| { | |
| "attention_backend": "PyTorch SDPA (Flash/Efficient/Math)", | |
| "src_tokenizer_type": "ByteLevelBPE", | |
| "tgt_tokenizer_type": "Unigram", | |
| "src_vocab_size_estimate": 10000, | |
| "tgt_vocab_size_estimate": 14938, | |
| "emb_size": 1536, | |
| "nhead": 8, | |
| "ffn_hid_dim": 4096, | |
| "num_encoder_layers": 8, | |
| "num_decoder_layers": 8, | |
| "dropout": 0.1, | |
| "max_len": 384, | |
| "batch_size_per_gpu": 128, | |
| "effective_batch_size": 768, | |
| "num_epochs": 50, | |
| "learning_rate": 5e-05, | |
| "weight_decay": 0.01, | |
| "grad_clip_norm": 1.0, | |
| "validation_split": 0.05, | |
| "random_seed": 42, | |
| "patience": 5, | |
| "precision": "bf16-mixed", | |
| "gpus": 6, | |
| "strategy": "ddp", | |
| "num_workers": 2, | |
| "pytorch_version": "2.6.0+cu124", | |
| "sdpa_available": true, | |
| "bos_token_id": 1, | |
| "eos_token_id": 2, | |
| "pad_token_id": 0, | |
| "tgt_vocab_size": 14938, | |
| "src_vocab_size": 10000 | |
| } |