See axolotl config

axolotl version: 0.6.0

base_model: mrcuddle/Dark-Hermes3-Llama3.2-3B
dataloader_num_workers: 4
datasets:
- dataset_prepared_path: last_run_prepared
  path: llamafactory/alpaca_en
  type: alpaca
eval_steps: 500
evaluation_strategy: steps
fp16: true
gradient_accumulation_steps: 8
gradient_checkpointing: false
learning_rate: 2e-5
load_in_4bit: false
logging_dir: /content/outputs/logs
logging_steps: 10
lr_scheduler: cosine
lr_scheduler_type: cosine
micro_batch_size: 1
num_train_epochs: 3
optimizer: paged_adamw_8bit
output_dir: /content/outputs
overwrite_output_dir: true
per_device_train_batch_size: 4
save_steps: 500
save_total_limit: 2
use_peft: false
val_set_size: 0.05
warmup_steps: 100

content/outputs

This model is a fine-tuned version of mrcuddle/Dark-Hermes3-Llama3.2-3B on the llamafactory/alpaca_en dataset. It achieves the following results on the evaluation set:

Loss: 1.1205

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 2e-05
train_batch_size: 1
eval_batch_size: 1
seed: 42
gradient_accumulation_steps: 8
total_train_batch_size: 8
optimizer: Use paged_adamw_8bit with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: cosine
lr_scheduler_warmup_steps: 100
num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss
No log	0.0002	1	2.4030
1.2572	0.0814	500	1.1935
1.3061	0.1629	1000	1.1865
1.2733	0.2443	1500	1.1864
1.265	0.3258	2000	1.1753
1.2436	0.4072	2500	1.1542
1.2935	0.4887	3000	1.1448
1.2595	0.5701	3500	1.1348
1.2896	0.6515	4000	1.1295
1.2081	0.7330	4500	1.1236
1.2451	0.8144	5000	1.1212
1.2134	0.8959	5500	1.1205
1.2437	0.9773	6000	1.1205