Qwen3-ASR-0.6B encodes the jfk.wav fixture into 777 audio tokens via
its mmproj, but the test harness defaulted BACKEND_TEST_CTX_SIZE to
512, so llama.cpp server rejected every transcription request with
"request (777 tokens) exceeds the available context size (512 tokens)".
Set BACKEND_TEST_CTX_SIZE=2048 on the llama-cpp transcription target
only — sherpa-onnx and vibevoice transcription targets don't go
through llama.cpp's slot/n_ctx and weren't failing.
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Assisted-by: Claude:claude-opus-4-7 [Claude Code]
streamTranscription previously emitted a done event with just `text`,
matching the OpenAI streaming spec exactly. Streaming clients that need
per-utterance timings or audio duration had to fall back to the
non-streaming JSON path — and that path is exactly the one that trips
on ResponseHeaderTimeout when whisper requests queue behind each other
on a SingleThread backend.
Extend the done event to additively carry `language`, `duration`, and
a `segments` array (id, start, end, text — start/end as float seconds,
matching TranscriptionSegmentSeconds). Empty / zero values are still
omitted; spec-compliant clients ignore the new fields.
This unblocks notary's streaming Transcribe (companion change in the
notary repo) so it produces the same TranscriptionResult shape as the
JSON path while sidestepping the queue-induced header timeouts.
Assisted-by: Claude:claude-opus-4-7 [Claude Code]
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Co-authored-by: Ettore Di Giacinto <mudler@localai.io>
* fix(distributed): make backend upgrade actually re-install on workers
UpgradeBackend dispatched a vanilla backend.install NATS event to every
node hosting the backend. The worker's installBackend short-circuits on
"already running for this (model, replica) slot" and returns the
existing address — so the gallery install path was skipped, no artifact
was re-downloaded, no metadata was written. The frontend's drift
detection then re-flagged the same backends every cycle (installedDigest
stays empty → mismatch → "Backend upgrade available (new build)") while
"Backend upgraded successfully" landed in the logs at the same time.
The user-visible symptom: clicking "Upgrade All" silently does nothing
and the same N backends sit on the upgrade list forever.
Two coupled fixes, one PR:
1. Force flag on backend.install. Add `Force bool` to
BackendInstallRequest and thread it through NodeCommandSender ->
RemoteUnloaderAdapter. UpgradeBackend (and the reconciler's pending-op
drain when retrying an upgrade) sets force=true; routine load events
and admin install endpoints keep force=false. On the worker, force=true
stops every live process that uses this backend (resolveProcessKeys
for peer replicas, plus the exact request processKey), skips the
findBackend short-circuit, and passes force=true into
gallery.InstallBackendFromGallery so the on-disk artifact is
overwritten. After the gallery install completes, startBackend brings
up a fresh process at the same processKey on a new port.
2. Liveness check on the fast path. installBackend's "already running"
branch read getAddr without verifying the process was alive, so a
gRPC backend that died without the supervisor noticing left a stale
(key, addr) entry. The reconciler then dialed that address, got
ECONNREFUSED, marked the replica failed, retried install — and the
supervisor said "already running addr=…" again. Loop forever, exactly
what we observed on a node whose llama-cpp process had died but whose
supervisor record persisted. Verify s.isRunning(processKey) before
trusting getAddr; if the entry is stale, stopBackendExact cleans up
and we fall through to a real install.
Backwards-compatible: the new Force field is omitempty, older workers
ignore it (their default behavior matches force=false). The signature
change on NodeCommandSender.InstallBackend is internal-only.
Verified: unit tests in core/services/nodes pass (108s suite). The
pre-existing core/backend build break (proto regen pending for
word-level timestamps) blocks core/cli and core/http/endpoints/localai
package tests but is unrelated to this change.
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Assisted-by: Claude:claude-opus-4-7 [Claude Code]
* test(e2e/distributed): pass force=false to adapter.InstallBackend
NodeCommandSender.InstallBackend gained a final force bool in the
upgrade-force commit; the e2e distributed lifecycle tests still called
the old 8-arg signature and broke compilation. These tests exercise the
routine install path (single replica, default behavior), so force=false
preserves their existing semantics.
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Assisted-by: Claude:claude-opus-4-7 [Claude Code]
---------
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Co-authored-by: Ettore Di Giacinto <mudler@localai.io>
* fix(http): log accurate status code when handler returns error
The custom xlog access-log middleware in API() reads res.Status
*before* Echo's central HTTPErrorHandler runs, so when a handler
returns an error without writing a response (e.g.
TranscriptEndpoint's `return err` on backend failure) the status
field stays at its default 200. The logged line then claims
status=200 while the client receives 500 — silently hiding every
500/503/etc. that bubbles up through Echo's error handler.
Mirror echo.DefaultHTTPErrorHandler's status derivation when
err != nil and the response hasn't been committed: default to 500,
upgrade to *echo.HTTPError.Code if applicable. The logged status now
matches what the client actually sees, so failed transcription
requests stop appearing as 200 in the access log.
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Assisted-by: Claude:claude-opus-4-7 [Claude Code]
* fix(transcription): log underlying error before returning 500 to client
ModelTranscriptionWithOptions surfaces real failures — gRPC errors
from a remote node, model load problems, ffmpeg conversion crashes —
but TranscriptEndpoint just did `return err`, so Echo turned it into
a 500 with a generic body and the original error was lost. Operators
chasing transcription failures across distributed mode were left
with "upstream returned 500" on the client and zero context anywhere
in the frontend's logs.
Add an xlog.Error before returning, recording model name, the staged
audio path, and the underlying error. Combined with the access-log
status fix, a failing transcription now leaves an audit trail (real
status code in the access line, real cause in an Error line) instead
of vanishing.
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Assisted-by: Claude:claude-opus-4-7 [Claude Code]
---------
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Co-authored-by: Ettore Di Giacinto <mudler@localai.io>
Bring the sglang Python backend up to feature parity with vllm by adding
the same engine_args:-map plumbing the vLLM backend already has. Any
ServerArgs field (~380 in sglang 0.5.11) becomes settable from a model
YAML, including the speculative-decoding flags needed for Multi-Token
Prediction. Validation matches the vllm backend's: keys are checked
against dataclasses.fields(ServerArgs), unknown keys raise ValueError
with a difflib close-match suggestion at LoadModel time, and the typed
ModelOptions fields keep their existing meaning with engine_args
overriding them.
Backend code:
* backend/python/sglang/backend.py: add _apply_engine_args, import
dataclasses/difflib/ServerArgs, call from LoadModel; rename Seed ->
sampling_seed (sglang 0.5.11 renamed the SamplingParams field).
* backend/python/sglang/test.py + test.sh + Makefile: six unit tests
exercising the helper directly (no engine load required).
Build / CI / backend gallery (cuda13 + l4t13 paths are now first-class):
* backend/python/sglang/install.sh: add --prerelease=allow because
sglang 0.5.11 hard-pins flash-attn-4 which only ships beta wheels;
add --index-strategy=unsafe-best-match for cublas12 so the cu128
torch index wins over default-PyPI's cu130; new pyproject.toml-driven
l4t13 install path so [tool.uv.sources] can pin torch/torchvision/
torchaudio/sglang to the jetson-ai-lab index without forcing every
transitive PyPI dep through the L4T mirror's flaky proxy (mirrors the
equivalent fix in backend/python/vllm/install.sh).
* backend/python/sglang/pyproject.toml (new): L4T project spec with
explicit-source jetson-ai-lab index. Replaces requirements-l4t13.txt
for the l4t13 BUILD_PROFILE; other profiles still go through the
requirements-*.txt pipeline via libbackend.sh's installRequirements.
