Slot config UX Phase 3: non-manual chat templates (model-level + per-slot override)#802
Open
thinmintdev wants to merge 12 commits into
Open
Slot config UX Phase 3: non-manual chat templates (model-level + per-slot override)#802thinmintdev wants to merge 12 commits into
thinmintdev wants to merge 12 commits into
Conversation
Per-slot mtp override in container flag-building + capability-gated MTP pill. 2 TDD tasks + a bench-retune issue. Bases on Phase 1 (#796). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add SlotConfig.mtp (bool | None) so a slot can force MTP on/off
independently of its profile. resolve_profile_flags gains an
mtp_override param (None = inherit profile). _profile_image_and_flags
accepts the override and recomputes flags from the profile's raw .flags
when set, bypassing any pre-expanded resolved_flags on ResolvedProfile.
container_spec passes slot_cfg.get("mtp") through.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Adds an MTP PillToggle inside the Inference FieldGroup (after Reasoning).
Renders only when slot.device starts with "gpu-rocm" AND the loaded model's
tags include "mtp". Toggle writes { mtp: next } via PUT /config then fires
a non-blocking restart (mirrors profile-change pattern). Tests C7i/C7j added
to slot-drawer-profile-v3.spec.ts; Page type re-exported from apiMock fixture.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…s persisted state
Model-level + slot-override chat templates, store-backed library, container --chat-template-file emission. 5 TDD tasks. Container load needs CT105 validation. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add chat_template to ModelDefaults and SlotConfig (slot override wins over model default; 'auto'/None both resolve to None for GGUF-embedded template). Add module-level resolve_chat_template() near resolve_profile_flags. 3 TDD tests green, ruff clean. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…eeding Adds GET/POST /api/chat-templates backed by <model_store_root>/chat-templates/, seeds chatml and llama3 bundled templates at startup (absent-only, non-fatal on read-only store), and validates custom template ids against a strict regex to prevent path traversal. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…template
Add `chat_template_path: str | None` param to `_llama_launch_plan`; when
set, inserts `["--chat-template-file", path]` into the command after
flag_tokens but before extra_tokens so a manual override in
[server].extra_args still wins on a later duplicate.
In `container_spec`, resolve the effective template id via
`resolve_chat_template(slot_cfg, model_info)` (slot override >
model defaults.chat_template > None/'auto') and compute the host path
as `Path(model_store_root()) / "chat-templates" / f"{tmpl_id}.jinja"`,
then pass it through to `_llama_launch_plan`. The model store is mounted
identical-path into the container so the host path == container path.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Add a chat_template <select> to the RecipeEditorModal seeded from
GET /api/chat-templates; the selected value is written into
defaults.chat_template on Save (PUT /api/models/{id}).
- NEW ui/src/api/hooks/useChatTemplates.ts — useQuery wrapper
- endpoints.ts — chatTemplates: '/api/chat-templates'
- model-modals.jsx — chatTemplate state + <select> + onSave write
- hooks/index.ts — re-export useChatTemplates
- NEW tests/e2e/specs/model-recipe-template-v3.spec.ts — 2 tests (red→green)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Part A: surface chat_template from TOML config in slot-view aggregator alongside mtp/enable_thinking (tri-valued: string/None/absent → null). Tests: TestConfigEnrichment.test_config_fields_surfaced + test_absent_config_fields_surface_as_defaults. Part B: Template row in the Model FieldGroup of EditSlotDrawer. - Read-only display shows model-level default (model.defaults.chat_template or "auto") with an [Override] button. - [Override] reveals a select fed by useChatTemplates() (auto + catalogue). - Override state seeds from slot.chat_template; starts in override mode when the slot has an existing override. - Save includes chat_template in slotBody only when changed (dirty-tracks against slot.chat_template, mirrors profileChanged pattern). - A chat_template change fires the non-blocking restart (mirrors MTP/profile). E2E: C7k added to slot-drawer-profile-v3.spec.ts. All gates: 49 pytest, 39 playwright (chromium), typecheck 0, build 0, ruff clean. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…-storm, Array.isArray guard The recipe modal is mounted while a model is merely selected, so an ungated useChatTemplates fired its (unmocked-in-test / failing) fetch on the models list and detached row controls. And when the endpoint returns a non-array, the '|| []' / '?? []' guards let a non-array through → '.filter is not a function' crashed the modal. Fixes: enabled:open gating, retry:false, and Array.isArray guards at both the recipe-editor and slot-drawer template selects. Restores the models-v3 recipe + delete-cascade e2e (full suite 239 passed).
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.
Phase 3 (final) of the slot-config UX redesign (spec/plan in
docs/superpowers/). Builds on Phase 1 (#796) + Phase 2 (#800).What changes — no more hand-editing TOML for chat templates
chatml,llama3) and seed<model-store>/chat-templates/on startup (absent-only, skipped silently on a read-only store).GET /api/chat-templateslists them +auto;POSTaccepts a custom paste (id-validated).defaults.chat_template.chat_template(slot wins). Fires the non-blocking restart.resolve_chat_template(slot_cfg, model_info)picks the effective id (slot > model > none=auto); the container provider emits--chat-template-file <store>/chat-templates/<id>.jinja. The store is mounted identical-path read-only into the container, so no path translation.auto/none → no flag (GGUF-embedded template).chat_templatein the slot-list payload so the override seeds from disk.The wiring is fully unit/e2e tested here, but this VM has no GPU/container runtime — that llama-server in the container actually loads the mounted
--chat-template-filemust be validated on CT105 before relying on it.Verification
test_chat_template_resolve+test_chat_templates(catalog, seeding, POST validation incl. path-traversal reject) +test_container_chat_template+ existing container (40) + slot_view (with newchat_templateassert) — all green. ruff check + format clean.npm run typecheck0 ·npm run build0 · full Playwright e2e 239 passed / 6 skipped / 0 failed (new recipe-template + slot-override specs).useChatTemplatesfetched on the models list and a non-array response slipped past|| []→ crashed the modal. Fixed withenabled:opengating,retry:false, andArray.isArrayguards.Series complete
Phases 1–3 of the slot-config UX redesign are now all up. Follow-up #799 (MTP bundle bench-retune) remains.
🤖 Generated with Claude Code