mirror of
https://github.com/mudler/LocalAI.git
synced 2026-07-30 18:09:05 -04:00
feat/vllm-cpp-engine-args
7295 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
localai-org-maint-bot
|
b5b47da709 |
fix(vllm-cpp): tolerate Apple Clang folding warning
Keep the GNU constant-folding diagnostics visible on Darwin without allowing vllm.cpp's global -Werror to fail the Metal backend build. Assisted-by: Codex:gpt-5 [systematic-debugging] |
||
|
Ettore Di Giacinto
|
40e7c76ec2 |
chore(vllm-cpp): bump the vllm.cpp pin to main; GGUF speculative decoding is real now
The pin sat at f384edcd while vllm.cpp main moved a long way. The ABI is unchanged at v10 and both POD structs are field-identical to the pinned commit (verified by diffing vllm_model_params and vllm_sampling_params across the range), so the Go mirror needs no edit and this is a clean bump. What it picks up matters for this backend: - MTP speculative decoding from a GGUF target, gated end to end on GPU. - DFlash speculative decoding with a GGUF draft AND a GGUF target. - NVFP4 GGUF: dequant, plus a native fp4 compute path for dense and full-attention projections. On the 27B that closed a cross-container divergence entirely (the GGUF and safetensors builds of the same quantization run now emit identical tokens) and halved peak RSS. - A real engine fix: the GDN speculative state gather/scatter was mis-striding the widened conv row, so speculation silently corrupted the target's own recurrent state on CPU. Docs corrected accordingly. The section previously told users that mtp and dflash are rejected on a .gguf target and called it a gap in the engine's GGUF loader. That is no longer true, and leaving it would send people to safetensors for no reason. A head-less GGUF is still refused, and the text now says so with the actual cause. The real-library ABI handshake was re-run against a libvllm.so built at the exact pinned commit rather than a stale one: 43 specs pass, reported ABI 10. make lint clean. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Assisted-by: Claude Code:claude-opus-5 [ClaudeCode] |
||
|
localai-org-maint-bot
|
f5b38e28cc | Merge branch 'master' into feat/vllm-cpp-engine-args | ||
|
Richard Palethorpe
|
49ef40a187 |
feat(classifier/VAD): support voice control on low power devices (#10804)
* feat(llama-cpp): route Score through the slot loop Score previously bypassed the slot loop with a direct llama_decode: a conflict guard aborted the whole process if scoring raced generation, the config validator had to reject score alongside chat/completion/embeddings, and every candidate re-decoded the full shared prompt. Add SERVER_TASK_TYPE_SCORE to the (patched) upstream server so score tasks are scheduled like any other slot work: generation and scoring serialize naturally, the shared prompt is decoded once per call, and the slot's prompt cache carries the conversation prefix across calls. Context checkpoints at the score boundary and at the cache-divergence point keep SWA/hybrid/recurrent models (e.g. LFM2.5) from re-prefilling the whole prompt per candidate: warm-turn scoring on a 6-option set drops from ~8s to ~0.5s on a desktop CPU. The conflict guard and the validation split are removed; declaring score with generation usecases on one config is now supported and shares the slot cache. Assisted-by: Claude:claude-fable-5 Signed-off-by: Richard Palethorpe <io@richiejp.com> * feat(realtime): classifier wire types and pipeline config Wire types and YAML config for realtime classifier mode: sessions carry a localai_classifier extension (options with canned replies/tool calls, softmax threshold, normalization, history trimming, fallback modes, and a deterministic wake-word address gate), mirrored by pipeline.classifier in the model YAML and surfaced in the config-meta registry. The localai.classifier.result server event reports the full score distribution per turn. Assisted-by: Claude:claude-fable-5 Signed-off-by: Richard Palethorpe <io@richiejp.com> * feat(realtime): classifier response flow Classifier-mode responses: instead of autoregressive generation, each user turn is prefill-scored against the option list (router.ScoreClassifier prompt/candidate shapes over the Score primitive) and the winning option's canned reply and tool call are emitted through the existing response machinery. Below-threshold turns take the configured fallback (none / canned reply / generate); empty transcripts and unaddressed turns (wake word not mentioned) skip scoring entirely. The scoring probe defaults to the latest user message only — small scorers echo canned replies from prior turns back as the top option otherwise. Built for hardware that can afford prompt processing but not decode: with slot-based Score the option list stays KV-cached across turns, so a turn costs roughly one forward pass over the new words. session_update_error events now carry the validation cause instead of a generic message. Assisted-by: Claude:claude-fable-5 Signed-off-by: Richard Palethorpe <io@richiejp.com> * fix(realtime): bound the VAD tick's scan window and buffer retention The VAD tick loop re-scanned the entire input buffer every 300ms and only trimmed it on zero-segment ticks or commits. Audio that keeps producing segments without a committing pause (steady noise a mic pipeline lets through, music, continuous speech) grew the buffer toward the 100MB cap with each tick rescanning all of it — O(n^2), measured at ~3.3ms of silero per buffered second: past ~90s retained, ticks run back to back and pin ~4 cores until the stream stops. Silero's recurrent state only carries a few hundred ms of context, so rescanning old audio buys nothing. Clip the slice handed to the VAD to the largest silence the commit test can need to measure (server_vad silence window or the semantic eagerness fallback) plus a warm-up margin, and rebase the returned segment times so every downstream consumer keeps whole-buffer coordinates. An open turn whose clipped window is all silence now commits (the silence outran the window) instead of being discarded as no-speech. Independently, retain at most 90s of raw buffer, rebasing the live-feed and EOU cursors on trim — this also bounds the previously unbounded VAD-error path. Turn boundaries are otherwise unchanged: no forced commits, no new coordinator states. pipeline.turn_detection.vad_window_sec can widen the scan window; values below the automatic floor are ignored. The tick body is extracted into vadTick so specs can drive turn detection synchronously (same shape as classifySoundWindow); the babble reproduction that pinned 4 cores now plateaus under 10% of one core. Assisted-by: Claude:claude-fable-5 Signed-off-by: Richard Palethorpe <io@richiejp.com> * fix(backend): let per-model threads override the global default ModelOptions overrode