fix: HF vision mmproj sidecars (.mmproj) + remove backend-switch surface (WS-5)#1089
Merged
Merged
Conversation
The Add-model-by-HF inspect endpoint only admitted files ending in `.gguf`, so repos that ship the projector with a bare `.mmproj` extension (e.g. `mmproj-*-F32.mmproj`) had it dropped from the variant list. The Add-by-HF modal builds its vision picker from those variants, so it showed "no mmproj files in repo" and a vision pull silently shipped without a projector. Inspect now also admits `*.mmproj` sidecars (guarded on the "mmproj" token so nothing unrelated leaks into the main quant list), labelled "mmproj sidecar". This matches the local-directory scanner (`discover._is_mmproj_sidecar`), which already matches mmproj by name regardless of extension. No UI change needed — the modal already routes `/mmproj/i` variants into the vision dropdown and out of the main list. Co-Authored-By: Claude Opus 4.8 <[email protected]>
Backend identity is owned by the slot's profile, so a dedicated
backend-switch was a second way to express what profiles already own.
Removes `POST /api/slots/{name}/backend` and its two endpoint-only
helpers, the `useSlotBackend` hook + `SlotBackendSwitchResponse` type +
`slotBackend` endpoint constant, and the switch UX in `slots.jsx` /
`inference-pane.jsx`.
The amber `backend_mismatch` chip is KEPT (detection is unchanged,
sourced from the slot_view aggregator) but now opens the slot editor's
profile picker via `onEdit` instead of firing a one-click switch — the
sanctioned way to change a slot's backend.
Deletes the dedicated endpoint test file, drops
`test_backend_flip_reconciles_profile_via_route` (the device->profile
reconciliation it covered still lives in `update_config`), and removes
the two `/backend` typed-error params.
Co-Authored-By: Claude Opus 4.8 <[email protected]>
FO-1 (WS-4 straggler): `cli/slot_commands.py` hardcoded `range(8081, 8100)` for its free-port scan, diverging from the API's `_next_free_slot_port`. Point it at the shared `SLOT_PORT_RANGE_START/END` constants in `hardware.stats` (port scan + help string) so the two pools can never drift. FO-2: correct a stale `container.py` module docstring that still claimed `ContainerProvider` reuses an inherited `render_systemd_override` — that default was deleted in WS-15; unit rendering is owned by `_render_unit_from_plan`. Also records the finish-out resolutions (FO-1..FO-4) in the handoff doc. Co-Authored-By: Claude Opus 4.8 <[email protected]>
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 logically separate changes, one commit each.
1.
fix(models)— HF vision.mmprojsidecars (bug)The Add-model-by-HF inspect endpoint only surfaced
.gguffiles, so repos that ship the projector with a bare.mmprojextension (e.g.jcbtc/chadrock-35b-ace-saber-rocmfp4-mtp→mmproj-CHADROCK-35B-Ace-Saber-F32.mmproj) had it dropped from the variant list. The Add-by-HF modal builds its vision picker from those variants, so it showed "no mmproj files in repo" and a vision pull silently shipped without a projector.Inspect now also admits
*.mmprojsidecars (guarded on themmprojtoken so nothing unrelated leaks into the main quant list), matching the local-directory scanner (discover._is_mmproj_sidecar) which already matches by name regardless of extension. No UI change needed. Two regression tests added (one uses the exact repo layout above).2.
refactor(slots)— remove the backend-switch surface (WS-5)Backend identity is owned by the slot's profile, so a dedicated backend-switch was a second way to express what profiles already own. Removes
POST /api/slots/{name}/backend+ its endpoint-only helpers, theuseSlotBackendhook/type +slotBackendendpoint const, and the switch UX inslots.jsx/inference-pane.jsx.The amber
backend_mismatchchip is kept (detection unchanged, from the slot_view aggregator) but now opens the slot editor's profile picker instead of firing a switch. Deletes the dedicated endpoint test file and trims the corresponding route/typed-error cases.3.
chore(slots)— FO-1/FO-2 cleanupSLOT_PORT_RANGE_START/ENDconstants instead of a hardcodedrange(8081, 8100)(was diverging from the API pool).container.pydocstring referencing therender_systemd_overridedefault deleted in WS-15.Verification
tests/api+tests/slotssuites pass; 1527 tests collect with no import breakage.tsctypecheck clean,vite buildsucceeds,.mjsdash tests pass.🤖 Generated with Claude Code