* backend/python/sglang/requirements-l4t13.txt: removed; superseded
by pyproject.toml.
* backend/python/sglang/requirements-cublas{12,13}{,-after}.txt: pin
sglang>=0.5.11 (Gemma 4 floor); add cu130 torch index for cublas13
(new files) and cu128 torch index for cublas12 (default PyPI now
ships cu130 torch wheels by default and breaks cu12 hosts).
* backend/index.yaml: add cuda13-sglang and cuda13-sglang-development
capability mappings + image entries pointing at
quay.io/.../-gpu-nvidia-cuda-13-sglang.
* .github/workflows/backend.yml: new cublas13 sglang matrix entry,
mirroring vllm's cuda13 build.
Model gallery + docs:
* gallery/sglang.yaml: base sglang config template, mirrors vllm.yaml.
* gallery/sglang-gemma-4-{e2b,e4b}-mtp.yaml: Gemma 4 MTP demos
transcribed verbatim from the SGLang Gemma 4 cookbook MTP commands.
* gallery/sglang-mimo-7b-mtp.yaml: MiMo-7B-RL with built-in MTP heads
+ online fp8 weight quantization, verified end-to-end on a 16 GB
RTX 5070 Ti at ~88 tok/s. Uses mem_fraction_static: 0.7 because the
MTP draft worker's vocab embedding is loaded unquantised and OOMs
the static reservation at sglang's 0.85 default.
* gallery/index.yaml: three new entries (gemma-4-e2b-it:sglang-mtp,
gemma-4-e4b-it:sglang-mtp, mimo-7b-mtp:sglang).
* docs/content/features/text-generation.md: new SGLang section with
setup, engine_args reference, MTP demos, version requirements.
* .agents/sglang-backend.md (new): agent one-pager covering the flat
ServerArgs structure, the typed-vs-engine_args precedence, the
speculative-decoding cheatsheet, and the mem_fraction_static gotcha
documented above.
* AGENTS.md: index entry for the new agent doc.
Known limitation: the two Gemma 4 MTP gallery entries ship a recipe
that doesn't yet run on stock libraries. The drafter checkpoints
(google/gemma-4-{E2B,E4B}-it-assistant) declare
model_type: gemma4_assistant / Gemma4AssistantForCausalLM, which
neither transformers (<=5.6.0, including the SGLang cookbook's pinned
commit 91b1ab1f... and main HEAD) nor sglang's own model registry
(<=0.5.11) registers as of 2026-05-06. They will start working when
HF or sglang upstream registers the architecture -- no LocalAI
changes needed. The MiMo MTP demo and the non-MTP Gemma 4 paths work
today on this build (verified on RTX 5070 Ti, 16 GB).
Assisted-by: Claude:claude-opus-4-7 [Read] [Edit] [Bash] [WebFetch] [WebSearch]
Signed-off-by: Richard Palethorpe <io@richiejp.com>
The previous omegaconf pin only addressed one symptom of a deeper problem:
chatterbox-tts upstream depends on `russian-text-stresser` (unpinned git URL),
which transitively pins `spacy==3.6.*` and other ancient packages. That cascade
forces pip to backtrack through Jinja2/MarkupSafe/omegaconf into Python-2-era
sdists that no longer build (e.g. ruamel.yaml<0.15, Jinja2 2.6 importing the
long-removed `setuptools.Feature`).
Install chatterbox-tts itself with --no-deps in install.sh and list its real
runtime deps explicitly in each requirements-*.txt, dropping the optional
russian-text-stresser. This unblocks the darwin (and other) builds without
playing whack-a-mole on each newly-discovered transitive pin.
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Assisted-by: Claude:claude-opus-4-7 [Claude Code]
LocalRecall switched its PDF extractor from gen2brain/go-fitz (cgo +
static libmupdf) to klippa-app/go-pdfium with the WebAssembly backend
(no cgo, no static libs, no glibc symbol issues). This bump pulls in
the new shape via LocalAGI@main → LocalRecall@a7724fe.
Why this matters: the previous go-fitz approach broke aarch64 LocalAI
builds with `__isoc23_strtol undefined reference` link errors —
go-fitz's bundled libmupdf static library is compiled against
glibc ≥ 2.38 (C2x strtol intrinsics) but the LocalAI builder uses
older glibc. The WASM backend has no native dependencies; same Go
binary works on every architecture.
Indirect graph: drops gen2brain/go-fitz, ebitengine/purego,
jupiterrider/ffi; adds klippa-app/go-pdfium + tetratelabs/wazero.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The previous pin in requirements.txt was ineffective: installRequirements
runs a separate `pip install --requirement` per file, so resolution does
not carry over to the per-profile file where chatterbox-tts is declared.
With chatterbox-tts's unpinned `omegaconf` dep, pip backtracked through
1.x sdists into ruamel.yaml<0.15, whose Python-2-era setup.py fails on
Python 3.10+.
Pin omegaconf==2.3.0 next to chatterbox-tts in every profile file
(matches what upstream chatterbox uses). Drop the dead pin from
requirements.txt.
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Assisted-by: Claude:claude-opus-4-7 [Claude Code]
* fix(auth): cascade user deletion across all owned data on PostgreSQL
Deleting a user from the admin UI in distributed mode (PostgreSQL auth
DB) returned "user not found" even when the user clearly existed. The
old handler ignored result.Error and only checked RowsAffected, so a
foreign-key constraint violation surfaced as a misleading 404.
Two issues drove this:
1. invite_codes.created_by / used_by reference users(id) but the
InviteCode model declared the FKs without ON DELETE CASCADE. On
PostgreSQL the engine therefore rejected the user delete with NO
ACTION whenever the user had ever issued or consumed an invite. On
SQLite (default in single-node mode) FKs are not enforced, so the
bug never appeared there.
2. Several owned tables were never cleaned up regardless of dialect:
user_permissions and quota_rules relied on CASCADE that does not
fire under SQLite, and usage_records have no FK at all and were
left orphaned in every dialect.
Introduce auth.DeleteUserCascade which runs the full cleanup in a
single transaction: drop invites authored by the user, NULL used_by on
invites they consumed (preserves the audit trail), and explicitly wipe
sessions, API keys, permissions, quota rules, and usage metrics before
deleting the user. The in-memory quota cache is invalidated after
commit so a recreated user with the same id never sees stale entries.
The HTTP handler now maps the helper's errors to proper status codes —
real failures surface as 500 with the cause instead of being swallowed
as "not found".
Add Ginkgo regression coverage in core/http/auth/users_test.go and
core/http/routes/auth_test.go covering invite cleanup, used_by
null-out, full data wipe, and the FK-enforced original failure mode
(via PRAGMA foreign_keys=ON to mirror PostgreSQL behavior on SQLite).
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Assisted-by: Claude:claude-opus-4-7 [Claude Code]
* chore(deps): bump LocalAGI/LocalRecall — pull in go-fitz PDF extraction
Pulls LocalAGI@main (facd888) and LocalRecall@v0.6.0. The latter
swaps PDF text extraction from dslipak/pdf to gen2brain/go-fitz
(libmupdf bindings) and wraps it in a 60s goroutine timeout —
previously certain PDFs (broken xref tables, encrypted, image-only
without OCR) would hang indefinitely inside r.GetPlainText() and
poison the upload queue.
