Users report Qwen3.8 outputting garbage after prolonged use
trashacct383 · reddit · 2026-08-27
A user reports that Qwen3.8-27B-FP8 starts outputting nonsense or garbage after a few hours of use on vLLM, requiring a restart to fix. The user provided detailed configuration flags (including FlashInfer backend, xgrammar, MTP speculative decoding) and noted that adding a repetition penalty helped slightly, but vLLM 28 seemed to worsen the issue, asking if others are experiencing the same.
More from Infra
- Selling B300 inference nets $40/hr vs $6/hr rental — but the math is disputed — TheZachMueller · 2026-10-15
- ASML CFO says US Match Act restricting DUV sales to China would backfire — pstAsiatech · 2026-10-15
- ASML sold out through 2028 but stock dips 2%; TL20 index up 57% YTD — TiernanRayTech · 2026-10-15
- Bloomberg breakdown: a 1GW Texas AI data center may need nearly $3B for overruns alone — rohanpaul_ai · 2026-10-15
- 7 System Design Algorithms to Know: Merkle Trees to CAP/PACELC — goyalshaliniuk · 2026-10-15
- MiniCPM-V 4.7 runs a 35B MoE multimodal model on phones with just 3GB memory — nevrekaraishwa2 · 2026-10-15