| 文件 | 最后提交记录 | 最后更新时间 |
|---|---|---|
fix(llm-client): fall back when the first stream event is a context overflow (#542) Streaming calls now classify the first stream event before the call is declared successful. If that event is an in-band error whose message matches the existing overflow phrases (is_context_overflow), the call fails as ContextWindowExceeded and the ordered candidate loop falls through. Signed-off-by: Francesco Parisio <92143809+fparisio@users.noreply.github.com> | 1 个月前 | |
fix(benchmark): preserve terminal failure pressure (#768) Signed-off-by: Alex Fournier <afournier@nvidia.com> | 18 天前 |
| 文件 | 最后提交记录 | 最后更新时间 |
|---|---|---|
| 1 个月前 | ||
| 18 天前 |