a set per-model threads value with the app-level --threads whenever the latter was non-zero — and WithThreads defaults it to the physical core count, so it always was. The YAML threads: knob has been dead config: a tiny VAD model could never opt down from the global pool size. SetDefaults already fills an unset per-model value from the app config, which is the intended precedence; resolve threads through a helper that honors it (explicit threads: 0 still means unset). Assisted-by: Claude:claude-fable-5 Signed-off-by: Richard Palethorpe <io@richiejp.com> * chore(gallery): single-thread the silero VAD Silero is a ~2MB recurrent model with no exploitable graph parallelism: measured per-call latency is identical at 1 and 10 ORT threads, while every extra pool thread just spin-waits between the realtime loop's frequent tiny inferences. Assisted-by: Claude:claude-fable-5 Signed-off-by: Richard Palethorpe <io@richiejp.com> * docs(realtime): classifier mode, VAD scan window, threads precedence Document the realtime classifier mode (options, threshold guidance, wake-word address gate, empty-transcript handling), the VAD scan window and 90s buffer retention (pipeline.turn_detection.vad_window_sec), the per-model threads precedence, and the M3 classifier note in the realtime state-machine design doc. Assisted-by: Claude:claude-fable-5 Signed-off-by: Richard Palethorpe <io@richiejp.com> * perf(llama-cpp): score all candidates in one batched decode One scoring call is now a single SERVER_TASK_TYPE_SCORE task: the slot decodes the shared prefix (prompt + longest common candidate token prefix) once, then forks one sequence per candidate off it (metadata-only for the unified KV cache, copy-on-write for recurrent state) and decodes every candidate's unique tail in one llama_decode. Previously each candidate was its own task that restored the boundary checkpoint and re-decoded its full tail sequentially, paying per-candidate task and decode overhead. The context reserves SERVER_SCORE_FORK_SEQS extra sequence ids (and recurrent-state cells) beyond the parallel slots via the new common_params::n_seq_score_forks. Forking requires the unified KV cache (already this backend's default) since per-sequence streams would shrink n_ctx_seq; an explicit kv_unified:false disables forking and Score calls that need it fail cleanly. Candidates beyond the fork/output budget decode in successive chunks. Wire contract and scores are unchanged: per-token logprobs are stitched from the shared region and the forked tails. Verified bitwise deterministic call-to-call and independent of candidate order (no cross-fork leakage via equal-length candidate swap); ranking matches the per-candidate implementation on the drone battery (winner softmax 0.99996 vs 0.99997), and >16-candidate chunking, prefix-of-another and empty candidates all pass. Measured on a desktop CPU: warm /api/score calls 0.52s -> 0.23s; warm realtime classifier turns 196-303ms. The 9-candidate drone turn decodes ~17 unique tail tokens in one batch instead of nine sequential ~220ms checkpoint-restore tasks. Assisted-by: Claude:claude-fable-5 Signed-off-by: Richard Palethorpe <io@richiejp.com> * fix(realtime): gate scoring capacity by model usecase Reserve llama.cpp scoring slots only for models that explicitly declare the score usecase, while allowing score to coexist with chat and completion. Reject incompatible unified-KV settings and classifier activation on models without scoring capacity. Propagate application defaults when resolving realtime and preload pipeline stages so unset thread counts are resolved consistently without overriding explicit model settings. Assisted-by: Codex:gpt-5 Signed-off-by: Richard Palethorpe <io@richiejp.com> * fix(ci): honor APT mirrors in the prebuilt llama-cpp compile step The builder-prebuilt path installs gcc-14 with apt directly and ignored the APT_MIRROR/APT_PORTS_MIRROR build args the from-source path already honors, so an ubuntu mirror outage broke every arm64 backend build. Pass the args into the stage and run apt-mirror.sh (already in the build context via COPY . /LocalAI) before the apt step. Assisted-by: Claude:claude-fable-5 Signed-off-by: Richard Palethorpe <io@richiejp.com> * feat(realtime): classifier argument slots via constrained completion Hybrid classify-then-complete: a classifier option's canned tool call can declare typed argument slots (number | enum | string, with defaults and prompt hints) referenced as "{{name}}" in the arguments template. When the option wins, the slots are filled by a short grammar-constrained completion that continues the exact scoring prompt — rendered by the same cached ScoreClassifier, so the llama.cpp prompt cache is already warm — with the chosen route JSON re-opened at the first slot field. A GBNF grammar pins the field skeleton and frees only the values; temperature 0, a couple dozen tokens at most (~300ms on a desktop CPU for two slots). Slot declarations and hints ride the option descriptions in the shared system prompt, informing scoring and the fill alike at no per-turn token cost. The localai.classifier.result event carries the final arguments and a fill_latency_ms. On inference failure the slots' defaults apply; a slot without a default fails the response (or falls through with fallback.mode: generate). Slot filling requires completion alongside score in the scoring model's known_usecases. Verified end-to-end on the Pi drone demo: "fly forward three meters" in distance mode classifies forward and infers {"distance": 3, "units": "meters"} in ~310ms, and the drone flies exactly 3 units. Assisted-by: Claude:claude-fable-5 Signed-off-by: Richard Palethorpe <io@richiejp.com> * feat(realtime): splice filled slot values into classifier replies A classifier option's spoken reply can now reference its tool's argument slots ("Going forward {{distance}} {{units}}."): the values inferred by the slot-fill completion — or the recovery defaults — are spliced into the reply as plain text before it is emitted, so what the assistant says confirms what it actually inferred. Placeholders without a value stay literal, and options without slots are untouched. FillToolArguments now returns the raw slot values alongside the spliced arguments JSON to make the reply templating possible. Assisted-by: Claude:claude-fable-5 Signed-off-by: Richard Palethorpe <io@richiejp.com> * fix(realtime): harden classifier slot completion Reserve context for constrained slot filling, size completions from their encoded output, and encode enum grammar literals as valid JSON. Reject empty enum values and cover the failure modes with regression tests. Assisted-by: Codex:gpt-5 Signed-off-by: Richard Palethorpe <io@richiejp.com> * feat(realtime): prewarm the classifier scoring prompt on registration Swapping a