NVFP4

/

Qwen3-30B-A3B-Thinking-2507-FP4

like 3

Follow

NVFP4 85

I think there is an error on the config, did not work with sglang like the others fp4 models that has the structure like:
"kv_cache_scheme": {
"dynamic": false,
"num_bits": 8,
"type": "float"
}