| { | |
| "_name_or_path": "/usr/local/google/home/vishvak/Projects/datamux-pretraining-root/checkpoints/release_checkpoints/bert/base/gaussian_hadamard_index/sst2/10_gaussian_hadamard_index_5e-05_1_0_3_20000_128_4/checkpoint-15000", | |
| "architectures": [ | |
| "MuxedBertForSequenceClassification" | |
| ], | |
| "attention_probs_dropout_prob": 0.1, | |
| "binary_hadamard_epsilon": 0.0, | |
| "demuxing_variant": "index", | |
| "finetuning_task": "sst2", | |
| "gaussian_hadamard_norm": 1.0, | |
| "gradient_checkpointing": false, | |
| "head_temperature": 100.0, | |
| "hidden_act": "gelu", | |
| "hidden_dropout_prob": 0.1, | |
| "hidden_size": 768, | |
| "hierarchical_softmax_buckets": 100, | |
| "initializer_range": 0.02, | |
| "intermediate_size": 3072, | |
| "layer_norm_eps": 1e-12, | |
| "learn_muxing": 0, | |
| "legacy_demuxing": false, | |
| "max_position_embeddings": 512, | |
| "model_type": "bert", | |
| "muxing_variant": "gaussian_hadamard", | |
| "num_attention_heads": 12, | |
| "num_hidden_demux_layers": 3, | |
| "num_hidden_layers": 12, | |
| "num_instances": 10, | |
| "pad_token_id": 0, | |
| "position_embedding_type": "absolute", | |
| "retrieval_loss_coeff": 0.0, | |
| "retrieval_percentage": 1.0, | |
| "task_loss_coeff": 1.0, | |
| "transformers_version": "4.4.2", | |
| "type_vocab_size": 2, | |
| "use_cache": true, | |
| "use_hierarchical_softmax": false, | |
| "vocab_size": 30522 | |
| } | |