architecture:
    backbone_dtype: bfloat16
    gradient_checkpointing: true
    intermediate_dropout: 0.0
    pretrained: true
    pretrained_weights: ''
augmentation:
    neftune_noise_alpha: 0.0
    random_parent_probability: 0.0
    skip_parent_probability: 0.0
    token_mask_probability: 0.0
dataset:
    add_eos_token_to_answer: true
    add_eos_token_to_prompt: true
    add_eos_token_to_system: true
    answer_column: output
    chatbot_author: spacedome
    chatbot_name: spacedome
    data_sample: 1.0
    data_sample_choice:
    - Train
    - Validation
    id_column: None
    limit_chained_samples: false
    mask_prompt_labels: true
    only_last_answer: false
    parent_id_column: None
    personalize: true
    prompt_column:
    - user_prompt
    prompt_column_separator: \n\n
    system_column: system_prompt
    text_answer_separator: <|answer|>
    text_prompt_start: <|prompt|>
    text_system_start: <|system|>
    train_dataframe: /workspace/data/user/spacedome_v2/spacedome_v2.parquet
    validation_dataframe: None
    validation_size: 0.02
    validation_strategy: automatic
environment:
    compile_model: false
    deepspeed_allgather_bucket_size: 1000000
    deepspeed_method: ZeRO2
    deepspeed_reduce_bucket_size: 1000000
    deepspeed_stage3_param_persistence_threshold: 1000000
    deepspeed_stage3_prefetch_bucket_size: 1000000
    find_unused_parameters: false
    gpus:
    - '0'
    huggingface_branch: main
    mixed_precision: true
    mixed_precision_dtype: bfloat16
    number_of_workers: 120
    seed: -1
    trust_remote_code: true
    use_deepspeed: false
experiment_name: sydney
llm_backbone: mistralai/Ministral-8B-Instruct-2410
logging:
    log_all_ranks: false
    log_step_size: absolute
    logger: None
    neptune_project: ''
    wandb_entity: ''
    wandb_project: ''
output_directory: /workspace/output/user/sydney/
prediction:
    batch_size_inference: 0
    do_sample: false
    max_length_inference: 2048
    max_time: 0.0
    metric: BLEU
    metric_gpt_model: gpt-3.5-turbo-0301
    metric_gpt_template: general
    min_length_inference: 5
    num_beams: 1
    num_history: 4
    repetition_penalty: 1.0
    stop_tokens: ''
    temperature: 0.0
    top_k: 0
    top_p: 1.0
problem_type: text_causal_language_modeling
tokenizer:
    add_prompt_answer_tokens: false
    max_length: 8192
    padding_quantile: 1.0
    tokenizer_kwargs: '{"use_fast": true, "add_prefix_space": false}'
training:
    attention_implementation: auto
    batch_size: 8
    differential_learning_rate: 1.0e-05
    differential_learning_rate_layers:
    - embed
    - head
    - backbone
    drop_last_batch: true
    epochs: 2
    evaluate_before_training: false
    evaluation_epochs: 1.0
    freeze_layers: []
    grad_accumulation: 1
    gradient_clip: 0.0
    learning_rate: 0.0001
    lora: true
    lora_alpha: 32
    lora_dropout: 0.1
    lora_r: 16
    lora_target_modules: ''
    lora_unfreeze_layers: []
    loss_function: TokenAveragedCrossEntropy
    min_learning_rate_ratio: 0.0
    optimizer: AdamW
    save_checkpoint: best
    schedule: Cosine
    train_validation_data: false
    use_dora: true
    use_rslora: false
    warmup_epochs: 0.0
    weight_decay: 0.0