Repository navigation
[PR] docs: refresh stale comments after #3900 / kernel 0.0.151 - #4203
ranxianglei wants to merge 2 commits into
Conversation
- budget.ts: the #453 output-budget clamp still runs only on the rebuilding lanes (openai/responses/google), but its old rationale ("Anthropic enforces its input limit independently of max_tokens") stopped being true with #3900's universal output-headroom reservation — reword to the current boundary. - wire-drop-warn.ts: kernel floor is 0.0.151 in kernel/package.json, not 0.0.150. Fixes #4202
📦 Built Package ArtifactBranch: Option A — Install this exact build (recommended)npm install -g billion-context@0.0.4998-pr.4203.239d3d6This exact version is immutable and reproducible. The rolling Option B — Download artifact
tar xzf billion-context-pr4203.tgz
npm install -g packageInstall instructions are refreshed on each push. |
(ai: vllm-qwen/qwen3.8-27b) 查重: 无。检索 open+closed issue/PR(关键词 stale comment / budget / wire-drop-warn / 0.0.150),无其他条目处理这两处注释漂移;本 PR 的立项依据就是 #4202,对应关系正确。 标签: pr + scope/wiring + p3(纯注释刷新,wire/请求管道域文件,无行为影响)。 这是 [PR] 类唤醒,未重做工作——拉取分支 diff 事实核对(逐项对过代码,非照抄正文):
回归审查:触碰路径承担的既有行为清单为空(注释不参与运行);高危模式(语义漂移/fail-fast 变吞错/阈值时序输出格式变化)逐项不适用。注释内容与代码事实的一致性已按上面三项逐条钉住。 CI(head f08bf3e):除 基线:分支落后 master 4 个提交(v0.1.204 发版 bump + #4198 revert),这 4 个提交均不触碰本 PR 的两个文件,GitHub 判定 mergeable=True、可干净合并;注释级 PR 无需 rebase,合并时自然收敛。 结构化八问:
判定:可合并。 唯一红 check 是 #3665 类 Windows runner 抖动(零断言失败,master 同车道亦红),不构成阻塞;如需洁癖绿,重跑该 job 即可。 |
(ai: zai/glm-5.3)
Comment-only follow-up from the v0.1.204 release review (PR #4190). Re-scoped after the owner's #4198 revert of #3900 landed mid-review:
src/wire-drop-warn.ts— kernel floor referenced as 0.0.150;kernel/package.jsonis 0.0.151.src/server/budget.ts— the plugin mode: no hard backstop when the agent ignores the compress nudge — context grows past the window and upstream 400s #453 clamp comment's "Non-Anthropic only — Anthropic enforces its input limit independently of max_tokens" rationale was stale while [BUG] acp-proxy upstream HTTP 400: Request with 1,198,269 tokens exceeds the model's 1,048,576 token limit, consistently | [BUG] acp-proxy upstream HTTP 400:请求 1,198,269 tokens 超过模型上限 1,048,576,一直复现——是插件压缩未生效,还是 DSH 的问题? #3900's universal reservation existed, but [PR] Revert "#3900: reserve output headroom on every wire (drop the Anthropic exemption)" (#4184) — unblock v0.1.204 #4198 reverted that; the comment now keeps the (again-accurate) rationale and records the revert trail for traceability.No behavior change (comments only); typecheck clean.
Fixes #4202