andreasgru/vllm-qwen38-flash-next-kv-quant
fp8_e4m3 and nvfp4 KV cache for the QSA path of Qwen3.8-Flash-Next in vLLM: container patch, kernel self-tests, paired measurements on one RTX PRO 6000 (SM120). Pool 1.889x / 3.059x at identical VRAM.
Python
1 contributors
Apache License 2.0
Social mentions
Recent discussions about this repository across the web
Repository activities
repository's daily and monthly activities across stars, forks, merged PRs, issues, and closed issues