Pure dep bump, no LocalAI source changes. Indirect graph picks up
go-fitz + purego + ffi; drops dslipak/pdf.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---------
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Co-authored-by: Ettore Di Giacinto <mudler@localai.io>
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(downloader): list supported URL schemes when input is unrecognized
The error message previously read "does not look like an HTTP URL",
but the downloader actually supports file://, huggingface://, hf://,
ollama://, oci://, and github:// in addition to http(s)://. Users who
type a bare filename or a typo'd scheme (e.g. fle:// instead of file://)
get the misleading impression that only HTTP is accepted.
Reference the existing prefix constants directly via strings.Join so
the scheme list cannot drift when new prefixes are added.
Refs #9683.
Signed-off-by: Tai An <antai12232931@outlook.com>
* fix(downloader): normalize uri.go to LF line endings
Resolves the noisy diff and golangci-lint errcheck warnings on lines I did not actually modify.
* fix(downloader): preserve trailing newline at end of file
---------
Signed-off-by: Tai An <antai12232931@outlook.com>
Without an upper-floor pin, pip's resolver backtracks through omegaconf 1.x
sdists when installing chatterbox-tts. Old 1.x setups depend on
ruamel.yaml<0.15, whose setup.py uses Python-2-era names (Str, Bytes) and
fails to build on Python 3.10+, breaking the darwin python backend build.
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Assisted-by: Claude:claude-opus-4-7 [Claude Code]
FindAndLockNodeWithModel previously ordered candidate replicas by
in_flight ASC, available_vram DESC. The primary key is correct, but the
tiebreaker meant that whenever in_flight tied — the common case at low
to moderate concurrency where requests don't overlap — the node with
the largest available VRAM won every pick. With autoscaling placing
replicas of the same model on multiple nodes, the fattest GPU node
ended up taking nearly all the load while the others sat idle.
Insert last_used ASC between the two existing tiers. last_used is
already refreshed inside the same transaction that increments in_flight
(and by TouchNodeModel on cache hits in the router), so the
"oldest-used" replica naturally rotates through the candidate set —
strict round-robin without a schema change. available_vram DESC is
demoted to a final tiebreaker for cold starts where last_used is
identical across replicas.
Placement queries (FindNodeWithVRAM, FindLeastLoadedNode, and the
*FromSet variants) have the same fattest-GPU bias on tiebreakers but
are higher-cost to fix consistently. Deferred to a follow-up so the
routing fix can land first — for the user-observed symptom routing was
the dominant cause anyway.
Test: registry_test.go adds a focused spec that loads three replicas
on three nodes with 24/16/8 GB VRAM and asserts each is picked at
least twice across 9 in_flight-tied calls.
Assisted-by: claude-code:claude-opus-4-7 [Read] [Edit] [Bash] [Grep]
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Co-authored-by: Ettore Di Giacinto <mudler@localai.io>
Two unrelated CI breakages bundled together since both are one-liners:
- rerankers: bump torch 2.4.1 -> 2.7.1 on cpu/cublas12. The unpinned
transformers resolves to 5.x, whose moe.py registers a custom_op with
string-typed `'torch.Tensor'` annotations that torch 2.4.1's
infer_schema rejects, blocking the gRPC server from starting and
failing all 5 backend tests with "Connection refused" on :50051.
Matches the version used by the transformers backend.
- vllm-omni: strip fa3-fwd from the upstream requirements/cuda.txt
before resolving on aarch64. fa3-fwd 0.0.3 ships only an
x86_64 wheel and has no sdist, making the cuda profile unsatisfiable
on Jetson/SBSA. fa3-fwd is a soft runtime dep — vllm-omni's
attention backends fall back to FA2 then SDPA when it's missing.
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Co-authored-by: Ettore Di Giacinto <mudler@localai.io>
* feat: Rework VRAM estimation and use known_usecases in gallery
Signed-off-by: Richard Palethorpe <io@richiejp.com>
Assisted-by: Claude:claude-opus-4-7[1m] [Claude Code]
* chore(gallery): regenerate gallery index and add known_usecases to model entries
Signed-off-by: Richard Palethorpe <io@richiejp.com>
---------
Signed-off-by: Richard Palethorpe <io@richiejp.com>
Routes mudler/vibevoice.cpp-models and similar repos to the vibevoice-cpp
backend. Detects via repo name ("vibevoice.cpp"/"vibevoice-cpp"), file
listing (vibevoice-*.gguf + tokenizer.gguf), or preferences.backend
override. Defaults to the realtime TTS model; preferences.usecase=asr
selects the ASR/diarization variant. Bundles the required tokenizer.gguf
and (for TTS) a voice prompt, emitting the Options[] entries the backend
expects. Registered ahead of VibeVoiceImporter so the C++ bundles aren't
swallowed by the older Python-backend substring match.
Assisted-by: claude-code:claude-opus-4-7 [Read] [Edit] [Write] [Bash]
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Co-authored-by: Ettore Di Giacinto <mudler@localai.io>
* fix(docs): correct broken Hugo relrefs
The Hugo build has been failing on master since the relevant pages
landed:
- text-generation.md:720 referenced `/docs/features/distributed-mode`,
but Hugo `relref` paths are relative to the content root, not the
rendered URL. Drop the `/docs/` prefix so the lookup matches the
existing `features/...` form used elsewhere in the file.
- audio-transform.md:144 referenced `tts.md`; the actual page is
`text-to-audio.md`.
Assisted-by: Claude:claude-opus-4-7[1m]
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* fix(kokoros): stub Diarize and AudioTransform Backend trait methods
The recent backend.proto additions (Diarize, AudioTransform,
AudioTransformStream) extended the gRPC Backend trait, breaking
kokoros-grpc compilation with E0046 because the Rust implementation
hadn't picked up the new methods. Add Unimplemented stubs matching the
existing pattern for non-applicable RPCs in this TTS-only backend.
Assisted-by: Claude:claude-opus-4-7[1m]
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* fix(vibevoice-cpp): track upstream ABI + wire 1.5B voice cloning
Two recent commits in mudler/vibevoice.cpp reshaped the vv_capi_tts
signature without a corresponding bump on the LocalAI side:
3bd759c "1.5b: unify into a single tts entry point" inserted a
ref_audio_path parameter between voice_path and dst_wav_path.
ad856bd "1.5b: multi-speaker dialog support" promoted that to a
(const char* const* ref_audio_paths, int n_ref_audio_paths)
pair for per-speaker conditioning.
Because purego resolves symbols by name and not by signature, the
build kept linking; at runtime the misaligned arguments turned the
TTS->ASR closed-loop test into a SIGSEGV inside cgo. Track HEAD
explicitly and bring the bridge in line with it:
* Update the CppTTS purego binding to the 9-arg form. purego
marshals []*byte as a **char by handing the C side the underlying
array address; nil/empty maps to NULL, which matches the C
contract for "no reference audio" on the realtime-0.5B path.
* Add a `ref_audio` gallery option (comma-separated, repeatable)
that the 1.5B path consumes for runtime voice cloning. Multiple
entries are interpreted as one WAV per speaker (Speaker 0..n-1).
* TTSRequest.Voice now routes by extension/shape: `.wav` or a
comma-separated list goes to ref_audio_paths; anything else stays
on voice_path (realtime-0.5B's pre-baked voice gguf).
* Pin VIBEVOICE_CPP_VERSION to ad856bd and wire the Makefile into
the existing bump_deps matrix so future upstream rolls land as
reviewable PRs instead of a silent CI break.
Assisted-by: Claude:claude-opus-4-7[1m]
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* refactor(vibevoice-cpp): use ModelOptions.AudioPath for 1.5B ref audio
Use the existing audio_path field from ModelOptions (already plumbed
through config_file's `audio_path:` YAML and consumed by other audio
backends like kokoros) instead of inventing a custom `ref_audio:`
Options[] string. Multi-speaker setups stay on a single comma-
separated value.
