Add FLM multi-modal support (ASR + embeddings) with test refactor#1270
Merged
Add FLM multi-modal support (ASR + embeddings) with test refactor#1270
Conversation
…isper CI Add FLM multi-modal support (ASR + embeddings) in the C++ server and restructure the test capability system to support it. C++ implementation: - fastflowlm_server: add ASR transcription and embedding inference support - router: route audio transcription and embedding requests to FLM backend - server_models.json: add whisper-v3-turbo-FLM and embed-gemma-300m-FLM models Test refactor: - capabilities.py: modality-first CAPABILITIES dict with backward-compat flat alias - server_base.py: thread modality through parse_args/run_server_tests - server_whisper.py: use capability system for model lookup and skip decorators - server_llm.py, server_sd.py: pass modality to run_server_tests - CI: add flm-whisper matrix entry, explicit --wrapped-server whispercpp Co-Authored-By: Claude Opus 4.6 <[email protected]>
Use label-based checks (transcription, image, speech) consistently for all model types, matching the pattern already used for embeddings and reranking. This decouples the UI view routing from specific recipe names, so any recipe with the right labels gets the right view. Co-Authored-By: Claude Opus 4.6 <[email protected]>
The WebSocket realtime layer is backend-agnostic — it buffers audio and calls the same audio_transcriptions() method used by HTTP. FLM already implements IAudioServer so no code changes needed beyond flipping the capability flag. Verified with test_006 and test_007 passing. Co-Authored-By: Claude Opus 4.6 <[email protected]>
superm1
reviewed
Mar 2, 2026
Co-Authored-By: Claude Opus 4.6 <[email protected]>
WhisperCpp with NPU backend takes exclusive hold of the NPU, like RyzenAI. This ensures whispercpp evicts all NPU servers on load, FLM evicts whispercpp NPU servers before starting, and unknown NPU recipes default to evicting all rather than just one server. Co-Authored-By: Claude Opus 4.6 <[email protected]>
…nade into jfowers/flm-mm
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
3 participants
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
CAPABILITIES["llm"]["flm"]vsCAPABILITIES["whisper"]["flm"]), disambiguating backends that serve multiple modalitiesflm-whisperCI matrix entry and make existing whisper entries explicit with--wrapped-server whispercppChanges
C++ server:
fastflowlm_server.cpp/h: FLM ASR transcription and embedding inferencerouter.cpp/h: Route audio/embedding requests to FLM backendserver_models.json: Addwhisper-v3-turbo-FLMandembed-gemma-300m-FLMmodel entriesTest refactor:
capabilities.py: Modality-firstCAPABILITIESdict with backward-compat flat aliasserver_base.py: Threadmodalitythroughparse_args/run_server_tests; adddefault_wrapped_serverfor backward compatserver_whisper.py: Dynamic model viaget_test_model("audio"),@skip_if_unsupporteddecorators forrai_cache/realtime_websocketserver_llm.py,server_sd.py: Passmodality=torun_server_testsCI:
flm-whispermatrix entry on[rai300_400, Windows]--wrapped-server whispercppexplicitly🤖 Generated with Claude Code