feat(lib-ollama-client): DEFAULT_NUM_CTX → 32768 (Phase C / 順位 98 root cause fix)#143
Merged
Conversation
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Pro Run ID: 📒 Files selected for processing (1)
✅ Files skipped from review due to trivial changes (1)
📝 WalkthroughWalkthroughOllama クライアントのコンテキストサイズを DEFAULT_NUM_CTX=8192 から 32768 に上げ、関連テストとドキュメントを更新し、push-runner の quality gate タイムアウトを 180→600 秒に延長、Phase C の検証結果と次フェーズ用 TODO を追記しました。 変更内容DEFAULT_NUM_CTX 段階的増加と Phase C 完了
🎯 3 (中程度) | ⏱️ ~20 分関連する可能性のあるプルリクエスト
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. 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 |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 1
🧹 Nitpick comments (1)
src/lib-ollama-client/src/lib.rs (1)
128-139: 💤 Low value詳細な進化履歴の配置を検討してください。
DEFAULT_NUM_CTXの doc comment に 12 行の詳細な進化履歴が記録されていますが、このレベルの詳細情報はコード内コメントより ADR-038 やdocs/local-llm-offload-history.mdに配置する方が適切かもしれません。推奨: コード内には以下の essential 情報のみを残し、詳細な dogfood 経緯は ADR に委譲する構成:
/// Ollama 既定の `num_ctx` (2048) は本リポジトリの lint-screen prompt に対して不足。 /// mistral:7b の theoretical max である 32768 を default とする。 /// 進化履歴の詳細は ADR-038 参照。 pub const DEFAULT_NUM_CTX: u32 = 32768;ただし、現状の詳細コメントも「コード近傍で判断根拠を即座に確認できる」利点があるため、プロジェクトの方針次第です。
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@src/lib-ollama-client/src/lib.rs` around lines 128 - 139, The doc comment for DEFAULT_NUM_CTX in lib.rs contains an overly long evolution history; trim it to the essential rationale (that Ollama's default 2048 is insufficient for our lint-screen prompts and we choose mistral:7b's theoretical max 32768 as the default) and replace the long timeline with a brief pointer to ADR-038 or docs/local-llm-offload-history.md for full details; locate the constant named DEFAULT_NUM_CTX in src/lib-ollama-client/src/lib.rs, remove the 12-line history block, leave a 2–3 line explanatory comment and add a single-line reference to the ADR/doc file.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@docs/todo-summary.md`:
- Line 78: The rank-109 mismatch: update the todo item so both sources agree on
Tier; locate the entry for rank 109 (the summary line showing "🚀 Tier 1") and
the detailed todo entry that currently labels "PR `#142` post-merge-feedback
T2-#1" and make them consistent—preferably change the detailed todo label
"T2-#1" to "T1-#1" (or alternatively change the summary Tier to Tier 2) and
ensure the text mentions "Tier 1" where Severity High/Phase C/D dogfood
precondition is asserted so both the summary and detailed todo reflect the same
Tier and label.
---
Nitpick comments:
In `@src/lib-ollama-client/src/lib.rs`:
- Around line 128-139: The doc comment for DEFAULT_NUM_CTX in lib.rs contains an
overly long evolution history; trim it to the essential rationale (that Ollama's
default 2048 is insufficient for our lint-screen prompts and we choose
mistral:7b's theoretical max 32768 as the default) and replace the long timeline
with a brief pointer to ADR-038 or docs/local-llm-offload-history.md for full
details; locate the constant named DEFAULT_NUM_CTX in
src/lib-ollama-client/src/lib.rs, remove the 12-line history block, leave a 2–3
line explanatory comment and add a single-line reference to the ADR/doc file.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: 212f08bf-ad7a-46d7-ae47-9817aebf9808
📒 Files selected for processing (5)
docs/local-llm-offload-analysis.mddocs/todo-summary.mddocs/todo6.mdpush-runner-config.tomlsrc/lib-ollama-client/src/lib.rs
Owner
Author
|
@coderabbitai review |
✅ Actions performedReview triggered.