No behavior change beyond the gallery key name; per-call routing via
TTSRequest.Voice is unchanged.
Assisted-by: Claude:claude-opus-4-7[1m]
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
---------
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Co-authored-by: Ettore Di Giacinto <mudler@localai.io>
* fix(backend): resolve relative draft_model paths against the models dir
The main model file and mmproj are joined with the configured models
directory before reaching the backend, but draft_model was sent
verbatim. With a relative draft_model in the YAML config, llama.cpp
opens the path from the backend process's CWD and fails with "No such
file or directory", forcing users to hard-code an absolute path.
Mirror the existing mmproj resolution: if draft_model is relative,
join it with modelPath. Absolute paths are passed through unchanged.
Adds an e2e regression test against the mock backend that asserts the
main model file, mmproj, and draft_model all arrive at the backend
resolved to absolute paths.
Closes#9675
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Assisted-by: Claude:claude-opus-4-7-1m [Read] [Edit] [Bash] [Write]
* fix(backend): always join draft_model with models dir (drop IsAbs shortcut)
The previous commit kept absolute draft_model paths intact via an
IsAbs check. That left a path-traversal vector open: a user-supplied
YAML config could set draft_model to /etc/passwd (or any other host
file the backend process can read) and the path would be sent through
unchanged.
filepath.Join cleans the leading slash from absolute components, so
joining unconditionally — the way mmproj already does — keeps the
result rooted at the configured models directory regardless of input.
Adds a second e2e spec that feeds an absolute draft_model into the
mock backend and asserts the path is clamped under modelsPath.
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Assisted-by: Claude:claude-opus-4-7-1m [Read] [Edit] [Bash]
---------
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Co-authored-by: Ettore Di Giacinto <mudler@localai.io>
In distributed mode the React UI's "Upgrade All" button fanned every
detected outdated backend out to every healthy backend node, including
nodes that never had that backend installed. On heterogeneous clusters
this surfaced as platform errors (e.g. mac-mini-m4 asked to upgrade
cpu-insightface-development, which has no darwin/arm64 variant) and left
forever-retrying pending_backend_ops rows.
DistributedBackendManager.UpgradeBackend now queries ListBackends()
first, builds the target node-ID set from SystemBackend.Nodes, and only
fans out to those nodes — every per-node primitive
(adapter.InstallBackend, the pending-ops queue, BackendOpResult) is
unchanged. enqueueAndDrainBackendOp gains an optional targetNodeIDs
allowlist; Install/Delete keep their fan-to-everyone semantics by
passing nil. If no node reports the backend installed, UpgradeBackend
now returns a clear "not installed on any node" error instead of
producing a stuck queue.
Adds Ginkgo coverage for the smart fan-out: backend on a subset of
nodes goes only to those nodes; backend on no node returns the new
error and never sends a NATS install request.
Assisted-by: Claude:claude-opus-4-7 [Claude Code]
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Two related runtime fixes for Python backends that JIT-compile CUDA
kernels at first model load (FlashInfer, PyTorch inductor, triton):
1. libbackend.sh: replace `source ${EDIR}/venv/bin/activate` with a
minimal manual setup (_activateVenv: export VIRTUAL_ENV, prepend
PATH, unset PYTHONHOME) computed from $EDIR at runtime. `uv venv`
and `python -m venv` both bake the create-time absolute path into
bin/activate (e.g. VIRTUAL_ENV='/vllm/venv' from the Docker build
stage), so sourcing activate on a relocated venv — copied out of
the build container and unpacked at an arbitrary backend dir —
prepends a stale, non-existent path to $PATH. Pip-installed CLI
tools (e.g. ninja, used by FlashInfer's NVFP4 GEMM JIT) are then
never found and the load aborts with FileNotFoundError. Doing the
env setup ourselves matches what `uv run` does internally and
sidesteps the relocation problem entirely. Generic — every Python
backend benefits.
2. vllm/run.sh: replace ninja's default -j$(nproc)+2 with an adaptive
MAX_JOBS = min(nproc, (MemAvailable-4)/4). Each concurrent
nvcc/cudafe++ peaks at multiple GiB; the default OOM-kills on
memory-tight hosts (e.g. a 16 GiB desktop loading a 27B NVFP4
model) but underutilises 100-core / 1 TB boxes. User-set MAX_JOBS
still wins. Also pin NVCC_THREADS=2 unless overridden.
Refs: https://github.com/vllm-project/vllm/issues/20079
Assisted-by: Claude:claude-opus-4-7 [Edit] [Bash]
* feat(vllm): build vllm from source for Intel XPU
Upstream publishes no XPU wheels for vllm. The Intel profile was
silently picking up a non-XPU wheel that imported but errored at
engine init, and several runtime deps (pillow, charset-normalizer,
chardet) were missing on Intel -- backend.py crashed at import time
before the gRPC server came up.
Switch the Intel profile to upstream's documented from-source
procedure (docs/getting_started/installation/gpu.xpu.inc.md in
vllm-project/vllm):
- Bump portable Python to 3.12 -- vllm-xpu-kernels ships only a
cp312 wheel.
- Source /opt/intel/oneapi/setvars.sh so vllm's CMake build sees
the dpcpp/sycl compiler from the oneapi-basekit base image.
- Hide requirements-intel-after.txt during installRequirements
(it used to 'pip install vllm'); install vllm's deps from a
fresh git clone of vllm via 'uv pip install -r
requirements/xpu.txt', swap stock triton for
triton-xpu==3.7.0, then 'VLLM_TARGET_DEVICE=xpu uv pip install
--no-deps .'.
- requirements-intel.txt trimmed to LocalAI's direct deps
(accelerate / transformers / bitsandbytes); torch-xpu, vllm,
vllm_xpu_kernels and the rest come from upstream's xpu.txt
during the source build.
- requirements.txt: add pillow + charset-normalizer + chardet --
used by backend.py and missing on the Intel install profile.
- run.sh: 'set -x' so backend startup is visible in container
logs (the gRPC startup error path was previously opaque).
Also adds a one-line docs example for engine_args.attention_backend
under the vLLM section, since older XE-HPG GPUs (e.g. Arc A770)
need TRITON_ATTN to bypass the cutlass path in vllm_xpu_kernels.
Tested end-to-end on an Intel Arc A770 with Qwen2.5-0.5B-Instruct
via LocalAI's /v1/chat/completions.
Assisted-by: Claude:claude-opus-4-7 [Claude Code]
Signed-off-by: Richard Palethorpe <io@richiejp.com>
* feat(vllm): add multi-node data-parallel follower worker
vLLM v1's multi-node story is one process per node sharing a DP
coordinator over ZMQ -- the head runs the API server with
data_parallel_size > 1 and followers run `vllm serve --headless ...`
with matching topology. Today LocalAI can already configure DP on the
head via the engine_args YAML map, but there's no way to bring up the
follower nodes -- so the head sits waiting for ranks that never
handshake.
Add `local-ai p2p-worker vllm`, mirroring MLXDistributed's structural
precedent (operator-launched, static config, no NATS placement). The
worker:
- Optionally self-registers with the frontend as an agent-type node
tagged `node.role=vllm-follower` so it's visible in the admin UI
and operators can scope ordinary models away via inverse
selectors.
- Resolves the platform-specific vllm backend via the gallery's
"vllm" meta-entry (cuda*, intel-vllm, rocm-vllm, ...).
