chore(issue-template): bug.yml asks for lisa --version first - #18
Merged
Conversation
The bug-report template told reporters to run \`git rev-parse --short HEAD\` for the version field — only works for people who cloned the repo, breaks for everyone who installed via npm or homebrew (both documented install paths). Since 84ba4b7 added \`lisa --version\`, every install path can report a stable version string with one command. Prefer it; keep the git sha fallback for dev-checkout reporters who want commit-level precision. Closes #17 Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
oratis
added a commit
that referenced
this pull request
Aug 1, 2026
一、第十一轮:2606.04703(迭代坍缩)3 票核定,文献 🔍 全部清零 - CE1 坍缩现象 3-0 确认:3 轮迭代内化后掉到 base 以下 (WebWalkerQA 21.0→18.7→8.5 vs base 16.6;Qwen3-4B/8B web agent) - 我方 2 处修正(自我推翻 17→19): #18 「off-policy from 高质量 teacher」→ teacher 是同一策略条件化于 经验池的自教师 + rejection sampling 只留成功轨迹,非外部强模型 #19 「坍缩诱因是无差别写入、门是对症解」(我方 💭)→ 论文归因是 instance 伪影/注入错位/on-policy 反应式监督,且其稳定配方本身 含 rejection sampling(任务成功)正确性门。「门能推迟坍缩」降级 为我方待验证假设,DESIGN #12 与论文草稿引用纪律已相应改写 - CE4:全文无 per-user 隔离(单一共享经验池)——利好 P4 主张 - rejection sampling 门属结果 oracle·训练数据准入类(STaR/LOCI 族), 不触「MDL × 推理期按条持久写入」两轴,空白声明不受影响 二、P1 实证复核通过(p1_eval.py, Qwen2.5-1.5B MPS) - 核心域 G1-G3:margin +4.44 nats,23/24,p=1e-6,AUC(z) 0.911 ——复现 p0e 手工集(+4.08, 19/22, p=0.00043),程序化生成不损信号 - G4 AUC(z) 0.500 精确复现 P0f 能力边界;G5 +0.23 小而一致(8/8) - 过滤 2 补跑(Qwen tokenizer):0 词被标记 - 过滤 3 补跑(零样本探针):AUC 0.394 ≈ 无先验信号 - 结论:生成器可放量供 P2 三臂对照 REPORT 8→11 轮全链同步;§5 文献风险清零;§6 推进到 P2 Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
oratis
added a commit
that referenced
this pull request
Aug 4, 2026
两份草稿里的裸 arXiv 号与命名占位符全部换成 \cite{}（31 处），
交叉核验：正文引用的 24 个 key **全部**在 bib 中，无悬空引用。
★ 会场填写纪律（写在文件抬头）——这条是自我推翻 #9 换来的：
当初把 LMLM 写成「NeurIPS 2025 主会」，实为 CCFM workshop Oral。
⟹ booktitle/journal 只在两种证据之一存在时才填：
(a) arXiv 元数据的 comment/journal-ref 自陈；(b) 我方 r9/r10/r11 全文核查记录。
其余一律 @misc + eprint。我方文档声称过、但两条证据都没有的，
标 % VENUE-UNVERIFIED 并保留 @misc（目前 2 条：Delétang ICLR'24、SEMA CVPR'25）。
其它纪律写进各条目的 note 字段，让引用时无法绕过：
· chen2024generativeadapter — 不得写"天然按用户隔离"（#16，全文无 isolation 一词）
· kujanpaa2024injection — 不得写"多规模超过 RAG"（#14，HotpotQA 上 RAG 全面占优）
· ye2026opcd — 不得写"更好保留 OOD"（#15，是退化更少；checkpoint 按 test acc 挑）
· tan2025dyprag — 不得写"平均互相干扰"（#17，说反了）
· chen2026continual — teacher 是自教师不是外部强模型；坍缩归因不得改写（#18/#19）
· huang2024sharpening — DPI 论点须归 cover1999elements，不得写成该文定理（#10）
· stein2026gates + huang2022selfimprove — 禁止写"尚无 GT-free 写入准入门控"（#13）
· li2025selfsizing — 🚨 必须钉 v3，v1/v2 是另一篇
· colaco2026ratedistortion — 🔴🔴 最高优先权风险，同构提案；投稿前必须重扫
元数据来源：arXiv Atom API（2026-08-04）；Farquhar 一条走 Crossref DOI 取全字段。
未入库的 6 个背景引用（LOCI/DEN/CN-DPM/DER++/GSS/MIR/iCaRL）已在正文标 %TODO-BIB，
不静默留着——它们只出现在族 5/6 概览句里，不承载我方主张。
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes #17.
What
The bug-report template's version field told reporters:
That only works for people who cloned the repo. Anyone who installed via npm or homebrew — both documented install paths — has no git checkout, and ends up either typing "n/a" or abandoning the report.
Since
84ba4b7shipped `lisa --version`, every install surface can report a stable version with one command:Change
- type: input id: lisa-version attributes: - label: LISA commit / version - description: Output of \`git rev-parse --short HEAD\` - placeholder: e.g. ac3a541 + label: LISA version + description: Output of \`lisa --version\` (or \`git rev-parse --short HEAD\` if you're running from a clone) + placeholder: e.g. 0.2.0Dev-checkout reporters who want commit-level precision can still send the git sha — it's mentioned as the fallback.
Scope
🤖 Generated with Claude Code