fix(content): correct two citation/spec errors; fix the rule that missed one - #87
Merged
Merged
Conversation
…sed one Three live posts corrected, and the guardrail I added in #86 replaced because it would not have caught the case that prompted it. 1. minicpm5-2b-review-2026: cited arXiv:2506.07900 as MiniCPM5-2B's technical report. That paper is "MiniCPM4: Ultra-Efficient LLMs on End Devices" (June 2025) — the previous generation. The card's other tag, 2602.09003, is "Data Science and Technology Towards AGI Part I: Tiered Data Management". Neither describes MiniCPM5, so the post now says so instead of promoting a near-miss. Both titles read off arxiv.org/abs/<id>. 2. glm-5-3-review-2026 and glm-5-3-flash-vs-qwen3-8-flash-next-comparison-2026: both listed the context length as 300K. That number is the evaluation context for the HLE benchmark in the cards' footnotes; config.json gives max_position_embeddings = 1048576 for both models. Same failure as the pricing bug — a footnote figure promoted to a spec. 3. The #86 rule said "prefer the newest arxiv tag unless the card ties an older one to this version". Applied to MiniCPM5-2B that picks 2602.09003, which is also wrong. Any ordering heuristic fails here, because the tag list carries no guarantee at all. Replaced with: fetch arxiv.org/abs/<id>, read the title, and cite it only if the title names this model and version. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vru5xPN55JPU7dyVS3uswd
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
sameerkhansf
enabled auto-merge (squash)
September 8, 2026 19:00
|
Caution Review failedThe pull request is closed. ℹ️ Recent review info⚙️ Run configurationConfiguration used: Path: .coderabbit.yaml Review profile: CHILL Plan: Advanced Run ID: 📒 Files selected for processing (5)
📝 WalkthroughWalkthroughThe changes update citation verification guidance, correct GLM-5.3 context claims, and correct the MiniCPM5 technical-report citation. The compiled workflow metadata reflects the guidance change. ChangesContent corrections
Estimated code review effort: 2 (Simple) | ~10 minutes ✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
This branch was successfully deployed
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.
Three live posts corrected. Also replaces the guardrail I added in #86, because it would not have caught the case that prompted it.
1. #85 cites the wrong paper (now live)
The post lists
arXiv:2506.07900as MiniCPM5-2B's "Technical Report". I read both tags off the model card straight from arxiv.org:2506.079002602.09003The first describes the previous generation; the second is an unrelated lab paper. Neither is a MiniCPM5 technical report, and the card links none. The post now says that outright rather than promoting the closest-looking tag.
2. "300K context" is a footnote, not a spec — in two posts
Both GLM posts state a 300K context length. That figure is the evaluation context for the HLE benchmark in the model cards' footnotes. The authoritative value from
config.json:Both are 1M. Z.ai's own evaluations cap context per benchmark — 300K for HLE, 400K for DeepSWE and Terminal-Bench 3.0, 1M for NL2Repo and ALE — so 300K was never the model limit. Exact same failure as the pricing bug in #80: a number lifted out of a footnote and promoted to a spec.
3. My #86 rule was wrong
#86 said: "prefer the newest arxiv tag unless the card explicitly ties an older ID to this version."
Applied to MiniCPM5-2B that selects
2602.09003— the data-management paper. My rule would have produced a different wrong citation. Any ordering heuristic fails, because the tag list carries no guarantee about subject at all.Replaced with the only thing that actually works: fetch
https://arxiv.org/abs/<id>, read the title, cite it only if the title names this model and version, and otherwise state that no technical report is linked. I should have verified the rule against the case before shipping it in #86.Verification
citation_titleon arxiv.org/absmax_position_embeddingsread from each model'sconfig.jsonupdated:frontmatter and dated correction notes added to the two changed postsgh aw compile,npm run lint,npm run typecheck,npm run validate:contentall clean🤖 Generated with Claude Code
https://claude.ai/code/session_01Vru5xPN55JPU7dyVS3uswd
Summary by CodeRabbit
Documentation
Chores