mfirth/l3t

See axolotl config

axolotl version: 0.5.3.dev38+g5726141c

base_model: meta-llama/Llama-3.2-3B-Instruct

datasets:
  - path: axolotl_format_data_llama.json
    type: input_output
dataset_prepared_path: last_run_prepared
    
output_dir: ./models/llama
sequence_length: 4096

wandb_project: agent-v0
wandb_name: llama-3b

train_on_inputs: false
gradient_accumulation_steps: 2
micro_batch_size: 1
num_epochs: 5
optimizer: adamw_torch
learning_rate: 2e-5

bf16: true

logging_steps: 10
flash_attention: true

warmup_steps: 50
saves_per_epoch: 1
weight_decay: 0.0

deepspeed: axolotl/deepspeed_configs/zero3_bf16.json

special_tokens:
  pad_token: <|end_of_text|>

models/llama

This model is a fine-tuned version of meta-llama/Llama-3.2-3B-Instruct on the axolotl_format_data_llama.json dataset.

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
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• total_eval_batch_size: 4
• optimizer: Use OptimizerNames.ADAMW_TORCH 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: 50
• num_epochs: 5

Training results

Framework versions

• Transformers 4.46.3
• Pytorch 2.5.1+cu124
• Datasets 3.1.0
• Tokenizers 0.20.3