:05:58,615 - INFO - | |
| Running validation at step 10000... | |
| 2025-11-14 09:11:48,660 - INFO - Validation loss: 0.0076, perplexity: 1.01 | |
| 2025-11-14 09:11:48,661 - INFO - Qualitative metrics (n=5): | |
| 2025-11-14 09:11:48,661 - INFO - BLEU: 0.8612 | |
| 2025-11-14 09:11:48,661 - INFO - METEOR: 0.8815 | |
| 2025-11-14 09:11:48,661 - INFO - Edit Distance: 0.1106 | |
| 2025-11-14 09:11:48,661 - INFO - F-measure: 0.8881 | |
| 2025-11-14 09:11:48,662 - INFO - | |
| ====================================================================== | |
| 2025-11-14 09:11:48,662 - INFO - Qualitative Evaluation Samples: | |
| 2025-11-14 09:11:48,662 - INFO - ====================================================================== | |
| 2025-11-14 09:11:48,662 - INFO - | |
| Sample 1 (ID: sample_141920_chunk_1): | |
| 2025-11-14 09:11:48,662 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-14 09:11:48,662 - INFO - Generated: ' gave Q it four stars out of five and said that "Perhaps [the album\'s] seemingly illogical sequencing of songs makes sense if they wish to lure their audience into thinking it\'s as-you-were. But it\'s ...' | |
| 2025-11-14 09:11:48,663 - INFO - Ground Truth: 'Q gave it four stars out of five and said that "Perhaps [the album\'s] seemingly illogical sequencing of songs makes sense if they wish to lure their audience into thinking it\'s as-you-were. But it\'s n...' | |
| 2025-11-14 09:11:48,663 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-14 09:11:48,663 - INFO - | |
| Sample 2 (ID: sample_170543_chunk_2): | |
| 2025-11-14 09:11:48,663 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-14 09:11:48,663 - INFO - Generated: ', was Sirene Abou-Chakra, a Lebanese-American student who led the Arab-American Student Association. Other members included the woman president of the Michigan Student Assembly; the leader of Army ROT...' | |
| 2025-11-14 09:11:48,663 - INFO - Ground Truth: ', was Sirene Abou-Chakra, a Lebanese-American student who led the Arab-American Student Association. Other members included the woman president of the Michigan Student Assembly; the leader of Army ROT...' | |
| 2025-11-14 09:11:48,664 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-14 09:11:48,664 - INFO - | |
| Sample 3 (ID: sample_107152_chunk_9): | |
| 2025-11-14 09:11:48,664 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-14 09:11:48,664 - INFO - Generated: ' at the meeting Laymah headed. His investigator of goa is giant cutting, and he has a power to immobilise the opponents if he mind at his looks ino. The eye ga forfall are cut, Douel hets as both底的 b...' | |
| 2025-11-14 09:11:48,664 - INFO - Ground Truth: ' at the meeting Laymia headed. His weapon of choice is a giant ax, and he has the power to immobilise his opponents if they look at him in the eye. Oga falls for the trick, but Beel stops the ax and b...' | |
| 2025-11-14 09:11:48,665 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-14 09:11:48,665 - INFO - | |
| Sample 4 (ID: sample_069148_chunk_0): | |
| 2025-11-14 09:11:48,665 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-14 09:11:48,665 - INFO - Generated: '# Oriya (Unicode block)\nOriya is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0B01.....' | |
| 2025-11-14 09:11:48,665 - INFO - Ground Truth: '# Oriya (Unicode block)\nOriya is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0B01.....' | |
| 2025-11-14 09:11:48,666 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-14 09:11:48,666 - INFO - | |
| Sample 5 (ID: sample_103176_chunk_4): | |
| 2025-11-14 09:11:48,666 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-14 09:11:48,666 - INFO - Generated: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Maxis Redwood Shores ...' | |
