🐦 MagpieLM-8B-SFT-v0.1

About This Model

Training procedure

Training hyperparameters

• learning_rate: 2e-05
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 32
• total_train_batch_size: 128
• total_eval_batch_size: 4
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 51
• num_epochs: 2

Training results

Framework versions

• Transformers 4.45.0.dev0
• Pytorch 2.3.0+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1

See axolotl config

axolotl version: 0.4.1

yaml

1base_model: meta-llama/Meta-Llama-3.1-8B 2model_type: LlamaForCausalLM 3tokenizer_type: AutoTokenizer 4chat_template: llama3 5 6load_in_8bit: false 7load_in_4bit: false 8strict: false 9main_process_port: 0 10 11datasets: 12 - path: Magpie-Align/MagpieLM-SFT-Data-v0.1 13 type: sharegpt 14 conversation: llama3 15 16dataset_prepared_path: last_run_prepared 17val_set_size: 0.001 18output_dir: axolotl_out/MagpieLM-8B-SFT-v0.1 19 20sequence_len: 8192 21sample_packing: true 22eval_sample_packing: false 23pad_to_sequence_len: true 24 25wandb_project: SynDa 26wandb_entity: 27wandb_watch: 28wandb_name: MagpieLM-8B-SFT-v0.1 29wandb_log_model: 30hub_model_id: Magpie-Align/MagpieLM-8B-SFT-v0.1 31 32gradient_accumulation_steps: 32 33micro_batch_size: 1 34num_epochs: 2 35optimizer: paged_adamw_8bit 36lr_scheduler: cosine 37learning_rate: 2e-5 38 39train_on_inputs: false 40group_by_length: false 41bf16: auto 42fp16: 43tf32: false 44 45gradient_checkpointing: true 46gradient_checkpointing_kwargs: 47 use_reentrant: false 48early_stopping_patience: 49resume_from_checkpoint: 50logging_steps: 1 51xformers_attention: 52flash_attention: true 53 54warmup_ratio: 0.1 55evals_per_epoch: 5 56eval_table_size: 57saves_per_epoch: 58debug: 59deepspeed: 60weight_decay: 0.0 61fsdp: 62fsdp_config: 63special_tokens: 64 pad_token: <|end_of_text|> 65