| { | |
| "model_type": "mamba", | |
| "d_model": 512, | |
| "d_intermediate": 0, | |
| "n_layer": 6, | |
| "vocab_size": 1880, | |
| "ssm_cfg": { | |
| "layer": "Mamba2" | |
| }, | |
| "attn_layer_idx": [2, 5], | |
| "attn_cfg": { | |
| "num_heads": 8, | |
| "head_dim": 64, | |
| "rotary_emb_dim": 32, | |
| "causal": true, | |
| "softmax_scale": null, | |
| "qkv_proj_bias": true, | |
| "out_proj_bias": true, | |
| "mlp_dim": 0 | |
| }, | |
| "rms_norm": true, | |
| "residual_in_fp32": true, | |
| "fused_add_norm": true, | |
| "pad_vocab_size_multiple": 8, | |
| "tie_embeddings": true, | |
| "dropout_rate": 0.1, | |
| "eos_token_id": 2, | |
| "bos_token_id": 1, | |
| "pad_token_id": 3, | |
| "num_labels": 1, | |
| "summary_type": "cls_index", | |
| "summary_use_proj": true, | |
| "summary_activation": "tanh", | |
| "summary_proj_to_labels": true, | |
| "summary_first_dropout": 0.1, | |
| "summary_hidden_size": 128 | |
| } | |