? ''
: ? ''
  : ? ''
    : hydra:
        run:
          dir: .
        output_subdir: null
        job:
          chdir: false
      _target_: null
      job:
        name: ???
        mode: slurm
        work_dir: null
        dry_run: false
      slurm:
        time_limit: ???
        constraint:
        - h200
        memory: 200
        cpus_per_task: 16
        partition: null
        mail_user: user@example.com
      execution:
        nodes: null
        gpus_per_node: null
        num_gpus: null
        hostfile: null
        secrets_file: null
      model:
        name_or_path: ???
        finetuning_type: lora
      dataset:
        name: ???
        dir: null
        info_json: null
        template: default
        cutoff_len: 1024
        val_size: 0.1
        hf_hub_url: null
        formatting: alpaca
        ranking: false
        subset: null
        split: train
        folder: null
        num_samples: null
        columns:
          prompt: null
          query: null
          response: null
          history: null
          messages: null
          system: null
          tools: null
          images: null
          videos: null
          audios: null
          chosen: null
          rejected: null
          kto_tag: null
        tags:
          role: null
          content: null
          user: null
          assistant: null
          observation: null
          function: null
          system: null
      training:
        stage: sft
        do_train: true
        model_name_or_path: null
        finetuning_type: lora
        trust_remote_code: true
        dataset: null
        dataset_dir: null
        template: default
        cutoff_len: 1024
        val_size: 0.1
        preprocessing_num_workers: 1
        dataset_num_proc: 1
        dataloader_num_workers: 0
        streaming: false
        learning_rate: 5.0e-05
        num_train_epochs: 3.0
        per_device_train_batch_size: 1
        per_device_eval_batch_size: 1
        gradient_accumulation_steps: 8
        lr_scheduler_type: cosine
        warmup_ratio: 0.1
        warmup_steps: 0
        lora_rank: 8
        lora_alpha: 16
        lora_dropout: 0.05
        lora_target: all
        optim: adamw_torch
        bf16: true
        fp16: false
        output_dir: null
        save_strategy: epoch
        save_steps: 500
        save_total_limit: 3
        save_only_model: false
        eval_strategy: steps
        eval_steps: 500
        do_eval: true
        logging_steps: 10
        plot_loss: true
        report_to: none
        gradient_checkpointing: true
        ddp_timeout: 180000000
        include_num_input_tokens_seen: true
        overwrite_output_dir: true
        overwrite_cache: false
        seed: 42
      lora:
        rank: 8
        alpha: 16
        dropout: 0.05
        target: all
      output:
        experiment_dir: ./experiments
      merge:
        stage: export
        model_name_or_path: null
        adapter_name_or_path: null
        template: default
        export_dir: null
        export_size: 2
        export_device: auto
        export_legacy_format: false
        finetuning_type: lora
      wandb:
        project: null
        run_name: null
        entity: null
      hf:
        repo_id: null
        private: false
        upload_artifacts: true
      cleanup:
        checkpoints: false
        merged: false
job:
  name: lf_torch_test__interactive
  mode: local
  work_dir: null
  dry_run: false
slurm:
  time_limit: null
  constraint: null
  memory: null
  partition: null
  mail_user: null
execution:
  nodes: 2
  gpus_per_node: 2
  num_gpus: null
  hostfile: /scratch/zrs2020/LlamaFactoryHelper/experiments/lf_torch_test__interactive/hostfile_auto_generated.txt
  secrets_file: ./secrets.env
model:
  name_or_path: Qwen/Qwen2.5-0.5B
  finetuning_type: lora
lora:
  rank: 8
  alpha: 16
  dropout: 0.05
  target: all
dataset:
  name: my_custom_sft12
  dir: null
  info_json: null
  template: default
  cutoff_len: 8096
  val_size: 0.1
  hf_hub_url: TAUR-dev/D-SFT_C-BASELINE_r1_distillation-sft-data
  formatting: sharegpt
  ranking: false
  subset: null
  split: train
  folder: null
  num_samples: null
  columns:
    messages: conversations
  tags:
    role: role
    content: content
    user: user
    assistant: assistant
  tokenized_path: /scratch/zrs2020/LlamaFactoryHelper/experiments/lf_torch_test__interactive/tokenized/my_custom_sft12
  data_shared_file_system: true
output:
  experiment_dir: ./experiments
wandb:
  project: null
  run_name: interactive_test
  entity: null
hf:
  repo_id: TAUR-dev/testing_llamafactory_helper_quick_test__interactive
  private: false
cleanup:
  checkpoints: false
  merged: false
training:
  stage: sft
  do_train: true
  max_steps: 100
  do_eval: false
  save_strategy: steps
  save_steps: 50
  logging_steps: 10
  fp16: true
  bf16: false
  overwrite_output_dir: true
  per_device_train_batch_size: 1
  gradient_accumulation_steps: 1
  gradient_checkpointing: true
  preprocessing_num_workers: 16
  overwrite_cache: true
merge: {}