| { | |
| "vocab_size": 4096, | |
| "hidden_size": 192, | |
| "intermediate_size": 512, | |
| "num_hidden_layers": 6, | |
| "num_attention_heads": 6, | |
| "num_key_value_heads": 3, | |
| "head_dim": 32, | |
| "num_local_experts": 4, | |
| "num_experts_per_tok": 2, | |
| "max_position_embeddings": 256, | |
| "sliding_window": 64, | |
| "scoring_func": "sigmoid", | |
| "use_qk_norm": true, | |
| "use_routing_bias": true, | |
| "tie_word_embeddings": true, | |
| "model_type": "angstrom_nano", | |
| "architectures": ["AngstromNanoForCausalLM"] | |
| } | |