| 2025-11-14 09:11:48,666 - INFO - Ground Truth: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Maxis Redwood Shores ...' | |
| 2025-11-14 09:11:48,666 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-14 09:11:48,667 - INFO - | |
| Qualitative samples saved to: outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/qualitative_step_10000.jsonl | |
| 2025-11-14 09:12:29,802 - INFO - Saved checkpoint to outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/best_checkpoint.pt | |
| 2025-11-14 09:12:29,815 - INFO - New best validation loss: 0.0076, perplexity: 1.01 | |
| 2025-11-14 09:14:25,740 - INFO - Epoch 1 Step 10010 (Global: 10010): loss=0.0078, ppl=1.01, grad_norm=0.38, lr=4.64e-07 | |
| 2025-11-14 09:16:22,237 - INFO - Epoch 1 Step 10020 (Global: 10020): loss=0.0077, ppl=1.01, grad_norm=0.38, lr=4.42e-07 | |
| 2025-11-14 09:18:29,458 - INFO - Epoch 1 Step 10030 (Global: 10030): loss=0.0052, ppl=1.01, grad_norm=0.31, lr=4.20e-07 | |
| 2025-11-14 09:20:26,224 - INFO - Epoch 1 Step 10040 (Global: 10040): loss=0.0065, ppl=1.01, grad_norm=0.36, lr=3.98e-07 | |
| 2025-11-14 09:22:22,137 - INFO - Epoch 1 Step 10050 (Global: 10050): loss=0.0094, ppl=1.01, grad_norm=0.45, lr=3.78e-07 | |
| 2025-11-14 09:24:18,765 - INFO - Epoch 1 Step 10060 (Global: 10060): loss=0.0075, ppl=1.01, grad_norm=0.36, lr=3.57e-07 | |
| 2025-11-14 09:26:26,320 - INFO - Epoch 1 Step 10070 (Global: 10070): loss=0.0065, ppl=1.01, grad_norm=0.35, lr=3.38e-07 | |
| 2025-11-14 09:28:23,837 - INFO - Epoch 1 Step 10080 (Global: 10080): loss=0.0231, ppl=1.02, grad_norm=0.64, lr=3.18e-07 | |
| 2025-11-14 09:30:20,445 - INFO - Epoch 1 Step 10090 (Global: 10090): loss=0.0077, ppl=1.01, grad_norm=0.37, lr=3.00e-07 | |
| 2025-11-14 09:32:18,180 - INFO - Epoch 1 Step 10100 (Global: 10100): loss=0.0071, ppl=1.01, grad_norm=0.36, lr=2.82e-07 | |
| 2025-11-14 09:34:24,802 - INFO - Epoch 1 Step 10110 (Global: 10110): loss=0.0070, ppl=1.01, grad_norm=0.38, lr=2.64e-07 | |
| 2025-11-14 09:36:20,492 - INFO - Epoch 1 Step 10120 (Global: 10120): loss=0.0080, ppl=1.01, grad_norm=0.39, lr=2.47e-07 | |
| 2025-11-14 09:38:16,359 - INFO - Epoch 1 Step 10130 (Global: 10130): loss=0.0068, ppl=1.01, grad_norm=0.37, lr=2.31e-07 | |
| 2025-11-14 09:40:11,270 - INFO - Epoch 1 Step 10140 (Global: 10140): loss=0.0069, ppl=1.01, grad_norm=0.36, lr=2.15e-07 | |
| 2025-11-14 09:42:15,619 - INFO - Epoch 1 Step 10150 (Global: 10150): loss=0.0045, ppl=1.00, grad_norm=0.31, lr=2.00e-07 | |
| 2025-11-14 09:44:10,213 - INFO - Epoch 1 Step 10160 (Global: 10160): loss=0.0072, ppl=1.01, grad_norm=0.36, lr=1.85e-07 | |
| 2025-11-14 09:46:04,714 - INFO - Epoch 1 Step 10170 (Global: 10170): loss=0.0078, ppl=1.01, grad_norm=0.39, lr=1.71e-07 | |
| 2025-11-14 09:47:59,211 - INFO - Epoch 1 Step 10180 (Global: 10180): loss=0.0075, ppl=1.01, grad_norm=0.35, lr=1.58e-07 | |
| 2025-11-14 09:50:04,563 - INFO - Epoch 1 Step 10190 (Global: 10190): loss=0.0070, ppl=1.01, grad_norm=0.36, lr=1.45e-07 | |
| 2025-11-14 09:51:59,159 - INFO - Epoch 1 Step 10200 (Global: 10200): loss=0.0069, ppl=1.01, grad_norm=0.37, lr=1.32e-07 | |
| 2025-11-14 09:53:54,101 - INFO - Epoch 1 Step 10210 (Global: 10210): loss=0.0069, ppl=1.01, grad_norm=0.37, lr=1.20e-07 | |
| 2025-11-14 09:55:48,202 - INFO - Epoch 1 Step 10220 (Global: 10220): loss=0.0093, ppl=1.01, grad_norm=0.41, lr=1.09e-07 | |
| 2025-11-14 09:57:52,671 - INFO - Epoch 1 Step 10230 (Global: 10230): loss=0.0066, ppl=1.01, grad_norm=0.35, lr=9.81e-08 | |
| 2025-11-14 09:59:46,742 - INFO - Epoch 1 Step 10240 (Global: 10240): loss=0.0114, ppl=1.01, grad_norm=0.43, lr=8.79e-08 | |
| 2025-11-14 10:01:41,838 - INFO - Epoch 1 Step 10250 (Global: 10250): loss=0.0074, ppl=1.01, grad_norm=0.36, lr=7.83e-08 | |