- Runs vLLM as a child process so the heartbeat goroutine survives
until vLLM exits; forwards SIGINT/SIGTERM so vLLM can clean up its
ZMQ sockets before we tear down.
- Validates --headless + --start-rank 0 is rejected (rank 0 is the
head and must serve the API).
Backend run.sh dispatches `serve` as the first arg to vllm's own CLI
instead of LocalAI's backend.py gRPC server -- the follower speaks
ZMQ directly to the head, there is no LocalAI gRPC on the follower
side. Single-node usage is unchanged.
Generalises the gallery resolution helper into findBackendPath()
shared by MLX and vLLM workers; extracts ParseNodeLabels for the
comma-separated label parsing both use.
Ships with two compose recipes (`docker-compose.vllm-multinode.yaml`
for NVIDIA, `docker-compose.vllm-multinode.intel.yaml` for Intel
XPU/xccl) plus `tests/e2e/vllm-multinode/smoke.sh`. Both vendors are
supported (NCCL for CUDA/ROCm, xccl for XPU) but mixed-vendor DP is
not -- PyTorch's process group requires every rank to use the same
collective backend, and NCCL/xccl/gloo don't interoperate.
Out of scope (deferred): SmartRouter-driven placement of follower
ranks via NATS backend.install events, follower log streaming through
/api/backend-logs, tensor-parallel across nodes, disaggregated
prefill via KVTransferConfig.
Assisted-by: Claude:claude-opus-4-7 [Claude Code]
Signed-off-by: Richard Palethorpe <io@richiejp.com>
* test(vllm): CPU-only end-to-end test for multi-node DP
Adds tests/e2e/vllm-multinode/, a Ginkgo + testcontainers-go suite
that brings up a head + headless follower from the locally-built
local-ai:tests image, bind-mounts the cpu-vllm backend extracted by
make extract-backend-vllm so it's seen as a system backend (no gallery
fetch, no registry server), and asserts a chat completion across both
DP ranks. New `make test-e2e-vllm-multinode` target wires the docker
build, backend extract, and ginkgo run together; BuildKit caches both
images so re-runs only rebuild what changed. Tagged Label("VLLMMultinode")
so the existing distributed suite isn't pulled along.
Two pre-existing bugs surfaced by the test:
1. extract-backend-% (Makefile) failed for every backend, because all
backend images end with `FROM scratch` and `docker create` rejects
an image with no CMD/ENTRYPOINT. Fixed by passing
--entrypoint=/run.sh -- the container is never started, only
docker-cp'd, so the path doesn't have to exist; we just need
anything that satisfies the daemon's create-time validation.
2. backend/python/vllm/run.sh's `serve` shortcut for the multi-node DP
follower exec'd ${EDIR}/venv/bin/vllm directly, but uv bakes an
absolute build-time shebang (`#!/vllm/venv/bin/python3`) that no
longer resolves once the backend is relocated to BackendsPath.
_makeVenvPortable's shebang rewriter only matches paths that
already point at ${EDIR}, so the original shebang slips through
unchanged. Fixed by exec-ing ${EDIR}/venv/bin/python with the script
as an argument -- Python ignores the script's shebang in that case.
The test fixture caps memory aggressively (max_model_len=512,
VLLM_CPU_KVCACHE_SPACE=1, TORCH_COMPILE_DISABLE=1) so two CPU engines
fit on a 32 GB box. TORCH_COMPILE_DISABLE is currently mandatory for
cpu-vllm: torch._inductor's CPU-ISA probe runs even with
enforce_eager=True and needs g++ on PATH, which the LocalAI runtime
image doesn't ship -- to be addressed in a follow-up that bundles a
toolchain in the cpu-vllm backend.
Assisted-by: Claude:claude-opus-4-7 [Claude Code]
Signed-off-by: Richard Palethorpe <io@richiejp.com>
* feat(vllm): bundle a g++ toolchain in the cpu-vllm backend image
torch._inductor's CPU-ISA probe (`cpu_model_runner.py:65 "Warming up
model for the compilation"`) shells out to `g++` at vllm engine
startup, regardless of `enforce_eager=True` -- the eager flag only
disables CUDA graphs, not inductor's first-batch warmup. The LocalAI
CPU runtime image (Dockerfile, unconditional apt list) does not ship
build-essential, and the cpu-vllm backend image is `FROM scratch`,
so any non-trivial inference on cpu-vllm crashes with:
torch._inductor.exc.InductorError:
InvalidCxxCompiler: No working C++ compiler found in
torch._inductor.config.cpp.cxx: (None, 'g++')
Bundling the toolchain in the CPU runtime image would bloat every
non-vllm-CPU deployment and force a single GCC version on backends
that may want clang or a different version. So this lives in the
backend, gated to BUILD_TYPE=='' (the CPU profile).
`package.sh` snapshots g++ + binutils + cc1plus + libstdc++ + libc6
(runtime + dev) + the math libs cc1plus links (libisl/libmpc/libmpfr/
libjansson) into ${BACKEND}/toolchain/, mirroring /usr/... layout. The
unversioned binaries on Debian/Ubuntu are symlink chains pointing into
multiarch packages (`g++` -> `g++-13` -> `x86_64-linux-gnu-g++-13`,
the latter in `g++-13-x86-64-linux-gnu`), so the package list resolves
both the version and the arch-triplet variant. Symlinks /lib ->
usr/lib and /lib64 -> usr/lib64 are recreated under the toolchain
root because Ubuntu's UsrMerge keeps them at /, and ld scripts
(`libc.so`, `libm.so`) hardcode `/lib/...` paths that --sysroot
re-roots into the toolchain.
The unversioned `g++`/`gcc`/`cpp` symlinks are replaced with wrapper
shell scripts that resolve their own location at runtime and pass
`--sysroot=<toolchain>` and `-B <toolchain>/usr/lib/gcc/<triplet>/<ver>/`
to the underlying versioned binary. That's how torch's bare `g++ foo.cpp
-o foo` invocation finds cc1plus (-B), system headers (--sysroot), and
the bundled libstdc++ (--sysroot, --sysroot is recursive into linker).
`run.sh` adds the toolchain bin dir to PATH and the toolchain's
shared-lib dir to LD_LIBRARY_PATH -- everything else (header search,
linker search, executable search) is encapsulated in the wrappers.
No-op for non-CPU builds, the dir doesn't exist there.
The cpu-vllm image grows by ~217 MB. Tradeoff is acceptable -- cpu-vllm
is already a niche profile (few users compared to GPU vllm) and the
alternative is a backend that crashes at first inference unless the
operator manually sets TORCH_COMPILE_DISABLE=1, which silently disables
all torch.compile optimizations.
Drops `TORCH_COMPILE_DISABLE=1` from tests/e2e/vllm-multinode -- the
smoke now exercises the real compile path through the bundled toolchain.
Test runtime is +20s for the warmup compile, still <90s end to end.
Assisted-by: Claude:claude-opus-4-7 [Claude Code]
Signed-off-by: Richard Palethorpe <io@richiejp.com>
* fix(vllm): scope jetson-ai-lab index to L4T-specific wheels via pyproject.toml
The L4T arm64 build resolves dependencies through pypi.jetson-ai-lab.io,
which hosts the L4T-specific torch / vllm / flash-attn wheels but also
transparently proxies the rest of PyPI through `/+f/<sha>/<filename>`
URLs. With `--extra-index-url` + `--index-strategy=unsafe-best-match`
uv would pick those proxy URLs for ordinary PyPI packages —
anthropic/openai/propcache/annotated-types — and fail when the proxy
503s. Master is hitting the same bug on its own l4t-vllm matrix entry.