session's classifier option list (a voice-switched command mode, for instance) made the next turns pay a full re-prefill of the new option-list prompt — measured 2.4s vs 0.3s warm on a desktop CPU, and worse: on hybrid-memory models like LFM2.5, whose state cannot be partially rewound (llama.cpp can only restore checkpoints), *every* probe change re-prefilled from scratch whenever the last checkpoint missed the probe boundary, so even same-list turns intermittently cost full prefills. Registering an option list (pipeline seed or session.update) now fires a best-effort background prewarm: two throwaway scores with distinct probes. The first prefills the new option-list prompt; the second, diverging exactly where per-turn probe text starts, plants the backend's rewind point (KV checkpoint) at the stable-prefix boundary that every real turn reuses. The prewarm hides behind the canned mode-switch reply — by the time it finishes speaking, the cache is warm. Idempotent per option set, detached from the registering request's lifetime. Measured on the drone demo (LFM2.5-1.2B, desktop CPU): first turn after a mode switch 2374ms -> 340ms; intermittent same-list full prefills (1.3-2.1s) all -> under 0.5s. For clients that swap lists frequently, options: [parallel:2] on the scoring model additionally keeps one slot per list via prefix-similarity routing (+26MB RSS, unified KV). Assisted-by: Claude:claude-fable-5 Signed-off-by: Richard Palethorpe <io@richiejp.com> * perf(llama-cpp): checkpoint scoring at the caller-declared stable prefix Hybrid-memory models (LFM2.5 shortconv, Qwen3.5 deltanet — where new small models are headed) cannot rewind their state, so any prompt-cache reuse that needs a rewind falls back to a full re-prefill. For classifier scoring that meant every probe change re-processed the whole option-list prompt: the server's checkpoints were placed reactively (at wherever the previous task happened to diverge), so a checkpoint past the next divergence was erased rather than restored — measured as intermittent 2-10s turns on prompts with a 95%+ common prefix. The classifier now computes the probe-invariant prompt prefix once (the byte-wise common prefix of two synthetic probe renders) and declares its length with every Score request; the server maps it to a token boundary and forces a KV checkpoint exactly there on each score prefill. That checkpoint sits at or before every future divergence under the same option list, so it always survives and always restores — repeat scoring costs probe+candidates regardless of how the probe changes. Also: - prewarm reruns on every option-list registration instead of memoizing per list: with boundary checkpoints a redundant rewarm costs two probe-sized decodes, while skipping one after a slot eviction (three lists sharing fewer slots evict in LRU cascades) silently moves a full re-prefill onto the user's next turn - new llama.cpp backend option rs_seq:N exposes bounded recurrent-state rollback outside speculative decoding; measured impractical for deltanet-scale states (65GB for 64 snapshots on Qwen3.5-4B) but cheap insurance for small-state models - docs: the multi-list recipe (parallel:N + sps:0.5 — the default slot similarity threshold funnels distinct lists onto one slot) Measured on the drone demo (LFM2.5-1.2B scorer, desktop CPU), steady state: every turn 285-421ms including mode switches, vs 2.4s post-switch and intermittent 1.3-2.9s re-prefills before. Assisted-by: Claude:claude-fable-5 Signed-off-by: Richard Palethorpe <io@richiejp.com> * fix(realtime): align classifier cache guidance Document the single-score prewarm behavior and clean the vendored score patch formatting. Assisted-by: Codex:gpt-5 Signed-off-by: Richard Palethorpe <io@richiejp.com> * fix(llama-cpp): guard score task for fork backends TurboQuant and Bonsai reuse the primary gRPC server against llama.cpp forks that do not carry LocalAI's slot-based Score patches. Compile the Score integration only for the patched primary backend and return UNIMPLEMENTED from fork builds instead of referencing absent task types and common_params fields. Assisted-by: Codex:gpt-5 [gh] Signed-off-by: Richard Palethorpe <io@richiejp.com> * fix(dev): generate gRPC code before commit lint The coverage phase regenerates ignored protobuf bindings, but lint runs first and can fail against missing or stale output. Generate the pinned bindings before lint so the gate always type-checks the current schema. Assisted-by: Codex:gpt-5 Signed-off-by: Richard Palethorpe <io@richiejp.com> --------- Signed-off-by: Richard Palethorpe <io@richiejp.com> |
||
|
Ettore Di Giacinto
|
df5e2a6d27 |
feat(vllm-cpp): move to vllm.cpp ABI v10 and expose jump-forward decoding
vllm.cpp landed ABI v10 on main while this branch was open. The backend's runtime handshake refuses any library whose reported ABI differs from the mirrors', so the pin and the Go PODs move together or not at all. v10 appends one int32, `enable_jump_forward`, AFTER the v9 fields. Nothing else in the config surface changed: the SGLang reconciliation that carried it explicitly dropped its own duplicate scheduler_policy int in favour of the v9 `scheduling_policy` string this branch already wires, and a diff of EngineParams and the server flags across the window turns up jump forward and nothing else. So the exposure is one new knob, `engine_args.enable_jump_forward` - SGLang's grammar-speed subset, which emits grammar-forced tokens without spending a model step and therefore only affects constrained requests. It is the SECOND tri-state on this struct, and it repeats the trap the first one had: 0 is not "off", it is "defer" (to the environment here, to the model capability for prefix caching), so an explicit `false` has to reach the engine as 2. The bool->tri-state helper and the log renderer are now shared rather than duplicated per field, and named for the encoding instead of for prefix caching, since the next tri-state will want them too. The docs say this outright, because "omitting the key" and "setting it false" being different is not guessable. The Go mirror grows the field plus an EXPLICIT trailing pad: the struct is 8-aligned and now ends on a lone int32, so it is 88 bytes rather than 84. The offset assertions cover it, and the real-library handshake spec (VLLM_CPP_LIBRARY against a CPU libvllm.so built at the new pin) confirms the version agrees: 43 specs pass, ABI reported 10. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Assisted-by: Claude Code:claude-opus-5 [ClaudeCode] |
||
|
Ettore Di Giacinto
|
8ea2b9d912 |
fix(vllm-cpp): resolve the DFlash draft path instead of missing the HF cache
