• Model creator: https://huggingface.co/mfirth/
• Original model: https://huggingface.co/mfirth/l3t-maybe-ok/

Original model description:

• generated_from_trainer datasets:
• axolotl_test.json model-index:
• name: models/llama_wm_v2 results: []

---

See axolotl config

axolotl version: 0.5.3.dev44+g5bef1906

yaml

1base_model: meta-llama/Llama-3.2-3B-Instruct 2 3plugins: 4 - axolotl.integrations.liger.LigerPlugin 5liger_rope: true 6liger_rms_norm: true 7liger_glu_activation: true 8liger_layer_norm: true 9liger_fused_linear_cross_entropy: true 10 11datasets: 12 - path: axolotl_test.json 13 type: input_output 14dataset_prepared_path: last_run_prepared 15 16output_dir: ./models/llama_wm_v2 17 18wandb_project: agent-v0 19wandb_name: llama-3b_wm_v2 20 21train_on_inputs: false 22gradient_checkpointing: true 23gradient_checkpointing_kwargs: 24 use_reentrant: false 25gradient_accumulation_steps: 2 26micro_batch_size: 4 27num_epochs: 5 28optimizer: adamw_torch 29learning_rate: 2e-5 30xformers_attention: 31flash_attention: true 32 33logging_steps: 5 34 35warmup_steps: 10 36saves_per_epoch: 1 37weight_decay: 0.0 38 39deepspeed: axolotl/deepspeed_configs/zero3_bf16_cpuoffload_all.json 40 41special_tokens: 42 pad_token: <|end_of_text|>

models/llama_wm_v2

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

• learning_rate: 2e-05
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• total_eval_batch_size: 32
• 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: 10
• num_epochs: 5

Training results

Framework versions

• Transformers 4.47.0
• Pytorch 2.5.1+cu124
• Datasets 3.1.0
• Tokenizers 0.21.0