{
  "activate_conditionality": null,
  "adam_beta1": 0.85,
  "adam_beta2": 0.98,
  "attention_dropout": 0.1,
  "auto_find_batch_size": false,
  "codecarbon": true,
  "cond_dropout": null,
  "cond_lr": null,
  "cond_wd": null,
  "condition_columns": null,
  "config": "/tmp/claude-1001/-home-cyprien/349b5436-5abc-447d-ba9d-b7a3a8831932/scratchpad/hf_cards/configs/CrystaLLM-pi_base/training_config.jsonc",
  "context_length": 1024,
  "data_seed": 1,
  "dataset_HF": "c-bone/lematerial_clean",
  "dataset_streaming": false,
  "deepspeed_config": "_config_files/deepspeed_default.json",
  "do_sample": "True",
  "early_stopping_patience": 10,
  "early_stopping_threshold": 1e-05,
  "embedding_dropout": 0.1,
  "eval_batch_size": 32,
  "eval_steps": 5000,
  "eval_strategy": "steps",
  "fp16": true,
  "gen_max_length": 1024,
  "grad_clip": 1.0,
  "gradient_accumulation_steps": 1,
  "greater_is_better": false,
  "input_parquet": null,
  "learning_rate": 0.001,
  "load_best_model_at_end": true,
  "logging_steps": 500,
  "lr_scheduler_kwargs": {
    "min_lr_rate": 0.001
  },
  "lr_scheduler_type": "cosine_with_min_lr",
  "max_return_attempts": 1,
  "max_samples": null,
  "max_steps": 1250000,
  "metric_for_best_model": "eval_loss",
  "model_ckpt_dir": "model_ckpts/cif-gpt2-small/checkpoint-400",
  "muon_lr": 0.02,
  "muon_momentum": 0.95,
  "n_embd": 512,
  "n_head": 8,
  "n_heads_sharing_slider": null,
  "n_hidden_cond": null,
  "n_layer": 8,
  "n_prefix_tokens": null,
  "num_return_sequences": 1,
  "optimizer": "adamw",
  "output_dir": "model_ckpts/mpdb-small-base-lematerial",
  "output_parquet": null,
  "pretrained_model_dir": null,
  "pretrained_tokenizer_dir": "HF-cif-tokenizer",
  "remove_CIFs_above_context": false,
  "remove_CIFs_with_unk": true,
  "report_to": "wandb",
  "residual_dropout": 0.1,
  "save_strategy": "steps",
  "save_total_limit": 2,
  "scoring_mode": "None",
  "seed": 1,
  "share_layers": null,
  "target_valid_cifs": 1,
  "temperature": 1.0,
  "top_k": 15,
  "top_p": 0.95,
  "torch_compile": true,
  "tracker_project": "CrystaLLM-pi",
  "train_batch_size": 32,
  "wandb_project_folder": "CrystaLLM_pi",
  "warmup_ratio": 0.02,
  "warmup_steps": 0,
  "weight_decay": 0.1
}