Thanks to visit codestin.com
Credit goes to github.com

Skip to content

Fix cached models missing from chat picker - #7680

Open
ousamabenyounes wants to merge 1 commit into
unslothai:mainfrom
ousamabenyounes:fix/issue-7676
Open

Fix cached models missing from chat picker#7680
ousamabenyounes wants to merge 1 commit into
unslothai:mainfrom
ousamabenyounes:fix/issue-7676

Conversation

@ousamabenyounes

@ousamabenyounes ousamabenyounes commented Jul 31, 2026

Copy link
Copy Markdown
Contributor

Summary

  • include usable Hugging Face cache-only models in the chat model picker
  • preserve active repository IDs and inactive snapshot load paths
  • deduplicate only complete cache entries with the same repository and model format
  • make projected local models searchable through the existing local sections

Validation

  • RED → GREEN: removing hf_cache projection fails with the reported model absent; restoring it passes.
  • RED → GREEN: repository-only dedup removes the GGUF fallback beside cached Safetensors; repository-plus-format keeps it.
  • RED → GREEN: untrimmed repository IDs leak a duplicate; normalized IDs deduplicate it.
  • npm test (77/77 passing)
  • npm run typecheck
  • npm run build
  • ESLint and Biome checks for all changed files
  • 100% statements, branches, functions, and lines for the new inventory projector
  • Chromium verification of search, inactive load-path preservation, and duplicate suppression

Full local suite: passed.

Closes #7676

@ousamabenyounes

ousamabenyounes commented Aug 15, 2026

Copy link
Copy Markdown
Contributor Author

Rebased onto current main and resolved the conflicts in pickers.tsx and
use-chat-picker-inventory.ts.

The conflict mattered: main had since hardened the local-model filter with a
diffusion-task exemption (canChat || studioPageForTask(row.task)) and an
allowedHiddenModelIds bypass. This rework keeps that filter verbatim and
only layers the fix on top:

  • add hf_cache to PICKER_LOCAL_SOURCES;
  • drop an hf_cache row only when the same repo+format is already surfaced as a
    complete cached model (isHfCacheDuplicate, keyed on repoId\nmodelFormat;
    partial / live-downloading cached rows are intentionally excluded so the cache
    copy stays visible until the download finishes);
  • project hf_cache rows under the models_dir source. This is required, not
    cosmetic: the picker buckets localModels by source into exactly
    lmstudio / models_dir / custom (pickers.tsx), so an entry left as
    hf_cache would match no bucket and stay invisible — the very bug this fixes.

The reusable matching/dedup logic moved into a pure, unit-tested module
(chat-picker-inventory-sources.ts).

Verification (local):

  • chat-picker-inventory-sources.test.ts: 5/5 pass; mutation-checked (breaking
    the format in the dedup key fails a case, so the test is not vacuous).
  • full frontend test suite: 2707/2707 pass (no regression).
  • tsc typecheck: clean.
  • eslint: 0 new errors vs the main baseline (pre-existing pickers.tsx
    warnings unchanged).
  • Not run in this environment: a real browser pass on the picker screen.

chatgpt-codex-connector[bot]
chatgpt-codex-connector Bot reviewed Aug 16, 2026
View reviewed changes

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 71b482bb14

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

Comment thread studio/frontend/src/features/model-picker/inventory/chat-picker-inventory-sources.ts
return (
row.source === LOCAL_MODEL_SOURCE.HF_CACHE &&
row.repoId !== null &&
completeCachedKeys.has(inventoryKey(row.repoId, row.modelFormat))

@chatgpt-codex-connector chatgpt-codex-connector Bot Aug 16, 2026

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P2 Badge Normalize non-GGUF cache rows before deduplicating

When a complete Diffusers/video repository is returned by both inventory endpoints, its cached row is classified as safetensors, while _scan_hf_cache reports the non-GGUF local twin as unknown. Requiring the formats to match therefore retains both rows; the task exemption in useChatPickerInventory then admits the unknown local row, so the Images/Video picker displays the same snapshot once as cached and again under Local models. Normalize this HF-cache format before comparison, or otherwise treat this known unknown/safetensors pair as a duplicate while preserving genuinely distinct GGUF copies.

Useful? React with 👍 / 👎.

@danielhanchen

danielhanchen commented Aug 19, 2026

Copy link
Copy Markdown
Member

Confirmed this hits use-chat-picker-inventory.ts where PICKER_LOCAL_SOURCES still omits hf_cache, so a cache-only repo never reaches the local lists, and the rebase kept the canChat and hidden-id filter intact. Will get this reviewed.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@chatgpt-codex-connector chatgpt-codex-connector[bot] chatgpt-codex-connector[bot] left review comments

At least 1 approving review is required to merge this pull request.

Assignees

No one assigned

Labels

affects UI# edge case#

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

[Bug] Local model not being displayed in list of local models, but can be loaded through Search Hub

2 participants

@ousamabenyounes @danielhanchen