dystrio/Qwen3.5-122B-A10B-CacheReady

Add model card

Add model card

Add model card

Use original tokenizer_config.json (Qwen2TokenizerFast)

Update weight index with vision encoder + MTP layers

Add missing vision/MTP weights (model.safetensors-00063-of-00063.safetensors)

Add missing vision/MTP weights (model.safetensors-00062-of-00063.safetensors)

Replace flattened text-only config with original multimodal config (routing patch only changes weights)

Fix top-level model_type: qwen3_5_moe_text → qwen3_5_moe (match original)

Fix tokenizer_class: TokenizersBackend → PreTrainedTokenizerFast

Copy preprocessor_config.json from original model

Copy merges.txt from original model

Copy vocab.json from original model

Copy video_preprocessor_config.json from original model

Copy LICENSE from original model

Add model card

Baked Physarum routing canonicalization for prefix cache determinism

initial commit