original/config.json
| 1 | { |
| 2 | "model_type": "privacy_filter", |
| 3 | "inference_contract_version": 1, |
| 4 | "encoding": "o200k_base", |
| 5 | "num_hidden_layers": 8, |
| 6 | "num_experts": 128, |
| 7 | "experts_per_token": 4, |
| 8 | "vocab_size": 200064, |
| 9 | "num_labels": 33, |
| 10 | "hidden_size": 640, |
| 11 | "intermediate_size": 640, |
| 12 | "head_dim": 64, |
| 13 | "num_attention_heads": 14, |
| 14 | "num_key_value_heads": 2, |
| 15 | "sliding_window": 257, |
| 16 | "bidirectional_context": true, |
| 17 | "bidirectional_left_context": 128, |
| 18 | "bidirectional_right_context": 128, |
| 19 | "initial_context_length": 4096, |
| 20 | "max_position_embeddings": 131072, |
| 21 | "default_n_ctx": 128000, |
| 22 | "rope_theta": 150000, |
| 23 | "rope_scaling_factor": 32.0, |
| 24 | "rope_ntk_alpha": 1.0, |
| 25 | "rope_ntk_beta": 32.0, |
| 26 | "param_dtype": "bfloat16" |
| 27 | } |