Views
No views yet
compressed-tensors) de huihui-ai/Huihui-Qwen3-VL-30B-A3B-Thinking-abliterated, a su vez basado en Qwen/Qwen3-VL-30B-A3B-Thinking.embed_tokens, visual.* (torre de visión completa, en fp16), lm_head, mlp.gate (router de expertos)1vllm serve nonoskygt/Qwen3-VL-30B-A3B-Thinking-abliterated-GPTQ-INT4 \
2 --tensor-parallel-size 2 \
3 --max-model-len 131072 \
4 --gpu-memory-utilization 0.93 \
5 --kv-cache-dtype fp8 \
6 --enable-prefix-caching --enable-chunked-prefill \
7 --trust-remote-code \
8 --dtype float16 \
9 --limit-mm-per-prompt '{"image":2,"video":0}' \
10 --reasoning-parser qwen3 \
11 --enable-auto-tool-choice --tool-call-parser hermescompressed-tensors se autodetecta, no hace falta --quantization.