{ "architecture_name": "OnnerReasearch", "model_name": "onner", "dataset_name": "indonesian-nlp/wikipedia-id", "vocab_size": 28000, "tokenizer_train_samples": 50000, "tokenizer_path": "onner_tokenizer.json", "cache_docs": 50000, "full_token_cache_path": "onner_tokens_full.pt", "train_token_cache_path": "onner_tokens_train.pt", "val_token_cache_path": "onner_tokens_val.pt", "val_ratio": 0.02, "seq_len": 512, "d_model": 768, "n_layers": 16, "n_heads": 12, "n_kv_heads": 4, "ffn_hidden": 2048, "dropout": 0.0, "rope_base": 10000.0, "mtp_n": 4, "mtp_loss_weights": [ 1.0, 0.3, 0.2, 0.1 ], "micro_batch_size": 8, "grad_accum_steps": 8, "max_steps": 500, "lr": 0.0003, "min_lr": 3e-05, "warmup_steps": 50, "weight_decay": 0.1, "grad_clip": 1.0, "eval_interval": 100, "eval_steps": 20, "save_interval": 500, "checkpoint_dir": "onner_checkpoints", "dataloader_workers": 0, "seed": 42, "parameter_count": 123964416, "parameter_count_millions": 123.964416, "model_type": "onner", "architecture": "OnnerReasearch", "library_name": "custom-pytorch" }