The engine resolves speculative_config.model against a directory containing config.json, or against ~/.cache/huggingface/hub/models--<org>--<repo>/ snapshots/*, and it never downloads. LocalAI keeps models in its own directory, so the repo-id spelling the vLLM docs teach - "z-lab/Qwen3.6-27B-DFlash" - misses the HF cache and dies deep inside the load with "draft checkpoint not found", which reads like a broken checkpoint rather than a model nobody fetched. Resolve it before the load call: the reference as given, then its last path segment under LocalAI's models dir (what LocalAI's own downloader produces), then the whole reference under the models dir. When none resolve, fail there naming both what was asked for and every location tried, so the message says what to do about it. mtp and ngram pass through untouched - neither has a separate draft checkpoint. A speculative_config that does not parse also passes through, because the engine owns config validation and produces the better error. Docs also gain the two limits that were missing and are easy to lose an afternoon to: speculation is Qwen3.5/3.6-only at this engine pin regardless of format, and mtp/dflash need a safetensors target. The latter is a gap in the engine's GGUF loader rather than a property of GGUF - the format carries MTP weights fine, llama.cpp reads them as nextn.* tensors plus a <arch>.nextn_predict_layers key - so the docs say that rather than implying GGUF cannot express it. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Assisted-by: Claude Code:claude-opus-5 [ClaudeCode] |
||
|
Ettore Di Giacinto
|
ed4478bd3f |
chore(vllm-cpp): pin vllm.cpp to the merged ABI v9 tip
974d9d72 was the pre-merge commit on the feature branch. eec09bed is the merged main tip carrying the same ABI v9 surface, and is the tree the capi suite (31/31) and the Go ABI handshake were actually verified against. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Assisted-by: Claude Code:claude-opus-5 [ClaudeCode] |
||
|
Ettore Di Giacinto
|
f13baee3c0 |
test(vllm-cpp): catch ABI pin/mirror skew without model weights
The Go PODs in govllmcpp.go are hand-written against one VLLM_ABI_VERSION and the Makefile pins the vllm.cpp commit that produces it. Nothing checked those two agree short of the e2e suite, which needs a model to run at all, so a pin bump could land with a stale mirror and only fail at a user's first load. VLLM_CPP_LIBRARY now drives a handshake spec that dlopens a built libvllm, binds every symbol, and compares the library's reported ABI against the mirrors'. No weights required. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Assisted-by: Claude Code:claude-opus-5 [ClaudeCode] |
||
|
Ettore Di Giacinto
|
5945b0eb11 |
feat(vllm-cpp): wire the full engine config surface through engine_args
The vllm-cpp backend could configure four of the engine's knobs - block size,
KV block count, max sequence length and max concurrent sequences - out of a
config surface that is considerably larger. Speculative decoding, prefix
caching, the chunked-prefill token budget, the scheduling policy and the
external KV connector were all reachable from vllm.cpp's own HTTP server and
from nothing LocalAI could write in a model config.
Part of that gap was the C ABI itself, which carried strictly less than
EngineParams does; that is fixed upstream in vllm.cpp ABI v9 (this bumps the pin
to it). The rest was here: the backend parsed a flat `options:` list with five
recognised keys and had no way to express a nested JSON document at all.
Configuration now goes through `engine_args:`, the same map the vLLM and SGLang
backends already take, with keys spelled as vLLM's own CLI flags - so a
`speculative_config` or `kv_transfer_config` block written for vLLM works
verbatim:
engine_args:
max_num_batched_tokens: 8192
enable_prefix_caching: true
scheduling_policy: lpm
speculative_config:
method: dflash
model: z-lab/Qwen3.6-27B-DFlash
num_speculative_tokens: 4
kv_transfer_config:
kv_connector: LMCacheConnector
kv_role: kv_both
kv_connector_extra_config: {host: 127.0.0.1, port: 65432}
The `options:` list keeps working, and now reads every key too, so no existing
config breaks; engine_args wins where both set the same key.
Two details worth calling out. `enable_prefix_caching: false` maps to the ABI's
force-OFF state (2), not the 0 that means "let the model capability decide" -
collapsing them would silently turn the cache ON for the dense architectures
that default it on. And cSamplingParams grows the ABI v8 logits-processor tail:
LocalAI installs no processor, but the C side reads those fields off the pointer
we hand it, so a Go struct that stopped short would have had the engine read 16
bytes past our allocation and call whatever sat there.
The importer gets the safetensors counterpart of the llama-cpp MTP hook: a
`vllm-cpp` import of a HuggingFace repo probes config.json and, on a checkpoint
that declares an MTP head, writes speculative_config {method: mtp} into the
generated engine_args. DFlash draft repositories are detected and refused with a
warning rather than configured as standalone models, since a drafter cannot
serve alone. The llama-cpp importer stops applying its own `spec_type:draft-mtp`
options when the chosen backend is vllm-cpp: those are llama.cpp option keys
vllm-cpp does not read, and vllm.cpp rejects MTP over a GGUF source anyway
because the `mtp.*` draft tensors only exist in the safetensors checkpoint.
docs/content/features/text-generation.md gains a vllm.cpp section - the backend
had no documentation page at all - covering the engine_args table, all three
speculative methods, LMCache, and the legacy options list.
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Assisted-by: Claude Code:claude-opus-5 [ClaudeCode]
|
||
|
localai-org-maint-bot
|
bc21f832aa |
gallery: add Laguna S 2.1 GGUF variants (#11188)
Add the official Q4_K_M and Q8_0 builds plus the DFlash speculative-decoding pairing for llama.cpp. Assisted-by: Codex:gpt-5 [Hugging Face API] Co-authored-by: localai-org-maint-bot <306269227+localai-org-maint-bot@users.noreply.github.com> |
||
|
mudler's LocalAI [bot]
|
967becb365 |
chore: ⬆️ Update ikawrakow/ik_llama.cpp to b054a8b983827c01aec59d4dc273a27c492c51c4 (#11175)
⬆️ Update ikawrakow/ik_llama.cpp Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: mudler <2420543+mudler@users.noreply.github.com> Co-authored-by: localai-org-maint-bot <bot-opensource@localaisrl.com> |
||
|
mudler's LocalAI [bot]
|
0569bb30a2 |
chore: ⬆️ Update ggml-org/whisper.cpp to 97c56f1dc1d1100a9d859c865a20c82d22f823ed (#11182)
⬆️ Update ggml-org/whisper.cpp Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: mudler <2420543+mudler@users.noreply.github.com> Co-authored-by: localai-org-maint-bot <bot-opensource@localaisrl.com> |
||
|
mudler's LocalAI [bot]
|
550545c03a |
chore: ⬆️ Update mudler/parakeet.cpp to e747acdaee69b916cef62263ae5f718bda9ff3f3 (#11181)
