| { | |
| "encoder_config": { | |
| "d_model": 1024, | |
| "d_intermediate": 0, | |
| "n_layer": 24, | |
| "vocab_size": 5000, | |
| "max_position_embeddings": 4096, | |
| "ssm_cfg": { | |
| "layer": "Mamba2" | |
| }, | |
| "attn_layer_idx": [ | |
| 6, | |
| 18 | |
| ], | |
| "attn_cfg": { | |
| "causal": false, | |
| "d_conv": 0, | |
| "head_dim": 64, | |
| "num_heads": 16, | |
| "num_heads_kv": 8, | |
| "out_proj_bias": false, | |
| "qkv_proj_bias": false, | |
| "rotary_emb_dim": 64 | |
| }, | |
| "rms_norm": true, | |
| "residual_in_fp32": true, | |
| "fused_add_norm": true, | |
| "pad_vocab_size_multiple": 8, | |
| "tie_embeddings": false | |
| }, | |
| "decoder_config": { | |
| "d_model": 1024, | |
| "d_intermediate": 0, | |
| "n_layer": 24, | |
| "vocab_size": 5000, | |
| "max_position_embeddings": 4096, | |
| "ssm_cfg": { | |
| "layer": "Mamba2" | |
| }, | |
| "attn_layer_idx": [ | |
| 6, | |
| 18 | |
| ], | |
| "attn_cfg": { | |
| "causal": true, | |
| "d_conv": 0, | |
| "head_dim": 64, | |
| "num_heads": 16, | |
| "num_heads_kv": 8, | |
| "out_proj_bias": false, | |
| "qkv_proj_bias": false, | |
| "rotary_emb_dim": 64 | |
| }, | |
| "rms_norm": true, | |
| "residual_in_fp32": true, | |
| "fused_add_norm": true, | |
| "pad_vocab_size_multiple": 8, | |
| "tie_embeddings": false | |
| }, | |
| "tie_word_embeddings": true, | |
| "seed": 0 | |
| } |