* feat(tts): stream Qwen text through audio pipelines
Feed cleaned LLM deltas into one persistent Qwen synthesis session.
Return audio before input ends and preserve the live RPC through model
cache, staging and in-flight wrappers. Stop generation on sink failures
or cancellation without replaying speech through a fallback.
Pin the incremental native fork and verify early audio, late text,
finalization and cancellation with lifecycle and real-model checks.
Assisted-by: Codex:gpt-6.1-sol
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* build(tts): use the organization Qwen fork
Fetch the reviewed incremental synthesis commit from localai-org so
backend builds use the organization-maintained fork.
Assisted-by: Codex:gpt-6.1-sol
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* feat(gallery): add a Qwen streaming assistant preset
Expose an English pipeline using existing VAD, ASR, LLM and named-voice
TTS models. Enable incremental text and audio delivery, and document
separate component installs because the gallery has no dependency
installer.
Assisted-by: Codex:gpt-6.1-sol
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* fix(gallery): use Parakeet TDT v3 for the assistant
Select the existing Parakeet TDT 0.6B v3 transcription entry and update
its prerequisite install command. Keep the preset's English voice and
incremental Qwen synthesis configuration.
Assisted-by: Codex:gpt-6.1-sol
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* fix(kokoros): reject unsupported backend RPCs
Implement the MotionStream and SlotCache methods required by the shared
protobuf service. Return Unimplemented so Kokoros compiles without
advertising support for either operation.
Assisted-by: Codex:gpt-6.1-sol
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* fix(tts): mark cleanup and logging results ignored
Explicitly ignore connection shutdown, test server shutdown and
Ginkgo diagnostic writer errors. These operations do not replace the
stream's inference or sink errors, and errcheck requires their return
values to be acknowledged.
Assisted-by: Codex:gpt-6.1-sol
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* feat(tts): stream buffered clauses in realtime pipelines
Start clause synthesis while LLM text is still arriving when both streaming flags are enabled. Keep explicit clause opt-out and native Qwen sessions, bound automatic queued work, and choose streaming support on resolved failover targets.
Assisted-by: Codex:gpt-6.1-sol
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* test(tts): bound synthetic voice clone generation
The sine-wave reference checks speaker conditioning and WAV output, but does not provide a meaningful speech EOS target. Limit that smoke request to 128 frames and verify the request limit reaches native synthesis. Leave production sampling defaults unchanged.
Assisted-by: Codex:gpt-6.1-sol
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* fix(tts): require explicit clause buffering opt-in
Keep whole-reply synthesis for nonnative engines when clause_chunking is unset or false. Use the bounded cancellable worker when explicitly enabled, with unary clauses when TTS streaming is disabled. Preserve native Qwen precedence for every clause setting.
Assisted-by: Codex:gpt-6.1-sol
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* feat(gallery): add VITS and Piper streaming assistants
Provide comparable full pipeline configurations with shared VAD, Parakeet transcription, and Qwen chat components. Enable clause buffering explicitly and document the separate prerequisite model installs and Talk selection.
Assisted-by: Codex:gpt-6.1-sol
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>
stable-diffusion.cpp PR #2148 added MOSS-Image support, which needs an
external tokenizer file passed via the --tokenizer CLI arg. The C++ binding
(gosd.cpp) never parsed a tokenizer option nor set ctx_params.tokenizer, so
MOSS-Image could not load through LocalAI.
- Wire the "tokenizer" option in gosd.cpp (parse + assign to ctx_params)
- Resolve "tokenizer:..." paths against the model directory in gosd.go,
the same way *_path options are resolved
- Bump the vendored stable-diffusion.cpp to 68285df (moss-image merge)
- Add moss-image-base and moss-image-sft gallery entries (int8_convrot)
FLUX.3 Image Klein is already in the gallery (added in #12736).
Co-authored-by: Ettore Di Giacinto <mudler@localai.io>
* chore(vllm-cpp): bump MLX pin to 0.32.4
The macOS runner's pip index no longer offers mlx==0.29.4 (it lists only
0.32.4), so the darwin vllm-cpp build fails at the MLX venv step. Move the
pin to 0.32.4 and update the README default to match.