⬆️ Update mudler/parakeet.cpp Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: mudler <2420543+mudler@users.noreply.github.com> Co-authored-by: localai-org-maint-bot <bot-opensource@localaisrl.com> |
||
|
mudler's LocalAI [bot]
|
c5e5141010 |
fix(ci): unbreak the sglang and darwin nemo backend builds (#11168)
* fix(sglang): keep nvidia-modelopt on a stable release Every cublas sglang image currently fails to build: Failed to build `nvidia-modelopt==0.46.0rc0` Call to `wheel_stub.buildapi.build_wheel` failed ModuleNotFoundError: No module named 'wheel_stub' sglang[all] pulls nvidia-modelopt in through its `diffusion` extra with no version bound of its own, and install.sh adds a GLOBAL --prerelease=allow so that flash-attn-4, which only ships 4.0.0b* wheels, can resolve. Unbounded plus prereleases-allowed picks 0.46.0rc0, whose build backend imports wheel_stub without declaring it in build-system.requires. EXTRA_PIP_INSTALL_FLAGS also starts with --no-build-isolation, so nothing installs wheel_stub and the build dies. Latest stable is 0.45.0 and resolves cleanly. Bounding this one package rather than dropping the global flag, because the flag is load-bearing for flash-attn-4 and this is the narrower change with the smaller blast radius. Raise the bound when 0.46.0 final ships. This is invisible on master because the backend build is path-filtered: sglang is only rebuilt when sglang changes. It surfaces on any PR touching a shared build input such as backend/backend.proto, which rebuilds the whole matrix. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Assisted-by: Claude:claude-opus-5 [Claude Code] * fix(nemo): build the darwin venv on Python 3.12 The darwin nemo image fails to build: ModuleNotFoundError: No module named 'maturin' nemo_toolkit pulls in text2num, a Rust extension built with maturin, whose macOS arm64 wheels start at cp311: 3.0.2 publishes cp311, cp312, cp313 and cp314 and no cp310. libbackend.sh defaults PYTHON_VERSION to 3.10, so pip finds no wheel, falls back to the sdist, and dies in the PEP 517 hook because EXTRA_PIP_INSTALL_FLAGS carries --no-build-isolation and nothing installs the build backend. Taking the prebuilt wheel avoids the source build entirely, so the runner needs no Rust toolchain. Darwin only, deliberately: the Linux profiles resolve a cp310 manylinux wheel for the same package and have no reason to move. The override is set after libbackend.sh is sourced and before installRequirements, the same shape sglang's install.sh already uses for its l4t13 profile. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Assisted-by: Claude:claude-opus-5 [Claude Code] * fix(nemo): pin the darwin portable-Python patch level too The 3.12 bump alone traded one failure for another: curl: (56) The requested URL returned error: 404 make[1]: *** [nemo-asr] Error 56 libbackend builds the portable-Python URL from cpython-${PYTHON_VERSION}.${PYTHON_PATCH}+${PY_STANDALONE_TAG}, and PYTHON_PATCH defaults to 18 because the default interpreter is 3.10.18. Setting only PYTHON_VERSION asked for a 3.12.18 that was never released. Patch 11, not the 12 that sglang/install.sh pairs with 3.12 for l4t13: at the 20250818 tag python-build-standalone published 3.12.12 for linux aarch64 but not for aarch64-apple-darwin, where 3.12.11 is the newest. Both URLs were checked against the release assets rather than assumed to match. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Assisted-by: Claude:claude-opus-5 [Claude Code] --------- Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Co-authored-by: Ettore Di Giacinto <mudler@localai.io> Co-authored-by: localai-org-maint-bot <bot-opensource@localaisrl.com> |
||
|
mudler's LocalAI [bot]
|
47c48e9409 |
chore: ⬆️ Update antirez/ds4 to 54b36ed9ba42da31b24f2d1a5feb075c2475dbb1 (#11178)
⬆️ Update antirez/ds4 Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: mudler <2420543+mudler@users.noreply.github.com> Co-authored-by: localai-org-maint-bot <bot-opensource@localaisrl.com> |
||
|
mudler's LocalAI [bot]
|
6aaf7db5f0 |
chore: ⬆️ Update mudler/depth-anything.cpp to 2028b47ac75a8659c6a9aa617baf09be193eb55f (#11179)
⬆️ Update mudler/depth-anything.cpp Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: mudler <2420543+mudler@users.noreply.github.com> Co-authored-by: localai-org-maint-bot <bot-opensource@localaisrl.com> |
||
|
mudler's LocalAI [bot]
|
193d49001b |
chore: ⬆️ Update CrispStrobe/CrispASR to 754b67289cf1137e3ed722885705f94132fc614f (#11180)
⬆️ Update CrispStrobe/CrispASR Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: mudler <2420543+mudler@users.noreply.github.com> Co-authored-by: localai-org-maint-bot <bot-opensource@localaisrl.com> |
||
|
localai-org-maint-bot
|
8f9184fbb2 |
feat(cli): support systemd socket activation (#11169)
* feat(cli): support systemd socket activation Serve the API from a single stream listener inherited through the systemd activation protocol while retaining the existing address bind path when no listener is provided. Validate activation metadata, preserve the public-bind safety check, and document an on-demand systemd setup. Assisted-by: Codex:gpt-5 * fix(cli): satisfy listener cleanup lint Make the best-effort close explicit so errcheck accepts the deferred systemd listener cleanup. Assisted-by: Codex:gpt-5 [golangci-lint] --------- Co-authored-by: localai-org-maint-bot <306269227+localai-org-maint-bot@users.noreply.github.com> |
||
|
mudler's LocalAI [bot]
|
84972cb745 |
chore(model-gallery): ⬆️ update checksum (#11176)
⬆️ Checksum updates in gallery/index.yaml Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: mudler <2420543+mudler@users.noreply.github.com> |
||
|
localai-org-maint-bot
|
034df6ceb1 |
fix(worker): report RAM alongside GPU memory (#11167)
* fix(worker): report RAM alongside GPU memory Assisted-by: Codex:gpt-5 * feat(ui): show worker RAM on node views Assisted-by: Codex:gpt-5 --------- Co-authored-by: localai-org-maint-bot <306269227+localai-org-maint-bot@users.noreply.github.com> |
||
|
localai-org-maint-bot
|
996bdcecdc |
fix(mlx-vlm): install torch dependencies on Metal (#11164)
Some Transformers processors used by MLX-VLM, including Qwen vision models, import both PyTorch and Torchvision. Include them in the Metal backend environment so model loading does not fail with missing-library errors. Assisted-by: Codex:gpt-5 Co-authored-by: localai-org-maint-bot <306269227+localai-org-maint-bot@users.noreply.github.com> |
||
|
walcz-de
|
4b4faa4ac7 |
feat(cloud-proxy): optional Anthropic prompt-cache breakpoints in translate mode (#11158)
The Anthropic translate provider builds the upstream request from scratch and
never emitted cache_control, so prompt caching was impossible for OpenAI-format
clients routed through cloud-proxy — even though the entire system prompt + tools
prefix is re-sent on every agentic turn.
