| { | |
| "model_name": "Qwen3-5M-MoE-2exp-active", | |
| "model_type": "Qwen3MoeForCausalLM", | |
| "tokenizer": "gpt2", | |
| "dtype": "bfloat16", | |
| "vocab_size": 50257, | |
| "hidden_size": 128, | |
| "num_layers": 8, | |
| "num_attention_heads": 4, | |
| "num_key_value_heads": 2, | |
| "head_dim": 32, | |
| "moe_intermediate_size": 1024, | |
| "num_experts": 64, | |
| "num_experts_per_tok": 2, | |
| "sliding_window": 512, | |
| "max_position_embeddings": 8192, | |
| "rope_theta": 500000, | |
| "layer_types": [ | |
| "sliding_attention", | |
| "sliding_attention", | |
| "full_attention", | |
| "sliding_attention", | |
| "sliding_attention", | |
| "full_attention", | |
| "sliding_attention", | |
| "sliding_attention" | |
| ], | |
| "max_window_layers": 6, | |
| "parameters_total": 208220928, | |
| "parameters_active": 13186816, | |
| "active_ratio": 0.03125, | |
| "positional_encoding": "rope", | |
| "normalization": "rmsnorm", | |
| "activation": "swiglu", | |
| "tie_word_embeddings": true | |
| } |