Two padding tokens doubled a hosted DeepSeek V4.1-Flash success rate, pointing to phase sensitivity
No_Pea_6625 · reddit · 2026-11-09
Running an identical 700-line Python refactor at temperature 0 across three DeepSeek V4.1-Flash hosts (9 runs each), the author got 7/9 clean on the official API vs 4/9 and 3/9 from resellers. Inspired by a Seed paper on phase sensitivity, adding one meaningless docstring line (two extra tokens) lifted Host B from 3/9 to 6/9 and Host A from 4/9 to 7/9. Same weights, same meaning, different prefix length, different quality—hard to attribute to quantization. Author admits n=9 and asks for cheaper validation designs.
More from Models
- Unverified claim: Opus 5.5 has only 10B active params; dev hands projects to Qwen-4-Flash — teortaxesTex · 2026-11-09
- Benchmarks are like perfume: why public leaderboards demand skepticism and private evals — intellectronica · 2026-11-09
- Running 8-25 Opus 5.5 threads all day used under 20% of a $200 Claude sub — cyrus_zei · 2026-11-09
- Grok Bot quietly throttled usage, then reset quotas after user backlash — xiaohu · 2026-11-09
- Is Opus 5.5 nerfed? Users report thinking patterns changed on November 3 — rzk · 2026-11-09
- DeepSeek-V4 and ERNIE 5.5 Flash Said to Use 4 Residual Streams with Sinkhorn Mixing — RobertTLange · 2026-11-09