| { |
| "model_class": "BERTModel", |
| "model_config": { |
| "name": "Albertina_M7", |
| "hidden_size": 768, |
| "ffn_factor": 4.0, |
| "vocab_size": 32768, |
| "bos_token_id": 5, |
| "eos_token_id": 6, |
| "pad_token_id": 0, |
| "mask_token_id": 4, |
| "masked_substitution_rate": 0.15, |
| "cloze_probability": 0.8, |
| "random_probability": 0.1, |
| "same_probability": 0.1, |
| "num_hidden_layers": 12, |
| "num_attention_heads": 12, |
| "tie_word_embeddings": false, |
| "rms_norm_eps": 1e-06, |
| "attention_type": [], |
| "max_position_embeddings": 1024, |
| "block_size_for_attention": 128, |
| "compile_flexattn": false, |
| "bias": false, |
| "default_layer": { |
| "attn_impl": "flash", |
| "sliding_window_size": null, |
| "positional_encoding": [ |
| "alibi", |
| "rope" |
| ], |
| "normalization": "rmsnorm", |
| "normalization_position": "pre", |
| "ffn_activation": "swiglu", |
| "hooks": {} |
| }, |
| "custom_layers": {} |
| }, |
| "training": { |
| "optimizer": "muon", |
| "lr_scheduling": true, |
| "lr": 0.0005, |
| "final_lr": 2e-05, |
| "hold_steps": 0.21, |
| "weight_decay": 0.01, |
| "scheduler": "custom", |
| "gradient_clip_val": 1.0, |
| "warmup_steps": 0.05, |
| "max_epochs": 1, |
| "accumulate_grad_batches": 16, |
| "seed": 27, |
| "save_every_n_steps": 5000, |
| "checkpoint_name": "albertina_M7" |
| }, |
| "tokenizer": { |
| "type": "huggingface", |
| "pretrained_name": "mrinaldi/Gettone", |
| "varlen_strategy": "padding" |
| }, |
| "data": { |
| "data_root": "/home/matteo/Albertone/Albertina/mini-albertina-2", |
| "batch_size": 48, |
| "num_workers": 1, |
| "mdat_strategy": "Gettone_1024", |
| "mdat_view": null |
| }, |
| "save_dir": "./checkpoints_ablations", |
| "wandb_project": "Albertina_Ablation_Studies", |
| "wandb_run_name": "Albertina_M7" |
| } |