Add an opt-in cache_prompt flag (ProxyOptions.cache_prompt; model YAML
proxy.cache_prompt: true). On a translate+anthropic model, buildAnthropicRequest
injects cache_control:{type:ephemeral} on the stable prefix — the system block,
the last tool, and the last message block (at most 3 of Anthropic's 4 allowed
breakpoints). Anthropic then serves the repeated prefix at the cache-read rate
(0.1x input) on subsequent calls, cutting cost on multi-turn/agentic workloads.
No effect in passthrough mode, for non-Anthropic providers, or when unset.
System is widened to any so it can carry the block form required to attach
cache_control, while still marshalling as a bare string when caching is off.
Adds a unit test asserting exactly three breakpoints when on and none when off,
and documents the option in docs/content/operations/cloud-proxy.md.
Assisted-by: Claude:opus-4.8
Signed-off-by: stefanwalcz <stefan.walcz@walcz.de>
|
||
|
mudler's LocalAI [bot]
|
0f7186f214 |
feat(ui): replace the stacked operations bar with a one-line strip and an Activity page (#11163)
* feat(ui): record finished gallery operations in a bounded history ring The operations panel drops an operation the moment it succeeds, so a user who steps away cannot tell whether an install finished, failed or was never started. OpCache now keeps the last 50 terminal operations, recorded from the point where an op leaves the cache. Assisted-by: Claude Code:claude-opus-5 [Read] [Edit] [Bash] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(ui): pin the history ring's dedupe, outcome order and start stamp Review of the history ring found four gaps. The dedupe guard and the bounded seen set were unreachable through the exported API and so had no coverage; an in-package spec file now drives opHistory directly. The outcome switch claimed an ordering was load bearing that nothing pinned, so an errored op that never reached Processed now has a spec. Two behaviour fixes come with it. StartedAt was the zero time for ops recovered from the store or replicated from a peer, since neither path stamps a start time, which would have rendered as a two-millennia duration; it now falls back to the finish time. Reusing a cache key with a fresh job ID orphaned the previous stamp, so Set and SetBackend now drop it. The comment on the outcome switch described a state the code cannot be in: CancelOperation sets Cancelled and Processed synchronously before the handler removes the entry, so status.Cancelled already covers the cancel endpoint. The !Processed clause stays for the dismiss endpoint firing on an in-flight op, and the comments now say so. Assisted-by: Claude Code:claude-opus-5 [Read] [Edit] [Bash] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(ui): record operations that end on a peer replica The NATS end event is the only signal a replica gets for an install another replica ran. Record from applyEnd too, deduped by job ID so the originating replica does not record its own broadcast twice. Three start-stamp defects in the same path go with it. applyEnd now drops the stamp unconditionally, since recordTerminal only cleans up on the path where it found a cache key and an end event can overtake the local Set. applyStart drops the stamp of the job whose cache key it replaces, which a peer-driven retry previously stranded. And recordTerminal reads the stamp once instead of testing Exists and then reading, so a concurrent record for the same job can no longer delete the stamp between the two and let the zero time overwrite the finish-time fallback, which the Activity page would render as a two-millennia run. Assisted-by: Claude Code:claude-opus-5 [Read] [Edit] [Bash] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(ui): do not guess the outcome of a peer operation with no local status A replica that restarts mid-operation hydrates its OpCache keys from PostgreSQL, but gallery statuses are in-memory only and come back empty. The end broadcast then landed on recordTerminal's nil-status branch, which reads a missing status as queued-and-removed and filed a successful install as cancelled. That reading is right locally and wrong on the peer path, where a missing status means the outcome was never held here. recordTerminal now takes the source of the terminal event and records nothing when the peer path finds no status, restoring what the replica did before the end event started recording. The local path is unchanged. Also move the ApplyEndForTest seam to the conventional export_test.go, and stop the dedupe spec from claiming to guard the ring's seen set: the local delete removes the status keys, so the broadcast that follows returns before reaching it. An in-package spec that calls recordTerminal twice does the pinning. Assisted-by: Claude Code:claude-opus-5 [Read] [Edit] [Bash] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(api): add GET and DELETE /api/operations/history Admin gated like the rest of the operations API. The live /api/operations payload is unchanged so the one second poll stays small. Assisted-by: Claude Code:claude-opus-5 [Read] [Edit] [Bash] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(ui): expose operation history through OperationsContext Fetched on demand and when the live list shrinks, never on the one second poll interval. Assisted-by: Claude Code:claude-opus-5 [Read] [Edit] [Bash] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(ui): detect operation departure by identity and ignore committing ops in the ETA gate Refetching history on a shrinking live count missed a completion that coincided with a start, which is the common case during a batch install. Track the live job IDs instead, so any departure triggers the refetch regardless of how the count moved. An operation that has finished downloading stays live at currentBytes == totalBytes for the whole commit and install phase and can never produce an estimate, so counting it in the all-or-nothing gate blanked every other operation's time remaining for as long as it lasted. Only operations still moving bytes get a vote. Assisted-by: Claude Code:claude-opus-5 [Read] [Edit] [Bash] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(ui): let only downloading operations gate the time remaining estimate Verifying pins an operation flat below its total for the whole sha256 pass: the AfterDownload hook reports completedBytes plus the finished file against a total summed over every file, then hashes synchronously without emitting progress. Files download sequentially, so a 15 shard model enters that window 14 times, and a byte comparison cannot see it because the counter is genuinely below the total throughout. Gating on phase closes resolving, verifying, committing and persisting in one predicate, so a quiet neighbour no longer blanks every other operation's estimate for minutes at a time. The byte clauses stay: a producer can report downloading with bytes already at the total. Assisted-by: Claude Code:claude-opus-5 [Read] [Edit] [Bash] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(ui): collapse the operations bar to a single line Four concurrent installs used to take four rows above every page. The strip now shows one operation, failure first, with a counter linking to Activity. The close button hides the strip and no longer cancels an install. Assisted-by: Claude Code:claude-opus-5 [Read] [Edit] [Bash] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(ui): keep the operations strip from widening the page and from muting a failure A long install error made the strip report a 1600px minimum width, which sized main-content to fit and gave every page under it a horizontal scrollbar. Inline-size containment plus shrinkable detail and bytes cells keep it inside the viewport. Hiding is no longer able to swallow the hidden job's own failure, a completed removal or staging says so instead of claiming an install, a cancelling operation renders as cancelling, and the live region no longer covers the per-second percentage. Assisted-by: Claude Code:claude-opus-5 [Read] [Edit] [Bash] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(ui): shrink main-content instead of containing the strip, and expose progress min-width on .main-content is what actually lets a long install error shrink, and unlike inline-size containment it has no browser support floor and no latent collapse if the strip ever lands in a shrink-to-fit context. It matches what .app-layout-chat .main-content already does, and it clears pre-existing horizontal overflow on narrow viewports as a side effect. The progress track is now a labelled progressbar, so assistive tech can read the value on demand rather than losing it to the aria-hidden that stopped the live region re-announcing every poll. Assisted-by: Claude Code:claude-opus-5 [Read] [Edit] [Bash] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(ui): add the live operation card for the Activity page Carries the detail the one-line strip has to drop: phase, bytes, the per-node breakdown for cluster installs, and a labelled Cancel button. Cancelling is destructive, so it gets a labelled button rather than a glyph. A cancelling operation drops its progress bar and its time estimate, the same call the strip makes: a percentage still climbing under "Cancelling" reads as the cancel not having taken. Assisted-by: Claude Code:claude-opus-5 [Read] [Edit] [Bash] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(ui): give the operation card a verb, live node disclosure and its per-node detail The card carried no verb, so an install, a removal and a staging op rendered as spinner plus name plus kind tag and were indistinguishable. It now runs the same verb and icon chain as the one-line strip, which is what stops the page that is meant to carry more detail from carrying less. The auto-expand default was evaluated once at mount. An operation is listed as soon as it is admitted but its nodes are filled in only when the fan-out starts reporting, so a card mounted at creation latched on the empty list and stayed collapsed. The default is a live expression now, and state holds only an explicit choice. Also: an optional onRetry gates a Retry button, so the page can own the install reconstruction without the card ever showing a control with nothing behind it; the disclosure moved above the region it controls and gained aria-controls; the toggle is gated at more than one node so the count is never "1 nodes"; an unmapped node status is passed through instead of being relabelled "Queued"; error text is clamped with the full string in the title; and file_name plus the per-node progress bar are rendered again, reviving three CSS rules that had gone dead along with the detail they styled. Assisted-by: Claude Code:claude-opus-5 [Read] [Edit] [Bash] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(ui): add the Activity page Live operations, unacknowledged failures and the record of what finished, at /app/activity in the Operate console. Cancelling an install now lives here behind a labelled button rather than on the strip, and a failed install can be retried: the retry dismisses the failure first so it still reaches the record, then reissues the model, backend or node-scoped backend install. The sidebar Operate entry carries the operation count. The console rail is only rendered on an Operate route and can be collapsed, so a badge there could vanish while operations were still running. Two follow-ups from review fold in here: a failed removal or staging job no longer reports a failed install on either the card or the strip, and the card's error text can shrink so one unbroken token cannot widen the card. Assisted-by: Claude Code:claude-opus-5 [Read] [Edit] [Bash] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(ui): dismiss operations by job, and stop the Activity page contradicting itself Dismissing resolved the job by display id, but /api/operations strips the "node:<nodeID>:" prefix before emitting, so a local install and a node-scoped install of one backend arrive as two jobs sharing one id. Dismissing by id retired whichever came first. That defeated the guarantee retry was built around: with the wrong job dismissed, the reinstall overwrote the acted-on failure's opcache entry in place, bypassing recordTerminal, while an unrelated failure vanished from Needs attention. dismissFailedOp, the card's dismiss control and the strip now all pass the jobID, which is what the endpoint takes. A filter matching nothing rendered the "nothing has ever run" empty state while the header counted the records the filter had hidden. The empty state is now gated on the All chip and a narrowed view gets its own message plus a way back; the header counts the instance rather than the chip, so selecting Backends no longer reports "Nothing running" over running model installs. Also: the summary drops a zero clause instead of rendering "0 needs attention" on the happy path and pluralises both counts; a record duration is floored at "< 1s" and rejected above a day, so a zero-value start stamp cannot render a span of millennia and a zero span cannot render "installed in" with nothing after it; a deletion cancelled mid-flight reports the cancellation rather than claiming it was removed; and the retry variant comment names the fix instead of calling the gap closed. Assisted-by: Claude Code:claude-opus-5 [Read] [Edit] [Bash] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * docs: document the Activity page and the operations history endpoints Adds an Activity page under Operations covering the one-line operations strip, the /app/activity sections and filters, per-operation cancel, retry and dismiss, the in-memory 50-entry record, and the sidebar count. Documents GET and DELETE /api/operations/history, and fills the gap in the admin-only endpoint list, which also omitted the pre-existing POST /api/operations/:jobID/dismiss. Corrects the distributed-mode install-watching section: the per-node breakdown now lives on the Activity page rather than on the strip, which rolls a fan-out up into a single phrase. Assisted-by: Claude Code:claude-opus-5 [Read] [Edit] [Bash] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * docs: correct nine details in the Activity page documentation The operations strip never renders a file name: its detail line is the error, the node roll-up, the target node, the phase or the queued note. Drops the stale clause in the distributed-mode section, where the per-node bullet is now the only place a file name is described. Scopes the phase vocabulary to artifact-backed gallery models, since a plain GGUF install emits no phase. Corrects the per-node list: the toggle exists for any fan-out of two or more workers and the four-node threshold only governs whether it starts open, while the N nodes tag needs more than one node. Notes that a cancelled operation can sit in the live section reading Cancelling, that cluster staging never reaches the record, and that Clear history appears only when the record has something in it. Names the operations response envelope, with a JSON example, so callers do not index a bare array, and stops describing the icon-only dismiss control as a labelled button. Assisted-by: Claude Code:claude-opus-5 [Read] [Edit] [Bash] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * docs: drop the unreachable Cancelling state and scope the byte claims An operation can only report isCancelled while it is unprocessed, but every writer of Cancelled sets Processed in the same breath, on the peer path as much as the local one, and the cache evicts cancelled entries before the handler sees them. The state cannot reach the page, so the live section is described again as running or queued operations. Byte counts come from the artifact bridge alone, the same producer as the phase, so a plain GGUF install, a removal and a backend install report none. Scopes both to artifact-backed gallery models and leaves the verb, the name and the percentage as what every operation shows. A worker backend install reports