Switch the l4t13 install path to a pyproject.toml that marks the
jetson-ai-lab index `explicit = true` and pins only torch, torchvision,
torchaudio, flash-attn, and vllm to it via [tool.uv.sources]. uv won't
consult the L4T mirror for anything else, so transitive deps fall back
to PyPI as the default index — no exposure to the proxy 503s.
`uv pip install -r requirements.txt` ignores [tool.uv.sources], so the
l4t13 branch in install.sh now invokes `uv pip install --requirement
pyproject.toml` directly, replacing the old requirements-l4t13*.txt
files. Other BUILD_PROFILEs continue using libbackend.sh's
installRequirements and never read pyproject.toml.
Local resolution test (x86_64, dry-run) confirms uv hits the L4T
index for torch and falls through to PyPI for everything else.
Assisted-by: claude-code:claude-opus-4-7-1m [Read] [Edit] [Bash] [Write]
Signed-off-by: Richard Palethorpe <io@richiejp.com>
---------
Signed-off-by: Richard Palethorpe <io@richiejp.com>
`int(x) * 1e9` returns a float because `1e9` is a float literal, but
TranscriptSegment.start/end are integer protobuf fields. This caused
every transcription request to fail with:
TypeError: 'float' object cannot be interpreted as an integer
Multiply first, then cast — `int(x * 1e9)` — to get an int as required.
* feat(api): add /v1/audio/diarization endpoint with sherpa-onnx + vibevoice.cpp
Closes#1648.
OpenAI-style multipart endpoint that returns "who spoke when". Single
endpoint instead of the issue's three-endpoint sketch (refactor /vad,
/vad/embedding, /diarization) — the typical client wants one call, and
embeddings can land later as a sibling without breaking this surface.
Response shape borrows from Pyannote/Deepgram: segments carry a
normalised SPEAKER_NN id (zero-padded, stable across the response) plus
the raw backend label, optional per-segment text when the backend bundles
ASR, and a speakers summary in verbose_json. response_format also accepts
rttm so consumers can pipe straight into pyannote.metrics / dscore.
Backends:
* vibevoice-cpp — Diarize() reuses the existing vv_capi_asr pass.
vibevoice's ASR prompt asks the model to emit
[{Start,End,Speaker,Content}] natively, so diarization is a by-product
of the same pass; include_text=true preserves the transcript per
segment, otherwise we drop it.
* sherpa-onnx — wraps the upstream SherpaOnnxOfflineSpeakerDiarization
C API (pyannote segmentation + speaker-embedding extractor + fast
clustering). libsherpa-shim grew config builders, a SetClustering
wrapper for per-call num_clusters/threshold overrides, and a
segment_at accessor (purego can't read field arrays out of
SherpaOnnxOfflineSpeakerDiarizationSegment[] directly).
Plumbing: new Diarize gRPC RPC + DiarizeRequest / DiarizeSegment /
DiarizeResponse messages, threaded through interface.go, base, server,
client, embed. Default Base impl returns unimplemented.
Capability surfaces all updated: FLAG_DIARIZATION usecase,
FeatureAudioDiarization permission (default-on), RouteFeatureRegistry
entries for /v1/audio/diarization and /audio/diarization, audio
instruction-def description widened, CAP_DIARIZATION JS symbol,
swagger regenerated, /api/instructions discovery map updated.
Tests:
* core/backend: speaker-label normalisation (first-seen → SPEAKER_NN,
per-speaker totals, nil-safety, fallback to backend NumSpeakers when
no segments).
* core/http/endpoints/openai: RTTM rendering (file-id basename, negative
duration clamping, fallback id).
* tests/e2e: mock-backend grew a deterministic Diarize that emits
raw labels "5","2","5" so the e2e suite verifies SPEAKER_NN
remapping, verbose_json speakers summary + transcript pass-through
(gated by include_text), RTTM bytes content-type, and rejection of
unknown response_format. mock-diarize model config registered with
known_usecases=[FLAG_DIARIZATION] to bypass the backend-name guard.
Docs: new features/audio-diarization.md (request/response, RTTM example,
sherpa-onnx + vibevoice setup), cross-link from audio-to-text.md, entry
in whats-new.md.
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Assisted-by: Claude:claude-opus-4-7 [Claude Code]
* fix(diarization): correct sherpa-onnx symbol name + lint cleanup
CI failures on #9654:
* sherpa-onnx-grpc-{tts,transcription} and sherpa-onnx-realtime panicked
at backend startup with `undefined symbol: SherpaOnnxDestroyOfflineSpeakerDiarizationResult`.
Upstream's actual symbol is SherpaOnnxOfflineSpeakerDiarizationDestroyResult
(Destroy in the middle, not the prefix); the rest of the diarization
surface follows the same naming pattern. The mismatched name made
purego.RegisterLibFunc fail at dlopen time and crashed the gRPC server
before the BeforeAll could probe Health, taking down every sherpa-onnx
test job — not just the diarization-related ones.
* golangci-lint flagged 5 errcheck violations on new defer cleanups
(os.RemoveAll / Close / conn.Close); wrap each in a `defer func() { _ = X() }()`
closure (matches the pattern other LocalAI files use for new code, since
pre-existing bare defers are grandfathered in via new-from-merge-base).
* golangci-lint also flagged forbidigo violations: the new
diarization_test.go files used testing.T-style `t.Errorf` / `t.Fatalf`,
which are forbidden by the project's coding-style policy
(.agents/coding-style.md). Convert both files to Ginkgo/Gomega
Describe/It with Expect(...) — they get picked up by the existing
TestBackend / TestOpenAI suites, no new suite plumbing needed.
* modernize linter: tightened the diarization segment loop to
`for i := range int(numSegments)` (Go 1.22+ idiom).
Verified locally: golangci-lint with new-from-merge-base=origin/master
reports 0 issues across all touched packages, and the four mocked
diarization e2e specs in tests/e2e/mock_backend_test.go still pass.
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Assisted-by: Claude:claude-opus-4-7 [Claude Code]
* fix(vibevoice-cpp): convert non-WAV input via ffmpeg + raise ASR token budget
Confirmed end-to-end against a real LocalAI instance with vibevoice-asr-q4_k
loaded and the multi-speaker MP3 sample at vibevoice.cpp/samples/2p_argument.mp3:
both /v1/audio/transcriptions and /v1/audio/diarization now succeed and
return correctly attributed speaker turns for the full clip.
Two latent issues surfaced once the diarization endpoint actually exercised
the backend with a non-trivial input:
1. vv_capi_asr only accepts WAV via load_wav_24k_mono. The previous code
passed the uploaded path straight through, so anything that wasn't
already a 24 kHz mono s16le WAV failed at the C side with rc=-8 and
the very unhelpful "vv_capi_asr failed". prepareWavInput shells out
to ffmpeg ("-ar 24000 -ac 1 -acodec pcm_s16le") in a per-call temp
dir, matching the rate the model was trained on; both AudioTranscription
and Diarize now route through it. This is the same shape sherpa-onnx
uses (utils.AudioToWav), but vibevoice needs 24 kHz rather than 16 kHz
so we don't reuse that helper.
2. The C ABI's max_new_tokens defaults to 256 when 0 is passed. That's
fine for a five-second clip but not for anything past ~10 s — vibevoice
stops mid-JSON, the parse fails, and the caller sees a hard error.
Pass a much larger budget (16 384 ≈ ~9 minutes of speech at the
model's ~30 tok/s rate); generation stops at EOS so this is a cap
rather than a target.
