cemig-temp/llama3.2-3b-instruct-tokenlab-cemigConvo-chatml-2048

See axolotl config

axolotl version: 0.13.0.dev0

base_model: meta-llama/Llama-3.2-3B-Instruct
trust_remote_code: true
strict: false

chat_template: chatml

load_in_8bit: false
load_in_4bit: false

plugins:
  - axolotl.integrations.liger.LigerPlugin

liger_rope: true
liger_rms_norm: true
liger_glu_activation: true
liger_layer_norm: true
liger_fused_linear_cross_entropy: true

datasets:
  - path: ./outputs/dataset_tokenlab/train
    type: chat_template
    weight: 0.8
  - path: ./outputs/dataset_cemig/train
    type: chat_template
    weight: 0.2

validation_datasets:
  - path: ./outputs/dataset_tokenlab/validation
    type: chat_template
    weight: 0.8
  - path: ./outputs/dataset_cemig/validation
    type: chat_template
    weight: 0.2

test_datasets:
  - path: ./outputs/dataset_tokenlab/test
    type: chat_template
    weight: 0.8
  - path: ./outputs/dataset_cemig/test
    type: chat_template
    weight: 0.2

val_set_size: 0.0

dataset_prepared_path: ./outputs/dataset_prepared
output_dir: ./outputs/cemig-sft-2gpu-chatml-2048/

sequence_len: 2048
sample_packing: true
pad_to_sequence_len: true
flash_attn: true

gradient_checkpointing: true
micro_batch_size: 4
gradient_accumulation_steps: 4

num_epochs: 2
optimizer: adamw_torch_fused
learning_rate: 1.0e-5
lr_scheduler: cosine
cosine_constant_lr_ratio: 0
cosine_min_lr_ratio: 0.1
warmup_ratio: 0.1
weight_decay: 0.0

bf16: true
tf32: true
save_only_model: true

logging_steps: 1
evals_per_epoch: 4
saves_per_epoch: 2

special_tokens:
  pad_token: <|finetune_right_pad_id|>

dataloader_num_workers: 4
dataloader_prefetch_factor: 2

wandb_project: llama32-3b-dados-cemig-chatml-2048
wandb_entity: null
wandb_name: llama3.2-3b-tokenlab-more-cemig-data-chatml-2048
wandb_log_model: checkpoint

outputs/cemig-sft-2gpu-chatml-2048/

This model is a fine-tuned version of meta-llama/Llama-3.2-3B-Instruct on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.6568
• Memory/max Active (gib): 30.79
• Memory/max Allocated (gib): 30.79
• Memory/device Reserved (gib): 35.42

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: 1e-05
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED 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: 7966
• training_steps: 79668

Training results

Training Loss	Epoch	Step	Validation Loss	Active (gib)	Allocated (gib)	Reserved (gib)
No log	0	0	1.9833	17.34	17.34	17.48
0.7873	0.2500	9959	0.7749	30.79	30.79	36.74
0.6617	0.5000	19918	0.6996	30.79	30.79	35.42
0.6314	0.7500	29877	0.6771	30.79	30.79	35.42
0.6538	1.0001	39836	0.6668	30.79	30.79	35.42
0.7352	1.2501	49795	0.6615	30.79	30.79	35.42
0.66	1.5001	59754	0.6585	30.79	30.79	35.42
0.6365	1.7501	69713	0.6568	30.79	30.79	35.42

Framework versions

• Transformers 4.57.1
• Pytorch 2.9.0+cu130
• Datasets 4.3.0
• Tokenizers 0.22.1