syv-ai/qwen38-27b-rtx3090

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.

1 best points · 1 comments · 1 Hacker News post · 2,598 GitHub stars · View on GitHub

Hacker News discussions

• I pushed Qwen3.8-27B to 99 tps on a single request and 1150 on batch on a 3090 (1 points, 1 comments, by mess1337, 2026-08-17)