Not verified on macOS: the Makefile was only dry-run on Linux.
Assisted-by: Claude Code:claude-sonnet-5-5
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* fix(vllm-cpp): pin MLX 0.32.0, the version the engine builds against
The macOS build fails against mlx 0.32.4:
metal_mlx_provider.mm:147:5: error: no matching member function
for call to 'set_data'
The vllm.cpp Metal provider calls the five-argument
array::set_data(Buffer, size, strides, flags, Deleter). The 0.32.4
headers only declare set_data(allocator::Data) and the Data overload
with size, strides, flags and offset; 0.32.0 still has the Buffer one.
The engine's own CI installs 'mlx==0.32.0' (ci.yml line 1798,
release.yml line 360), so pin that and update the README default.
Not verified on macOS: only the Makefile dry-run and the wheel
headers were checked on Linux.
Assisted-by: Claude Code:claude-sonnet-5-5
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* fix(vllm-cpp): install the MLX 0.32.0 wheel by URL and hash
The macOS runner's pip index offers only mlx 0.32.4, so
'pip install mlx==0.32.0' fails with "Could not find a version that
satisfies the requirement". 0.32.4 cannot be used: its headers lack the
five-argument array::set_data the engine calls.
Install the 0.32.0 wheel directly by URL with a sha256 check. The wheel
is mlx-metal (py3-none-macosx_14_0_arm64) from PyPI, sha256
5b64b20ac24b0c401f489de01e8209edc4d372125201f19314e6f39e385322aa. It
holds mlx/include, mlx/lib/libmlx.dylib and mlx/lib/mlx.metallib, which
is all the build and package.sh use, and it is not tied to a CPython
version. The URL and hash are Makefile variables; the stamp stays keyed
on MLX_VERSION.
Not verified on macOS: only the Makefile dry-run, the wheel contents and
its sha256 were checked on Linux.
Assisted-by: Claude Code:claude-sonnet-5-5
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>
* feat(stablediffusion-ggml): run MiniMax-H3 video with reference conditioning
Bump stable-diffusion.cpp to 2988060 and follow its tiling rename
(tile_size_x/y -> tile_size_w/h, rel_size_x/y -> rel_size_w/h).
GenerateVideo now passes params["ref_image"] and the request audio to
the MiniMax-H3 Ref2VA inputs (ref_images, ref_audios). The audio must
be a WAV file. params["ref_video"] is refused, since the proto cannot
carry a frame sequence yet.
stable-diffusion.cpp cannot tell the FL2VA and REF2VA DiTs apart, and a
reference input on FL2VA renders a wrong video without an error. The
backend now reads the video_partition option, or the model file name,
and fails early on that mix. Keyframes plus references are refused too.
Add gallery entries for the FL2VA and REF2VA Q4_K_M sets. They do not
enable any bf16 dequantisation.
Assisted-by: Claude:claude-sonnet-5-5 go vet gofmt
* feat(stablediffusion-ggml): take several reference images, videos and audio clips
Phase 1 passed one reference image and one reference audio clip to
stable-diffusion.cpp and refused ref_video. Add repeated reference fields
to GenerateVideoRequest (ref_images, ref_videos, ref_audios) so the
MiniMax-H3 Ref2VA path can use as many references as upstream accepts. A
reference video is a directory of frames with an fps and an optional
paired audio clip.
The /video endpoint stages the new inputs through the same path as
start_image, end_image and audio. Frames of one reference video are
written to a private directory under zero-padded names so the backend can
read them in order. The input count and frame count per request are
bounded.
The backend merges the new fields with params.ref_image, params.ref_video
and audio, which keep working. The FL2VA partition refuses any reference
input, REF2VA accepts all of them, and keyframes stay exclusive with
references. Reference audio is still WAV only.