3. As a defensive belt-and-braces, mirror AudioTranscription's existing
"fall back to a single segment if the model emits non-JSON text"
pattern in Diarize, so partial / unusual model output never produces
a 500. This kept the endpoint usable while diagnosing (1) and (2),
and is the right behaviour to keep.
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Assisted-by: Claude:claude-opus-4-7 [Claude Code]
* fix(vibevoice-cpp): pass valid WAVs through directly so ffmpeg is not required at runtime
Spotted by tests-e2e-backend (1.25.x): the previous fix forced every
incoming audio file through `ffmpeg -ar 24000 ...`, which meant the
backend container — which does not ship ffmpeg — failed even for the
existing happy path where the caller already uploads a WAV. The
container-side error was:
rpc error: code = Unknown desc = vibevoice-cpp: ffmpeg convert to
24k mono wav: exec: "ffmpeg": executable file not found in $PATH
Reading vibevoice.cpp's audio_io.cpp, `load_wav_24k_mono` uses drwav and
already accepts any PCM/IEEE-float WAV at any sample rate, downmixes
multi-channel input to mono, and resamples to 24 kHz internally. So the
only inputs that genuinely need an external converter are non-WAV
formats (MP3, OGG, FLAC, ...).
Detect WAVs by RIFF/WAVE magic at bytes 0..3 / 8..11 and pass them
straight through with a no-op cleanup; everything else still goes
through ffmpeg with the same 24 kHz mono s16le target. The result:
* Container builds without ffmpeg keep working for WAV uploads
(the e2e-backends fixture is jfk.wav at 16 kHz mono s16le).
* MP3 and other non-WAV inputs still get the new ffmpeg conversion
path so the diarization endpoint stays useful.
* If the caller uploads a non-WAV but ffmpeg isn't on PATH, the
surfaced error is still descriptive enough to act on.
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Assisted-by: Claude:claude-opus-4-7 [Claude Code]
* fix(ci): make gcc-14 install in Dockerfile.golang best-effort for jammy bases
The LocalVQE PR (bb033b16) made `gcc-14 g++-14` an unconditional apt
install in backend/Dockerfile.golang and pointed update-alternatives at
them. That works on the default `BASE_IMAGE=ubuntu:24.04` (noble has
gcc-14 in main), but every Go backend that builds on
`nvcr.io/nvidia/l4t-jetpack:r36.4.0` — jammy under the hood — now fails
at the apt step:
E: Unable to locate package gcc-14
This blocked unrelated jobs:
backend-jobs(*-nvidia-l4t-arm64-{stablediffusion-ggml, sam3-cpp, whisper,
acestep-cpp, qwen3-tts-cpp, vibevoice-cpp}). LocalVQE itself is only
matrix-built on ubuntu:24.04 (CPU + Vulkan), so it doesn't actually
need gcc-14 anywhere else.
Make the gcc-14 install conditional on the package being available in
the configured apt repos. On noble: identical behaviour to today (gcc-14
installed, update-alternatives points at it). On jammy: skip the
gcc-14 stanza entirely and let build-essential's default gcc take over,
which is what the other Go backends compile with anyway.
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Assisted-by: Claude:claude-opus-4-7 [Claude Code]
---------
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* feat(concurrency-groups): per-model exclusive groups for backend loading
Adds `concurrency_groups: [...]` to model YAML configs. Two models that share
a group cannot be loaded concurrently on the same node — loading one evicts
the others, reusing the existing pinned/busy/retry policy from LRU eviction.
Layered design:
- Watchdog (pkg/model): per-node correctness floor — on every Load(), evict
any loaded model that shares a group with the requested one. Pinned skips
surface NeedMore so the loader retries (and ultimately logs a clear
warning), instead of silently allowing the rule to be violated.
- Distributed scheduler (core/services/nodes): soft anti-affinity hint —
scheduleNewModel prefers nodes that don't already host a same-group
model, falling back to eviction only if every candidate has a conflict.
Composes with NodeSelector at the same point in the candidate pipeline.
Per-node, not cluster-wide: VRAM is a node-local resource, and two heavy
models running on different nodes is fine. The ConfigLoader is wired into
SmartRouter via a small ConcurrencyConflictResolver interface so the nodes
package keeps a narrow surface on core/config.
Refactors the inner LRU eviction body into a shared collectEvictionsLocked
helper and the loader retry loop into retryEnforce(fn, maxRetries, interval),
so both LRU and group enforcement share busy/pinned/retry semantics.
Closes#9659.
Assisted-by: Claude:claude-opus-4-7 [Claude Code]
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* fix(watchdog): sync pinned + concurrency_groups at startup
The startup-time watchdog setup lives in initializeWatchdog (startup.go),
not in startWatchdog (watchdog.go). The latter is only invoked from the
runtime-settings RestartWatchdog path. As a result, neither
SyncPinnedModelsToWatchdog nor SyncModelGroupsToWatchdog ran at boot,
so `pinned: true` and `concurrency_groups: [...]` only became effective
after a settings-driven watchdog restart.
Fix by adding both sync calls to initializeWatchdog. Confirmed end-to-end:
loading model A in group "heavy", then C with no group (coexists),
then B in group "heavy" now correctly evicts A and leaves [B, C].
Assisted-by: Claude:claude-opus-4-7 [Claude Code]
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* fix(test): satisfy errcheck on new os.Remove in concurrency_groups spec
CI lint runs new-from-merge-base, so the existing pre-existing
`defer os.Remove(tmp.Name())` lines are baseline-grandfathered but the
one introduced by the concurrency_groups YAML round-trip test is held
to errcheck. Wrap the remove in a closure that discards the error.
Assisted-by: Claude:claude-opus-4-7 [Claude Code]
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
---------
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Update quic-go from v0.54.1 to v0.59.0 to fix the crypto/tls session
ticket panic described in quic-go/quic-go#5572.
Co-dependency go-libp2p upgraded from v0.43.0 to v0.48.0 (required for
quic-go v0.59.0 compatibility).
Signed-off-by: Beshoy Girgis <shoy@1ds.us>
feat(audio-transform): add LocalVQE backend, bidi gRPC RPC, Studio UI
Introduce a generic "audio transform" capability for any audio-in / audio-out
operation (echo cancellation, noise suppression, dereverberation, voice
conversion, etc.) and ship LocalVQE as the first backend implementation.
Backend protocol:
- Two new gRPC RPCs in backend.proto: unary AudioTransform for batch and
bidirectional AudioTransformStream for low-latency frame-by-frame use.
This is the first bidi stream in the proto; per-frame unary at LocalVQE's
16 ms hop would be RTT-bound. Wire it through pkg/grpc/{client,server,
embed,interface,base} with paired-channel ergonomics.
LocalVQE backend (backend/go/localvqe/):
- Go-Purego wrapper around upstream liblocalvqe.so. CMake builds the upstream
shared lib + its libggml-cpu-*.so runtime variants directly — no MODULE
wrapper needed because LocalVQE handles CPU feature selection internally
via GGML_BACKEND_DL.
- Sets GGML_NTHREADS from opts.Threads (or runtime.NumCPU()-1) — without it
LocalVQE runs single-threaded at ~1× realtime instead of the documented
~9.6×.
- Reference-length policy: zero-pad short refs, truncate long ones (the
trailing portion can't have leaked into a mic that wasn't recording).
- Ginkgo test suite (9 always-on specs + 2 model-gated).
HTTP layer:
- POST /audio/transformations (alias /audio/transform): multipart batch
endpoint, accepts audio + optional reference + params[*]=v form fields.