| 2025-11-14 10:03:36,637 - INFO - Epoch 1 Step 10260 (Global: 10260): loss=0.0052, ppl=1.01, grad_norm=0.33, lr=6.92e-08 | |
| 2025-11-14 10:05:42,158 - INFO - Epoch 1 Step 10270 (Global: 10270): loss=0.0066, ppl=1.01, grad_norm=0.35, lr=6.06e-08 | |
| 2025-11-14 10:07:37,268 - INFO - Epoch 1 Step 10280 (Global: 10280): loss=0.0060, ppl=1.01, grad_norm=0.34, lr=5.27e-08 | |
| 2025-11-14 10:09:32,361 - INFO - Epoch 1 Step 10290 (Global: 10290): loss=0.0064, ppl=1.01, grad_norm=0.38, lr=4.53e-08 | |
| 2025-11-14 10:11:26,966 - INFO - Epoch 1 Step 10300 (Global: 10300): loss=0.0087, ppl=1.01, grad_norm=0.37, lr=3.84e-08 | |
| 2025-11-14 10:13:31,242 - INFO - Epoch 1 Step 10310 (Global: 10310): loss=0.0062, ppl=1.01, grad_norm=0.34, lr=3.21e-08 | |
| 2025-11-14 10:15:26,134 - INFO - Epoch 1 Step 10320 (Global: 10320): loss=0.0070, ppl=1.01, grad_norm=0.38, lr=2.64e-08 | |
| 2025-11-14 10:17:21,915 - INFO - Epoch 1 Step 10330 (Global: 10330): loss=0.0077, ppl=1.01, grad_norm=0.38, lr=2.12e-08 | |
| 2025-11-14 10:19:15,257 - INFO - Epoch 1 Step 10340 (Global: 10340): loss=0.0062, ppl=1.01, grad_norm=0.31, lr=1.66e-08 | |
| 2025-11-14 10:21:19,709 - INFO - Epoch 1 Step 10350 (Global: 10350): loss=0.0057, ppl=1.01, grad_norm=0.34, lr=1.26e-08 | |
| 2025-11-14 10:23:14,070 - INFO - Epoch 1 Step 10360 (Global: 10360): loss=0.0072, ppl=1.01, grad_norm=0.37, lr=9.12e-09 | |
| 2025-11-14 10:25:07,898 - INFO - Epoch 1 Step 10370 (Global: 10370): loss=0.0090, ppl=1.01, grad_norm=0.40, lr=6.20e-09 | |
| 2025-11-14 10:27:01,699 - INFO - Epoch 1 Step 10380 (Global: 10380): loss=0.0090, ppl=1.01, grad_norm=0.40, lr=3.84e-09 | |
| 2025-11-14 10:29:05,849 - INFO - Epoch 1 Step 10390 (Global: 10390): loss=0.0085, ppl=1.01, grad_norm=0.37, lr=2.05e-09 | |
| 2025-11-14 10:30:59,037 - INFO - Epoch 1 Step 10400 (Global: 10400): loss=0.0097, ppl=1.01, grad_norm=0.45, lr=8.11e-10 | |
| 2025-11-14 10:32:54,060 - INFO - Epoch 1 Step 10410 (Global: 10410): loss=0.0071, ppl=1.01, grad_norm=0.36, lr=1.38e-10 | |
| 2025-11-14 10:34:10,503 - INFO - Flushing 6 remainder batches from gradient accumulation | |
| 2025-11-14 10:34:10,508 - INFO - Rescaling gradients by 1.33x (compensating for 6/8 batches) | |
| 2025-11-14 10:34:10,768 - INFO - Remainder batch: loss=0.0059, ppl=1.01, grad_norm=0.48 | |
| 2025-11-14 10:34:10,784 - INFO - Epoch 1 training: loss=0.3403, ppl=1.41, grad_norm=0.79 | |
| 2025-11-14 10:34:10,796 - INFO - | |
| Running final validation... | |
| 2025-11-14 10:39:54,807 - INFO - Validation loss: 0.0076, perplexity: 1.01 | |
| 2025-11-14 10:39:54,807 - INFO - Qualitative metrics (n=5): | |
| 2025-11-14 10:39:54,807 - INFO - BLEU: 0.8907 | |
| 2025-11-14 10:39:54,808 - INFO - METEOR: 0.9243 | |
| 2025-11-14 10:39:54,808 - INFO - Edit Distance: 0.0913 | |
| 2025-11-14 10:39:54,808 - INFO - F-measure: 0.9133 | |
| 2025-11-14 10:39:54,808 - INFO - | |
| ====================================================================== | |
| 2025-11-14 10:39:54,808 - INFO - Qualitative Evaluation Samples: | |
| 2025-11-14 10:39:54,808 - INFO - ====================================================================== | |
| 2025-11-14 10:39:54,808 - INFO - | |
| Sample 1 (ID: sample_141920_chunk_1): | |
| 2025-11-14 10:39:54,808 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-14 10:39:54,808 - INFO - Generated: ' gave Q it four stars out of five and said that "Perhaps [the album\'s] seemingly illogical sequencing of songs makes sense if they wish to lure their audience into thinking it\'s as-you-were. But it\'s ...' | |
