Skip to content

fix(tinfoil): accept the auto effort Tinfoil lists for Inkling 2 Small - #13842

Merged
rekram1-node merged 3 commits into
devfrom
fix/tinfoil-auto-effort
Nov 16, 2026
Merged

rekram1-node merged 3 commits into
devfrom
fix/tinfoil-auto-effort

Conversation

@rekram1-node

@rekram1-node rekram1-node commented Nov 16, 2026 •

Copy link
Copy Markdown
Collaborator

The Tinfoil sync started failing today. Tinfoil's /v1/models now lists auto in supported_efforts for inkling-2-small, and the schema only took the catalog's effort levels, so the whole parse threw.

auto isn't in Tinfoil's reasoning docs (https://docs.tinfoil.sh/guides/reasoning.md lists the accepted values per model, and Inkling isn't there), and it isn't a catalog level. Renaming it to default would make clients send reasoning_effort: "default", which Tinfoil doesn't document. So the schema accepts auto and drops it from the published list. Any other unknown value still fails the sync.

Dry run: 0 created, 0 updated, 0 removed, 11 unchanged. Inkling 2 Small has no Tinfoil file today, so nothing published changes. The test covers both the parse and the reasoning_options buildTinfoilModel publishes.

@github-actions

Copy link
Copy Markdown
Contributor

Action items

  • [low] [possible mistake] packages/core/src/sync/providers/tinfoil.ts:27 - Check: Dropping auto from supported_efforts is justified only if reasoning_effort: "auto" is truly equivalent to omitting the parameter on Tinfoil. Why: The file's own comment (tinfoil.ts:92-93) documents supported_efforts as "the reasoning_effort values Tinfoil accepts", so after this change every published Tinfoil effort list silently omits a value the host explicitly accepts. That is only lossless if the PR body's claim ("the same as sending no effort") holds, and that claim has no cited Tinfoil docs or live probe; the repo also has a first-class effort value for exactly this host-published "model picks its depth" level (default in packages/core/src/schema-v2.ts:136-145, published verbatim for hosts that document one, e.g. providers/groq/models/qwen/qwen3.6-27b.toml:14 with reasoning_effort = none|default). Action: Cite Tinfoil documentation or a live probe (one request with reasoning_effort: "auto", one without) confirming auto behaves identically to omission. If auto is instead a distinct selectable depth, keep it representable — map it to the catalog's default effort value and add default to EFFORT_ORDER (tinfoil.ts:90) — rather than discarding it.

@rekram1-node

Copy link
Copy Markdown
Collaborator Author

Good point. auto now maps to the catalog's default effort level (the same one aiand and aimlapi publish) instead of being dropped, so the list keeps everything Tinfoil accepts. The schema also takes default, because the fetch output is parsed twice. Dry run is still clean.

@github-actions

Copy link
Copy Markdown
Contributor

Action items

  • [medium][possible mistake] packages/core/src/sync/providers/tinfoil.ts:29 - Check: Effort values published for Tinfoil must reflect the reasoning_effort values its wire actually accepts. Why: The transform renames the catalog token auto to catalog default, and tinfoilReasoningOptions (which replaces the authored effort list for every tracked Tinfoil model) will publish effort = […, "default"] — a value Tinfoil's /v1/models does not list. The only published default today (providers/groq/models/qwen/qwen3.8-27b.toml:20) is a literal Groq wire value, while Eden AI's sync (ACCEPTED_EFFORTS, edenai.ts:34) treats default as non-sendable when deriving from lab entries, so the catalog convention for default is not settled. The PR body's own rationale ("the same as sending no effort, so it isn't a level") argues against publishing it as a level, and the unchanged comment at tinfoil.ts:96 ("the catalog's supported_efforts are the reasoning_effort values Tinfoil accepts") now contradicts the code. Action: Cite Tinfoil's reasoning docs (e.g. https://docs.tinfoil.sh/guides/reasoning.md, already cited by existing Tinfoil files) for what reasoning_effort: "auto" does and whether default is accepted; keep the mapping only with that evidence, updating the line 96 comment to state the auto→default rename (and noting the wire token for future hand-authored file headers), otherwise drop auto as the PR body describes. Also cover the mapping end-to-end through buildTinfoilModel (the new test at packages/core/test/sync.test.ts:8921 only asserts the parse transform, not the published reasoning_options).
  • [low][violation] packages/core/src/sync/providers/tinfoil.ts:30 - Check: The PR body must describe the actual change and cite the source for factual claims about host behavior. Why: The body describes the superseded first-commit behavior ("it isn't a level. I accept it in the schema and drop it from the list"), and its stated rationale directly contradicts the final code, which publishes auto as a default level; a maintainer reading only the body would believe unknown catalog efforts are dropped. Action: Update the PR body to describe the final mapping (auto → default level in the published effort list) and attach the Tinfoil documentation supporting it.

@github-actions

Copy link
Copy Markdown
Contributor

No actionable findings.

@github-actions github-actions Bot added the reviewer: ready Automated review found no actionable items label Nov 16, 2026
@rekram1-node
rekram1-node merged commit 511bd7f into dev Nov 16, 2026
2 checks passed
@rekram1-node
rekram1-node deleted the fix/tinfoil-auto-effort branch November 16, 2026 21:41

This branch was successfully deployed

1 active deployment
release — 339594c2 Deployed Nov 16, 2026 by rekram1-node via review #19269
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

reviewer: ready Automated review found no actionable items

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant