See axolotl config

axolotl version: 0.4.1

yaml

1adapter: lora 2base_model: samoline/59b1b15a-698b-4f85-a1f0-ff3f3edf67d9 3bf16: true 4chat_template: llama3 5dataloader_num_workers: 8 6dataloader_pin_memory: true 7dataset_prepared_path: null 8datasets: 9- data_files: 10 - cc898cbe9e7a591d_train_data.json 11 ds_type: json 12 format: custom 13 path: /workspace/input_data/ 14 type: 15 field_input: input 16 field_instruction: instruct 17 field_output: output 18 format: '{instruction} {input}' 19 no_input_format: '{instruction}' 20 system_format: '{system}' 21 system_prompt: '' 22debug: null 23deepspeed: null 24device_map: auto 25dynamic_lora_per_layer: true 26early_stopping_patience: 3 27eval_max_new_tokens: 128 28eval_steps: 100 29eval_table_size: null 30evaluation_strategy: steps 31flash_attention: true 32fp16: false 33fsdp: null 34fsdp_config: null 35gradient_accumulation_steps: 2 36gradient_checkpointing: true 37group_by_length: false 38hub_model_id: JoshMe1/59f4e3a8-1400-4b6a-aab8-36401a74a8be 39hub_repo: null 40hub_strategy: checkpoint 41hub_token: null 42learning_rate: 5.0e-05 43load_in_4bit: false 44load_in_8bit: true 45local_rank: null 46logging_steps: 10 47lora_alpha: 128 48lora_dropout: 0.1 49lora_fan_in_fan_out: null 50lora_model_dir: null 51lora_r: 64 52lora_target_linear: true 53lr_finder: true 54lr_scheduler: cosine 55lr_scheduler_args: [] 56max_grad_norm: 1.0 57max_memory: 58 0: 130GB 59max_steps: 30754 60micro_batch_size: 8 61mixed_precision: bf16 62mlflow_experiment_name: /tmp/cc898cbe9e7a591d_train_data.json 63model_type: AutoModelForCausalLM 64num_epochs: 10 65optimizer: adamw_bnb_8bit 66output_dir: miner_id_24 67pad_to_sequence_len: true 68resume_from_checkpoint: null 69s2_attention: null 70sample_packing: false 71save_steps: 100 72save_strategy: steps 73save_total_limit: 3 74scheduler: 75 factor: 0.5 76 monitor: eval_loss 77 patience: 1 78 threshold: 0.01 79 type: ReduceLROnPlateau 80sequence_len: 512 81strict: false 82tf32: false 83tokenizer_type: AutoTokenizer 84train_on_inputs: false 85training_stages: 86- learning_rate: 5.0e-05 87 name: warmup 88 num_train_epochs: 1 89- learning_rate: 5.0e-06 90 name: main 91trl: 92 ema: true 93 ema_decay: 0.999 94trust_remote_code: true 95val_set_size: 0.05 96wandb_entity: null 97wandb_mode: online 98wandb_name: fe5bc4e1-52ed-421a-affd-91e6a93d242b 99wandb_project: Gradients-On-Demand 100wandb_run: your_name 101wandb_runid: fe5bc4e1-52ed-421a-affd-91e6a93d242b 102warmup_steps: 3075 103weight_decay: 0.01 104xformers_attention: true 105

59f4e3a8-1400-4b6a-aab8-36401a74a8be

• Loss: 0.8564

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

• learning_rate: 5e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Use OptimizerNames.ADAMW_BNB 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: 3075
• training_steps: 30754

Training results

Framework versions

• PEFT 0.13.2
• Transformers 4.46.0
• Pytorch 2.5.0+cu124
• Datasets 3.0.1
• Tokenizers 0.20.1