See axolotl config

axolotl version: 0.4.1

yaml

1adapter: lora 2base_model: Qwen/Qwen1.5-0.5B-Chat 3bf16: auto 4chat_template: llama3 5dataset_prepared_path: null 6datasets: 7- data_files: 8 - 30a1f800cb30b101_train_data.json 9 ds_type: json 10 format: custom 11 path: /workspace/input_data/30a1f800cb30b101_train_data.json 12 type: 13 field_instruction: instruction 14 field_output: response 15 format: '{instruction}' 16 no_input_format: '{instruction}' 17 system_format: '{system}' 18 system_prompt: '' 19debug: null 20deepspeed: /workspace/axolotl/configs/deepspeed_stage2.json 21early_stopping_patience: null 22eval_max_new_tokens: 128 23eval_table_size: null 24evals_per_epoch: 4 25flash_attention: false 26fp16: null 27fsdp: null 28fsdp_config: null 29gradient_accumulation_steps: 8 30gradient_checkpointing: false 31group_by_length: false 32hub_model_id: PhoenixB/13c6e699-d8ae-4348-98f6-6fa5446b9dbf 33hub_repo: null 34hub_strategy: checkpoint 35hub_token: null 36learning_rate: 3e-5 37load_in_4bit: false 38load_in_8bit: false 39local_rank: null 40logging_steps: 1 41lora_alpha: 16 42lora_dropout: 0.05 43lora_fan_in_fan_out: null 44lora_model_dir: null 45lora_r: 8 46lora_target_linear: true 47lr_scheduler: cosine 48max_grad_norm: 1.0 49max_steps: 30 50micro_batch_size: 2 51mlflow_experiment_name: /tmp/30a1f800cb30b101_train_data.json 52model_type: AutoModelForCausalLM 53num_epochs: 10 54optimizer: adamw_bnb_8bit 55output_dir: miner_id_24 56pad_to_sequence_len: true 57resume_from_checkpoint: null 58s2_attention: null 59sample_packing: false 60saves_per_epoch: 4 61sequence_len: 512 62strict: false 63tf32: false 64tokenizer_type: AutoTokenizer 65train_on_inputs: false 66trust_remote_code: true 67val_set_size: 0.05 68wandb_entity: null 69wandb_mode: online 70wandb_name: 2ed8f0a3-fff0-40bb-a97b-5ba8a7efd7fb 71wandb_project: Gradients-On-Demand 72wandb_run: your_name 73wandb_runid: 2ed8f0a3-fff0-40bb-a97b-5ba8a7efd7fb 74warmup_steps: 10 75weight_decay: 0.01 76xformers_attention: null 77

13c6e699-d8ae-4348-98f6-6fa5446b9dbf

• Loss: 1.1665

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

• learning_rate: 3e-05
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 8
• total_train_batch_size: 128
• total_eval_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: 10
• training_steps: 30

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