Qwen/Qwen3.5-397B-A17B config.json

Fetched 2026-10-18 from the same commit as the card (8472618112ab). Top-level model_type qwen3_5_moe with a nested text_config of model_type qwen3_5_moe_text plus a vision_config. text_config: hidden_size 4096, 60 layers, 32 query heads, 2 kv heads, head_dim 256, 512 experts, 10 per token, moe_intermediate_size 1024, shared_expert_intermediate_size 1024, vocab 248320, silu. layer_types is 45 linear_attention and 15 full_attention, full_attention_interval 4. Linear block: linear_num_key_heads 16, linear_num_value_heads 64, key and value head dim 128. rope_parameters: rope_type default, rope_theta 10000000, partial_rotary_factor 0.25, mrope_interleaved true, mrope_section [11, 11, 10]. No rope_scaling key.