| 2025-11-14 10:39:54,808 - INFO - Ground Truth: 'Q gave it four stars out of five and said that "Perhaps [the album\'s] seemingly illogical sequencing of songs makes sense if they wish to lure their audience into thinking it\'s as-you-were. But it\'s n...' | |
| 2025-11-14 10:39:54,809 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-14 10:39:54,809 - INFO - | |
| Sample 2 (ID: sample_170543_chunk_2): | |
| 2025-11-14 10:39:54,809 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-14 10:39:54,809 - INFO - Generated: ', was Sirene Abou-Chakra, a Lebanese-American student who led the Arab-American Student Association. Other members included the woman president of the Michigan Student Assembly; the leader of Army ROT...' | |
| 2025-11-14 10:39:54,809 - INFO - Ground Truth: ', was Sirene Abou-Chakra, a Lebanese-American student who led the Arab-American Student Association. Other members included the woman president of the Michigan Student Assembly; the leader of Army ROT...' | |
| 2025-11-14 10:39:54,809 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-14 10:39:54,809 - INFO - | |
| Sample 3 (ID: sample_107152_chunk_9): | |
| 2025-11-14 10:39:54,809 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-14 10:39:54,809 - INFO - Generated: ' at the meeting Laymah headed. His investigator of goa is giant cutting, and he has a power to immobilise the opponents if his mind at they look in. the eye Oga falls for, the trick but Beel stops and...' | |
| 2025-11-14 10:39:54,809 - INFO - Ground Truth: ' at the meeting Laymia headed. His weapon of choice is a giant ax, and he has the power to immobilise his opponents if they look at him in the eye. Oga falls for the trick, but Beel stops the ax and b...' | |
| 2025-11-14 10:39:54,810 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-14 10:39:54,810 - INFO - | |
| Sample 4 (ID: sample_069148_chunk_0): | |
| 2025-11-14 10:39:54,810 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-14 10:39:54,810 - INFO - Generated: '# Oriya (Unicode block)\nOriya is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0B01.....' | |
| 2025-11-14 10:39:54,810 - INFO - Ground Truth: '# Oriya (Unicode block)\nOriya is a Unicode block containing characters for the Odia, Khondi and Santali languages of the state of Odisha in India. In its original incarnation, the code points U+0B01.....' | |
| 2025-11-14 10:39:54,810 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-14 10:39:54,810 - INFO - | |
| Sample 5 (ID: sample_103176_chunk_4): | |
| 2025-11-14 10:39:54,811 - INFO - Context: [Conv1D Residual compressed from 1000 tokens] | |
| 2025-11-14 10:39:54,811 - INFO - Generated: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Maxis Redwood Shores ...' | |
| 2025-11-14 10:39:54,811 - INFO - Ground Truth: ' |\n| The Sims 3: Generations | May 31, 2011 | Windows | Maxis Redwood Shores ...' | |
| 2025-11-14 10:39:54,811 - INFO - ---------------------------------------------------------------------- | |
| 2025-11-14 10:39:54,812 - INFO - | |
| Qualitative samples saved to: outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/qualitative_step_10417.jsonl | |
| 2025-11-14 10:40:41,699 - INFO - Saved checkpoint to outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252/best_checkpoint.pt | |
| 2025-11-14 10:40:41,711 - INFO - New best validation loss: 0.0076, perplexity: 1.01 | |
| 2025-11-14 10:40:41,715 - INFO - | |
| Training complete! | |
| 2025-11-14 10:40:41,715 - INFO - Final checkpoint is best, created symlink to save space (~2GB saved) | |
| 2025-11-14 10:40:41,716 - INFO - Best validation loss: 0.0076, perplexity: 1.01 | |
| 2025-11-14 10:40:41,716 - INFO - Checkpoints saved to outputs/production_conv1d_residual_t63_k5_reconstruction_20251112_221252 | |
| 2025-11-14 10:40:42,430 - INFO - W&B run finished | |