New
syv-ai/hyperqwen — Serving Qwen3.8-27B Optimized for a 24GB GPU
syv-ai/hyperqwen presents a vLLM patch and model preparation pipeline for serving Qwen3.8-27B on a single 24GB GPU, along with performance figures measured by the author.
Projects
syv-ai/hyperqwen
Serve large Qwen models fast on the GPUs you actually own. Qwen3.8-27B on a single 24 GB card with vLLM: 127 tok/s single-user (381 when the answer quotes the prompt), ~1,035 tok/s at 64 concurrent, 150k-262k context. vLLM patches, requant pipeline, benchmarks.
View on GitHub ↗- stars
- ★ 2K
- Forks
- 281
- Language
- Python
- License
- Apache-2.0
- Latest release
- —
Posts
1No matching stories.