Assisted-by: Claude:claude-sonnet-5-5 [Bash]
* feat(gallery): offer MiniMax-H3 by partition, DiT layout and format
Name each MiniMax-H3 entry after its partition (fl2va or ref2va), DiT
layout (unpruned or pruned) and format. Add the vllm.cpp builds that its
model page lists: pruned Q8_0 (renders), pruned Q6_K, Q5_0, Q4_K, Q3_K and
Q2_K (load only), and the pruned FL2VA NVFP4 file (loads only). The
earlier names minimax-h3-fl2va-q4 and minimax-h3-ref2va-q4 stay as
aliases. The two stable-diffusion.cpp entries become
minimax-h3-fl2va-unpruned-q4-ggml and minimax-h3-ref2va-pruned-q4-ggml,
and the latter now lists video input.
Fix two bad sources. The stable-diffusion.cpp entries took the VAEs from
a path that does not exist in the GGUF repository; they now use the
Comfy-Org copies. The vllm.cpp entries paired the fp16 video VAE weights
with a config from a different directory; they now use the official
weights and config from FL2VA/video_vae/source/.
Drop video_dequant_bf16 from every vllm.cpp entry. Expanding a quantised
DiT to bf16 is the suspected cause of a host out-of-memory error, so the
quantised weights stay resident. The model page records its renders with
the expansion on, and each description says so.
Assisted-by: Claude:claude-sonnet-5-5 [Bash]
* feat(gallery): add pruned FL2VA Q4 MiniMax-H3 for sd.cpp
Add minimax-h3-fl2va-pruned-q4-ggml, the 11.4 GB pruned FL2VA
counterpart of the existing pruned REF2VA entry.
Assisted-by: Claude Code:claude-sonnet-5-5 [rc]
---------
Co-authored-by: Ettore Di Giacinto <mudler@localai.io>
Add the GEM-X native backend with live pose and root displacement output, bounded duplex WebSocket streaming, session authorization, metrics and usage accounting. Share the embedded bidirectional RPC channel adapter across streaming backends.
Include backend build and gallery integration, API documentation, schemas and tests.
Assisted-by: Codex:gpt-6
Signed-off-by: Richard Palethorpe <io@richiejp.com>
(cherry picked from commit 016dc3d71a)
With DOT, EUCLID or MANHATTAN the backend returns Qdrant's raw scores,
not similarity-like values. Router, face and voice stores compare
against cosine thresholds, so log a warning at load when another
metric is configured, and document the behaviour on the stores page.
Assisted-by: Claude Code:claude-sonnet-5-5
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
StoresSet sent the whole batch in one Upsert under a single
request_timeout_ms, so a large batch could hit the timeout even though
every point was fine. Send at most 500 points per request and apply
the timeout to each request.
Assisted-by: Claude Code:claude-sonnet-5-5
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
A TLS 1.3 floor rejects Qdrant endpoints that only speak TLS 1.2, such
as some managed deployments and proxies. Use TLS 1.2 as the minimum.
When Load fails after the client was created, it closed the client but
left it set, so Free closed it a second time. Clear it after closing.
Assisted-by: Claude Code:claude-sonnet-5-5
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* test(omnivoice): expose ignored duration controls
Compile the real wrapper with a model-free engine fixture. Four failing
assertions expose ignored duration, postprocessing, and validation.
Assisted-by: nib:gpt-6-astra
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* feat(omnivoice): add bounded duration controls
Validate request controls before reference reads or native output.
Preserve engine defaults when controls are omitted. Bound forced
single-shot duration to 30 seconds as wrapper resource policy.
Exercise the real native boundary, streaming rejection, and defaults
with synthetic PCM. Document quantization and postprocessing limits.
Assisted-by: nib:gpt-6-astra
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* test(omnivoice): require fixture-free ordinary tests
Expose the ordinary suite regression before isolating native fixtures.
Require selected native tests to reject a missing fixture explicitly.
Assisted-by: nib:gpt-6-astra
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* fix(omnivoice): isolate native fixture tests
Ordinary Go tests must run without a compiled native fixture.
Select fixture specs explicitly in the dedicated gate while keeping
parser tests ordinary and native checks mandatory.
