base_model: mistralai/Mistral-7B-v0.1 model_type: MistralForCausalLM tokenizer_type: LlamaTokenizer is_mistral_derived_model: true load_in_8bit: true load_in_4bit: false strict: false datasets: - path: sci-datasets/arc_challange_train_alpaca.json ds_type: json type: alpaca - path: sci-datasets/camelai_biology_alpaca.json ds_type: json type: alpaca - path: sci-datasets/camelai_chemistry_alpaca.json ds_type: json type: alpaca - path: sci-datasets/camelai_physics_alpaca.json ds_type: json type: alpaca - path: sci-datasets/openbookqa_alpaca.json ds_type: json type: alpaca - path: sci-datasets/reclor_science_alpaca.json ds_type: json type: alpaca - path: sci-datasets/scibench_alpaca.json ds_type: json type: alpaca - path: sci-datasets/scienceqa_alpaca.json ds_type: json type: alpaca - path: sci-datasets/theoremqa_alpaca.json ds_type: json type: alpaca - path: sci-datasets/tiger_scienceeval_alpaca.json ds_type: json type: alpaca dataset_prepared_path: last_run_prepared val_set_size: 0 output_dir: ./science-mistral adapter: lora lora_model_dir: sequence_len: 8192 sample_packing: true pad_to_sequence_len: true lora_r: 64 lora_alpha: 32 lora_dropout: 0.05 lora_target_linear: true lora_fan_in_fan_out: lora_target_modules: - gate_proj - down_proj - up_proj - q_proj - v_proj - k_proj - o_proj wandb_project: huggingface wandb_entity: wandb_watch: wandb_name: wandb_log_model: # change # gradient_accumulation_steps: 4 micro_batch_size: 4 num_epochs: 1 optimizer: adamw_bnb_8bit lr_scheduler: cosine learning_rate: 0.0002 # change # train_on_inputs: false group_by_length: false bf16: true fp16: false tf32: false gradient_checkpointing: true early_stopping_patience: resume_from_checkpoint: local_rank: logging_steps: 1 xformers_attention: flash_attention: true warmup_steps: 10 saves_per_epoch: 2 debug: deepspeed: weight_decay: 0.1 fsdp: fsdp_config: special_tokens: bos_token: "" eos_token: "" unk_token: ""