See axolotl config

axolotl version: 0.4.0

yaml

1base_model: meta-llama/Meta-Llama-3-70B-Instruct 2model_type: LlamaForCausalLM 3tokenizer_type: AutoTokenizer 4 5load_in_8bit: false 6load_in_4bit: false 7strict: false 8 9hub_model_id: shisa-ai/shisa-llama3-70b-v1 10hub_strategy: end 11 12use_wandb: true 13wandb_project: shisa-v2 14wandb_entity: augmxnt 15wandb_name: shisa-llama3-70b-v1 16 17chat_template: llama3 18datasets: 19 - path: augmxnt/ultra-orca-boros-en-ja-v1 20 type: sharegpt 21dataset_prepared_path: last_run_prepared 22val_set_size: 0.05 23output_dir: ./outputs/basemodel-llama3-70b 24 25sequence_len: 4096 26sample_packing: true 27pad_to_sequence_len: true 28 29gradient_accumulation_steps: 2 30micro_batch_size: 2 31num_epochs: 3 32optimizer: paged_adamw_8bit 33lr_scheduler: linear 34learning_rate: 2e-5 35 36train_on_inputs: false 37group_by_length: false 38bf16: auto 39fp16: 40tf32: true 41 42gradient_checkpointing: true 43gradient_checkpointing_kwargs: 44 use_reentrant: false 45early_stopping_patience: 46resume_from_checkpoint: 47logging_steps: 1 48xformers_attention: 49flash_attention: true 50 51warmup_ratio: 0.1 52evals_per_epoch: 2 53eval_table_size: 54saves_per_epoch: 0 55debug: 56deepspeed: axolotl/deepspeed_configs/zero3_bf16.json 57weight_decay: 0.05 58fsdp: 59fsdp_config: 60special_tokens: 61 pad_token: <|end_of_text|> 62

shisa-llama3-70b-v1

• Loss: 0.4425

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

• learning_rate: 2e-05
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• distributed_type: multi-GPU
• num_devices: 16
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• total_eval_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 87
• num_epochs: 3

Training results

Framework versions

• Transformers 4.40.2
• Pytorch 2.3.0+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1