• Model creator: https://huggingface.co/Magpie-Align/
• Original model: https://huggingface.co/Magpie-Align/Llama-3-8B-OpenHermes-243K/

Original model description:

• axolotl
• generated_from_trainer model-index:
• name: Llama-3-8B-OpenHermes-243K results: []

---

See axolotl config

axolotl version: 0.4.0

yaml

1base_model: meta-llama/Meta-Llama-3-8B 2model_type: LlamaForCausalLM 3tokenizer_type: AutoTokenizer 4 5load_in_8bit: false 6load_in_4bit: false 7strict: false 8 9datasets: 10 - path: mahiatlinux/OpenHermes-v1-ShareGPT 11 type: sharegpt 12 conversation: llama3 13dataset_prepared_path: last_run_prepared 14val_set_size: 0.001 15output_dir: ./out_Llama-8B-Openhermes-243K 16 17sequence_len: 8192 18sample_packing: true 19eval_sample_packing: false 20pad_to_sequence_len: true 21 22wandb_project: SynDa 23wandb_entity: 24wandb_watch: 25wandb_name: Llama-3-8B-OpenHermes-243K 26wandb_log_model: 27hub_model_id: SynDa/Llama-3-8B-OpenHermes-243K 28 29gradient_accumulation_steps: 8 30micro_batch_size: 1 31num_epochs: 2 32optimizer: paged_adamw_8bit 33lr_scheduler: cosine 34learning_rate: 2e-5 35 36train_on_inputs: false 37group_by_length: false 38bf16: auto 39fp16: 40tf32: false 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_steps: 100 52evals_per_epoch: 3 53eval_table_size: 54saves_per_epoch: 1 55debug: 56deepspeed: 57weight_decay: 0.0 58fsdp: 59fsdp_config: 60special_tokens: 61 pad_token: <|end_of_text|> 62

Llama-3-8B-OpenHermes-243K

• Loss: 0.9845

Model description

Intended uses & limitations

Training and evaluation data

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: 8
• total_train_batch_size: 32
• 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: 100
• num_epochs: 2

Training results

Framework versions

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