|
5 tasks
aloekun
added a commit
that referenced
this pull request
May 11, 2026
…4) + workflow gap 反映 Phase D dogfood の 1 本目 (D-1)。3 件 ADR 採用 + Phase D 計画 land + workflow gap 反映を bundled。 順位 112 (PR #142 T3-#3): ADR-038 に 2 section 追記 - Diagnostic logging scope: eprintln (CLI 前提) と structured logging 移行条件 - 90% 閾値 rationale: 保守採用根拠 + Phase D 完了時の precision/recall 評価方針 順位 113 (PR #142 T3-#4): ADR-027 に metrics override 判断基準追記 - Incidental change vs Responsibility change の線引き - Override 記述様式 (Override / Reason / Rationale の 3 項目) 順位 114 (PR #143 T3-#1): ADR-040 新規作成 - mistral:7b context size 実測値 (8K 512MB 5-20s ↔ 32K 2GB 30-90s) - step_timeout 3.33x 比例係数の根拠 - num_ctx 選定 flow chart - CLAUDE.md ADR index 追加 - lib-ollama-client/src/lib.rs L128-139 dogfood evolution コメントを 4 行参照に短縮 Phase D 計画 land + workflow gap (新規): - analysis.md に D-1/D-2/D-3 PR 構成 + 計測手順 + 想定リスクを追加 - D-1 着手時に判明した workflow gap (jj auto-snapshot vs session-only opt-in) を計測手順に反映 - env var override (LINT_SCREEN_ENABLED) を todo8.md / todo-summary.md (順位 115) に backlog 登録 - D-1 自身は dogfood skip、env var override land 後に D-2 / D-3 で実 dogfood 実施 D-1 lint_screen 状態: skipped。Phase D guide §1 session-only opt-in workflow が jj auto-snapshot と 本質的に衝突するため、配列 env var override の cli-push-runner 実装 (順位 115) を D-2 前に land 必須。
6 tasks
aloekun
added a commit
that referenced
this pull request
May 12, 2026
…4) + workflow gap 反映 (#145) Phase D dogfood の 1 本目 (D-1)。3 件 ADR 採用 + Phase D 計画 land + workflow gap 反映を bundled。 順位 112 (PR #142 T3-#3): ADR-038 に 2 section 追記 - Diagnostic logging scope: eprintln (CLI 前提) と structured logging 移行条件 - 90% 閾値 rationale: 保守採用根拠 + Phase D 完了時の precision/recall 評価方針 順位 113 (PR #142 T3-#4): ADR-027 に metrics override 判断基準追記 - Incidental change vs Responsibility change の線引き - Override 記述様式 (Override / Reason / Rationale の 3 項目) 順位 114 (PR #143 T3-#1): ADR-040 新規作成 - mistral:7b context size 実測値 (8K 512MB 5-20s ↔ 32K 2GB 30-90s) - step_timeout 3.33x 比例係数の根拠 - num_ctx 選定 flow chart - CLAUDE.md ADR index 追加 - lib-ollama-client/src/lib.rs L128-139 dogfood evolution コメントを 4 行参照に短縮 Phase D 計画 land + workflow gap (新規): - analysis.md に D-1/D-2/D-3 PR 構成 + 計測手順 + 想定リスクを追加 - D-1 着手時に判明した workflow gap (jj auto-snapshot vs session-only opt-in) を計測手順に反映 - env var override (LINT_SCREEN_ENABLED) を todo8.md / todo-summary.md (順位 115) に backlog 登録 - D-1 自身は dogfood skip、env var override land 後に D-2 / D-3 で実 dogfood 実施 D-1 lint_screen 状態: skipped。Phase D guide §1 session-only opt-in workflow が jj auto-snapshot と 本質的に衝突するため、配列 env var override の cli-push-runner 実装 (順位 115) を D-2 前に land 必須。
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.
Summary
Phase d retirement への critical path Phase C (
DEFAULT_NUM_CTX増加) を land。Phase B で確定した「mistral:7b の context overflow が真因」を decisive に解消し、Phase D criteria (fallback rate < 50%) を classifier preview で達成。DEFAULT_NUM_CTX = 8192 → 32768(mistral:7b theoretical max)step_timeout = 180s → 600s(PR feat(cli-push-runner): Phase c MVP — pre-push lint-screen step (default OFF) #132 から 2 回目の bump)変更内容
Commit 1: docs only (前 turn carry-over)
順位 109-113 (PR #142 post-merge-feedback 採用分) を docs/todo6.md + docs/todo-summary.md に登録。
Commit 2: Phase C 実装
src/lib-ollama-client/src/lib.rsdoc comment を dogfood の進化記録 (2048 → 4096 → 8192 → 16384 → 32768) を含めて更新。Phase B 真因仮説 (overflow 時に Ollama は prompt_eval_count を num_ctx に clamp 報告) を inline で説明。
test 1 件 (
num_ctx_is_serialized_into_request_body) をnum_ctx:32768期待値に更新。lib-ollama-client 17/17 pass + cli-finding-classifier evals 20/20 pass。push-runner-config.tomlcargo test -- --ignored(12 件 mistral invoke) が num_ctx 32768 で local 269s 観測 → 180s 超過。600s への拡大で安全マージン込み。docs/local-llm-offload-analysis.mdPhase C row を ✅ 完了化 + Phase C smoke dogfood table を追加 (3 PRs replay 結果)。Phase D criteria (<50%) を classifier preview で 達成 表示。
Phase C smoke dogfood signal (32768 で 3 PRs replay、本 PR commit 2 後に取得)
auto_fix(real classification, finding detected)invalid severity: error= contract violation、num_ctx 起因ではない)auto_fix(real classification, finding detected)結果集計:
severity: "error"を返した = invalid)。Phase b' agreement 75% で説明可能な mistral semantic 精度の別問題、Phase C scope 外Critical path 進捗 (analysis.md 削除へ)
Test plan
cargo test -p lib-ollama-client17/17 passcargo test -p cli-finding-classifier --test lint_screen_evals20/20 pass (1 ignored = Ollama integration、別途 ignored test で確認)cargo test -- --ignored --test-threads=1全 pass (local 269s、step_timeout=600s で push pipeline pass).claude/cli-finding-classifier.exeで stderr に warn log emit されなくなったことを確認 (overflow 解消の subset 検証)Phase D 基準と残課題
Phase D criteria 達成判定の caveat:
残課題 (Phase E までに対応 / out-of-scope):
invalid severity: error) は semantic 精度問題、Phase b' agreement 75% で説明可能、Phase C scope 外Out of scope
invalid severity: error): mistral 出力崩壊系、prompt v4 検討 or model swap 別 phaselib-ollama-clientは本リポ専用、共有 crate 化は未計画Summary by CodeRabbit
バグ修正
ドキュメント