Persists inputs alongside the output in GeneratedContentDir/audio so the
React UI history can replay past (audio, reference, output) triples.
- GET /audio/transformations/stream: WebSocket bidi, 16 ms PCM frames
(interleaved stereo mic+ref in, mono out). JSON session.update envelope
for config; constants hoisted in core/schema/audio_transform.go.
- ffmpeg-based input normalisation to 16 kHz mono s16 WAV via the existing
utils.AudioToWav (with passthrough fast-path), so the user can upload any
format / rate without seeing the model's strict 16 kHz constraint.
- BackendTraceAudioTransform integration so /api/backend-traces and the
Traces UI light up with audio_snippet base64 and timing.
- Routes registered under routes/localai.go (LocalAI extension; OpenAI has
no /audio/transformations endpoint), traced via TraceMiddleware.
Auth + capability + importer:
- FLAG_AUDIO_TRANSFORM (model_config.go), FeatureAudioTransform (default-on,
in APIFeatures), three RouteFeatureRegistry rows.
- localvqe added to knownPrefOnlyBackends with modality "audio-transform".
- Gallery entry localvqe-v1-1.3m (sha256-pinned, hosted on
huggingface.co/LocalAI-io/LocalVQE).
React UI:
- New /app/transform page surfaced via a dedicated "Enhance" sidebar
section (sibling of Tools / Biometrics) — the page is enhancement, not
generation, so it lives outside Studio. Two AudioInput components
(Upload + Record tabs, drag-drop, mic capture).
- Echo-test button: records mic while playing the loaded reference through
the speakers — the mic naturally picks up speaker bleed, giving a real
(mic, ref) pair for AEC testing without leaving the UI.
- Reusable WaveformPlayer (canvas peaks + click-to-seek + audio controls)
and useAudioPeaks hook (shared module-scoped AudioContext to avoid
hitting browser context limits with three players on one page); migrated
TTS, Sound, Traces audio blocks to use it.
- Past runs saved in localStorage via useMediaHistory('audio-transform') —
the history entry stores all three URLs so clicking re-renders the full
triple, not just the output.
Build + e2e:
- 11 matrix entries removed from .github/workflows/backend.yml (CUDA, ROCm,
SYCL, Metal, L4T): upstream supports only CPU + Vulkan, so we ship those
two and let GPU-class hardware route through Vulkan in the gallery
capabilities map.
- tests-localvqe-grpc-transform job in test-extra.yml (gated on
detect-changes.outputs.localvqe).
- New audio_transform capability + 4 specs in tests/e2e-backends.
- Playwright spec suite in core/http/react-ui/e2e/audio-transform.spec.js
(8 specs covering tabs, file upload, multipart shape, history, errors).
Docs:
- New docs/content/features/audio-transform.md covering the (audio,
reference) mental model, batch + WebSocket wire formats, LocalVQE param
keys, and a YAML config example. Cross-links from text-to-audio and
audio-to-text feature pages.
Assisted-by: Claude:claude-opus-4-7 [Bash Read Edit Write Agent TaskCreate]
Signed-off-by: Richard Palethorpe <io@richiejp.com>
* fix(nodes/health): skip stale-marking already-offline nodes
The health monitor re-emitted "Node heartbeat stale" + "Marking stale
node offline" + MarkOffline on every cycle for nodes that were already
in the offline (or unhealthy) state. For an operator-stopped node this
flooded the logs with the same WARN+INFO pair every check interval.
Skip the staleness branch when the node is already StatusOffline /
StatusUnhealthy — the state is already what we'd write, so neither the
log lines nor the DB update carry information.
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* fix(worker): wait for backend gRPC bind before replying to backend.install
The backend supervisor used to wait up to 4s (20 × 200ms) for the
backend's gRPC server to answer a HealthCheck, then log a warning and
reply Success with the bind address anyway. On slower nodes (a Jetson
Orin doing first-boot CUDA init, large CGO library load) the gRPC
listener wasn't up yet, so the frontend's first LoadModel dial returned
"connect: connection refused" and the operator chased a phantom network
issue instead of a startup-timing one.
Two changes:
- Bump the readiness window to 30s. CUDA init on Orin/Thor first boot
measures in seconds, not milliseconds.
- On deadline-exceeded, stop the half-started process, recycle the
port, and return an error with the backend's stderr tail. The
frontend now gets a real failure with diagnostic context instead of
a misleading ECONNREFUSED on a downstream dial.
Process death during the wait window keeps its existing fast-fail path.
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* fix(distributed): route auto-upgrade through BackendManager + bump LocalAGI/LocalRecall
Two distributed-mode bugs that surfaced together in the orchestrator
logs:
1. Auto-upgrade always failed with "backend not found".
UpgradeChecker correctly routed CheckUpgrades through the active
BackendManager (so the frontend aggregates worker state), but the
auto-upgrade branch right below called gallery.UpgradeBackend
directly with the frontend's SystemState. In distributed mode the
frontend has no backends installed locally, so ListSystemBackends
returned empty and Get(name) failed for every reported upgrade.
Auto-upgrade now also goes through BackendManager.UpgradeBackend,
which fans out to workers via NATS.
2. Embedding-load failure on a remote node crashed the orchestrator.
When RAG init lazily called NewPersistentPostgresCollection and the
remote embedding worker was unreachable, LocalRecall called
os.Exit(1) inside the constructor, killing the orchestrator pod.
LocalRecall now returns errors instead, LocalAGI surfaces them as a
nil collection, and the existing RAGProviderFromState path returns
(nil, nil, false) — the same code path the agent pool already takes
when no RAG is configured. The orchestrator stays up; chat requests
degrade to "no RAG available" until the embedding worker recovers.
Bumps:
github.com/mudler/LocalAGI → e83bf515d010
github.com/mudler/localrecall → 6138c1f535ab
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
---------
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Co-authored-by: Ettore Di Giacinto <mudler@localai.io>
The LanguageSwitcher added in the i18n PR (#9642) lives in the sidebar
and also uses aria-haspopup="listbox" — same attribute the import-form
SearchableSelect uses. The Batch D / E tests' helper resolved the trigger
with `page.locator('button[aria-haspopup="listbox"]').first()`, which now
returns the language switcher (rendered first in DOM order, in the
sidebar) instead of the backend dropdown.
After clicking the wrong button, getByRole('option', { name: 'llama-cpp' })
naturally never resolves — language options aren't backend names — and
the test times out at 30s.
Scope the locator to the <main className="main-content"> wrapper so only
buttons inside the route's main content area match. The page layout has
the Sidebar outside <main>, so this cleanly excludes it.
Assisted-by: Claude:claude-opus-4-7[1m] [Claude Code]
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
The Backend dropdown is disabled while /backends/known is in flight
(disabled={isSubmitting || backendsLoading} in ImportModel.jsx). Until
now the disabled prop only guarded the internal onClick handler — there
was no `disabled` HTML attribute on the <button>, so the element
remained "actionable" from the outside.
That regressed the import-form-ux Batch D / E Playwright tests after
the i18next-suspense PR (#9642): suspending on the importModel
namespace defers the useEffect that fetches /backends/known, so when
the test calls backendTrigger.click() the button is rendered but
backendsLoading is still true. The click hits the no-op branch,
the dropdown stays closed, and `getByRole('option', { name: 'llama-cpp' })`
times out at 30s.
Surfacing the disabled state on the actual <button> makes Playwright
auto-wait until the dropdown is ready, fixes a11y (screen readers now
announce "disabled"), and removes the button from the tab order while
loading.
Assisted-by: Claude:claude-opus-4-7[1m] [Claude Code]
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