Assisted-by: nib:gpt-6-astra
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>
test: satisfy lint checks for upscale and diagnostics
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Co-authored-by: Ettore Di Giacinto <mudler@localai.io>
Update stable-diffusion.cpp to its merged Iris support commit and add the original F32 Iris weights with the Qwen3-VL Q4_K_M text encoder to the model gallery.
Assisted-by: nib:gpt-5.6-sol
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Co-authored-by: Ettore Di Giacinto <mudler@localai.io>
* feat(voice-detect): report the encoder family so audio-registered voices are fingerprinted
A voice registered from audio through the voice-detect backend had no
encoder fingerprint, so the parakeet-cpp backend could not tell whether
it was comparable with the loaded speaker model and could only fall back
to the file-name rule.
The voice-detect backend now binds the three new libvoicedetect accessors
with a symbol probe (an older library still loads and reports nothing),
copies the borrowed strings at once and never frees them. It also hashes
the model file once at load. VoiceEmbedResponse gains two optional
fields, encoder_family and encoder_weights ("sha256:<hex>", empty when
the model is not a plain file).
/v1/voice/register stores them as encoder_family and a new
encoder_weights field in the registry entry; model keeps the encoder
name, so the 1:N identify filter by name is unchanged for old entries.
A voice with a family is sent to the backend whatever its file name, and
the backend decides by family. /v1/voice/identify compares the family
when both the stored voice and the probe have one. Old entries load
without the fields and stay unfingerprinted.
Assisted-by: Claude:claude-sonnet-5-5 [Claude Code]
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* docs(voice-detect): say how a voice with weights but no family is treated
Assisted-by: Claude:claude-sonnet-5-5 [Claude Code]
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* fix(voice-detect): mark the model file open as intended for gosec
Assisted-by: Claude:claude-sonnet-5-5 [Claude Code]
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>
* feat(diarization): return sound events with include_sounds
A client that wants text, speakers, voice prints and sound events had to
make a diarization call and a separate sound call. Add an include_sounds
request field to /v1/audio/diarization that adds a sounds array of closed
events {start, end, label, confidence}, in seconds.
The parakeet-cpp backend runs a tagger-only scene stream over the clip,
the same stream and thresholds the live path uses, so a clip gives the
same events offline and live. A model with no sound_model companion, or a
backend that does not report sound events, fails with 501 and the stable
code include_sounds_unsupported instead of an empty list. The proto
carries sounds_included so an empty list still means "nothing heard".
The localai-proxy backend forwards the field. Swagger, docs and the
e2e mock backend are updated.
Assisted-by: Claude:claude-sonnet-5-5 [protoc swag go]
* feat(gallery): add parakeet-cpp-multilingual-diarization-speakers-sounds
Same as parakeet-cpp-multilingual-diarization-speakers (TDT 0.6B v3,
Nemotron-3-Diarization, WeSpeaker) plus a CED-Tiny sound_model, so one
model name serves /v1/audio/diarization with include_text,
include_speaker_profiles and include_sounds. It declares the
sound_classification usecase like the realtime scene entries.
Assisted-by: Claude:claude-sonnet-5-5
---------
Co-authored-by: Ettore Di Giacinto <mudler@localai.io>
The worker answered model.unload by calling Free on the first backend
process in its map, whatever model the request named. Every idle
scale-down or LRU eviction of one model could therefore empty another
model's backend on the same node. LocalAI still counted that model as
loaded, so its next request failed. A parakeet diarization model then
returned 501 "speaker profiles require a loaded speaker encoder" until
someone reloaded it by hand.
Resolve the target from the model name (all replicas), prefer an
address when the request carries one, and free nothing for an unknown
model.
Also let parakeet-cpp Diarize check the diarization model before the
speaker-profile capability. A backend with nothing loaded now answers
FailedPrecondition, which LocalAI treats as a stale replica and
reloads, instead of a final Unimplemented.
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Co-authored-by: Ettore Di Giacinto <mudler@localai.io>
Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com>
