| { | |
| "architectures": [ | |
| "ReformerModelWithLMHead" | |
| ], | |
| "attention_head_size": 128, | |
| "attn_layers": [ | |
| "local", | |
| "local", | |
| "lsh", | |
| "local", | |
| "local", | |
| "local", | |
| "lsh", | |
| "local", | |
| "local", | |
| "local", | |
| "lsh", | |
| "local" | |
| ], | |
| "axial_norm_std": 1.0, | |
| "axial_pos_embds": true, | |
| "axial_pos_embds_dim": [ | |
| 256, | |
| 768 | |
| ], | |
| "axial_pos_shape": [ | |
| 128, | |
| 512 | |
| ], | |
| "chunk_size_feed_forward": 0, | |
| "chunk_size_lm_head": 0, | |
| "eos_token_id": 2, | |
| "feed_forward_size": 4096, | |
| "hash_seed": null, | |
| "hidden_act": "relu", | |
| "hidden_dropout_prob": 0.2, | |
| "hidden_size": 1024, | |
| "initializer_range": 0.02, | |
| "is_decoder": true, | |
| "layer_norm_eps": 1e-12, | |
| "local_attention_probs_dropout_prob": 0.2, | |
| "local_attn_chunk_length": 128, | |
| "local_num_chunks_after": 0, | |
| "local_num_chunks_before": 1, | |
| "lsh_attention_probs_dropout_prob": 0.1, | |
| "lsh_attn_chunk_length": 256, | |
| "lsh_num_chunks_after": 0, | |
| "lsh_num_chunks_before": 1, | |
| "max_position_embeddings": 65536, | |
| "model_type": "reformer", | |
| "num_attention_heads": 8, | |
| "num_buckets": 512, | |
| "num_hashes": 4, | |
| "output_past": true, | |
| "pad_token_id": 0, | |
| "task_specific_params": { | |
| "text-generation": { | |
| "do_sample": true, | |
| "max_length": 100 | |
| } | |
| }, | |
| "vocab_size": 258 | |
| } |