architecture: backbone_dtype: float32 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: 0.2 dataset: add_eos_token_to_answer: false add_eos_token_to_prompt: false add_eos_token_to_system: false add_prompt_answer_tokens: false answer_column: score chatbot_author: H2O.ai chatbot_name: h2oGPT data_sample: 1.0 data_sample_choice: - Train - Validation limit_chained_samples: false mask_prompt_labels: true num_classes: 6 parent_id_column: None personalize: false prompt_column: - full_text system_column: None text_answer_separator: '' text_prompt_start: '' text_system_start: '' train_dataframe: /root/h2o-llmstudio/data/user/essay_train/essay_train.csv validation_dataframe: /root/h2o-llmstudio/data/user/essay_train/essay_train.csv validation_size: 0.1 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: 8 seed: -1 trust_remote_code: true use_deepspeed: false experiment_name: masked-mamba.1 llm_backbone: h2oai/h2ogpt-4096-llama2-7b logging: logger: Neptune neptune_project: samvelkoch/essay output_directory: /root/h2o-llmstudio/output/user/masked-mamba.1/ prediction: batch_size_inference: 0 metric: Accuracy problem_type: text_causal_classification_modeling tokenizer: add_prompt_answer_tokens: false max_length: 10240 padding_quantile: 1.0 tokenizer_kwargs: '{"use_fast": true, "add_prefix_space": false}' training: batch_size: 2 differential_learning_rate: 1.0e-05 differential_learning_rate_layers: - classification_head drop_last_batch: true epochs: 1 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: 16 lora_dropout: 0.05 lora_r: 4 lora_target_modules: '' lora_unfreeze_layers: [] loss_function: CrossEntropyLoss optimizer: AdamW save_checkpoint: last schedule: Cosine train_validation_data: false use_dora: false use_flash_attention_2: true warmup_epochs: 0.0 weight_decay: 1.0e-05