its bytes through fields the operations payload does not carry, so the distributed section now describes the percentage and the node roll-up, with per-file counts pointed at the per-node detail. Also: staging jobs carry no error, so they never reach Needs attention and Retry never had a staging case to exclude; an install that involves workers is no longer called node-scoped, which this page uses for node-targeted installs; and the record timestamps carry nanoseconds. Assisted-by: Claude Code:claude-opus-5 [Read] [Edit] [Bash] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * docs: state only the verb and the name as unconditional on the strip The percentage is as conditional as the bytes were: it renders only for a running operation that has reported progress, so a queued operation, a failed one and a removal never carry it. A removal in particular sits at progress zero for its whole visible life, since the delete path reports none and its completion is filtered out. Both the strip and the card paragraphs now lead with what always shows and list the rest as conditions. The Cluster chip matches on a node list that finished operations do not carry, so a fan-out install leaves the chip once it reaches the record. Scoped that claim to the live sections. Two more of the same shape, found by re-reading each clause alone: the strip also appears for a failure, which is not running, and the four-second hold only applies when nothing replaces the operation that just finished. Assisted-by: Claude Code:claude-opus-5 [Read] [Edit] [Bash] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(ui): stop reporting a cancelled install as installed, and make queued real Three defects that all trace to one root cause: `isCancelled: true` is unreachable from /api/operations. Every writer of Cancelled=true also sets Processed=true, the handler skips Processed && Cancelled, and OpCache.GetStatus evicts a cancelled op before the handler iterates it. Cancelling the last running operation put a green "Installed model X" on the strip for four seconds: the completion hold was guarded by `!previous.isCancelled`, which is dead. A cancellation deletes the operation server side, so the strip sees exactly what it sees on a completion, and nothing in the payload separates the two. The signal now comes from the side that issued the cancel: the operations context remembers the job IDs it cancelled (pruned after a minute) and the strip asks before it holds anything. A cancelled operation goes as soon as it stops; the record already reports it as cancelled. isQueued was set only when the gallery status was missing, but markQueued publishes a "queued" status at admission, so a queued op has a status for its whole queued life and the state was unreachable outside a microsecond window. Every operation waiting behind a running install rendered as "Installing model X" with a spinner. The queued phase is now the signal, via an exported PhaseQueued and a nil-safe OpStatus.IsQueued() next to the writer. With those two fixed, the Cancelling state has no way to be entered: cancelling is instantaneous from the API's point of view. Its branches, CSS, locale key and the isCancelled field itself are removed rather than left for a future reader to assume they work. Assisted-by: Claude Code:claude-opus-5 [Read] [Edit] [Bash] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(ui): keep a removal a removal, and say what an install is doing OpStatus.Deletion was set once, at admission, and lost on the next status write: UpdateStatus replaces the whole status and only carried Nodes forward. Every later writer (the worker's first write, the progress ticks, the failure path) leaves the field at its zero value, so the flag survived only the queued window, and both surfaces test isQueued first. The reachable consequence is that a failed removal reported itself as a failed install, which is exactly the shape the Activity page offers Retry for, and Retry installs: pressing it on a removal that failed re-downloaded the model. A running delete also rendered as "Installing model X" with a spinner, and a successful one as "Installed model X". Carry Deletion forward the way Nodes already is. A job is a delete or an install for its whole life; an unset flag means "no new information", not "this is an install". Pinned by Go specs on both the service and /api/operations: the existing Playwright specs were green only because they stubbed a payload the server could not emit. Also restore the operation's own status message on the Activity card. Phases and byte counters exist only on the managed-artifact path, so a legacy files: gallery model and every backend install rendered a sub-row with nothing in it but the verb. The strip stays terse on purpose. And give the strip's name a min-width floor: overflow: hidden zeroes its automatic minimum, so a long error squeezed the name down to "mod…" and the identity of the thing that broke was the first thing lost. primaryOperation is made module-private: its comment claimed the Activity page selected the same operation, but that page shows all of them, partitioned into failed and running, and never imported it. Assisted-by: Claude Code:Opus 5 [Read] [Edit] [Bash] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(activity): read the operations record from PostgreSQL The Activity page's record of finished installs and removals was a 50-entry in-memory ring per frontend replica. In distributed mode that is the wrong place for it: each replica keeps its own copy, a replica added by a scale-out or a rolling deploy starts empty and never backfills, and "Clear history" clears only the replica that served the request, so the record reappears on the next poll routed elsewhere. The data is already in gallery_operations. Read it from there. GalleryStore gains ListTerminal and ClearTerminal, sharing a lifted terminalStatuses set with CleanOld so there is one definition of "finished". ListTerminal orders by updated_at, when the operation reached its terminal status, because the record reports what finished and when. OpCache.History and ClearHistory dispatch on whether a store is wired, so the HTTP handlers and the OpRecord JSON shape are unchanged and the page needed no change. A failed store read falls back to the local ring rather than blanking the page, and ClearHistory empties the ring as well so a database blip cannot resurrect a record the admin just cleared. The name derivation in recordTerminal is lifted into operationDisplayName and used by both paths, so the ring and the store cannot name the same operation differently. Also fixes a pre-existing bug the store path made visible: the backend channel hardcoded op_type "backend_install" even for a removal, while the model channel derives model_install/model_delete from op.Delete. Both channels carry the same ManagementOp, whose Delete field the backend handler already branches on, so the backend channel now derives backend_delete the same way. Assisted-by: Claude Code:claude-opus-5 [Read] [Edit] [Bash] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(activity): keep a cancelled operation cancelled, and report a failed clear Review follow-up on the store-backed Activity record. A cancelled install was recorded as a failure. The cancel handler persists "cancelled" synchronously, then the handler goroutine unwinds with the context error and Start hands that to updateError unconditionally, which overwrote the row with "failed: context canceled". The page rendered a cancelled install as a red failure card offering Retry, with a raw context error as the reason. Fixed in GalleryStore rather than in Start, because an operation finishes once and the paths that retire one are not mutually exclusive: UpdateStatus now refuses to rewrite a row that already reached a terminal status. That also pins updated_at to when the operation really finished, which is the key the record is ordered by, and Create's upsert now freezes the same columns so a worker dequeuing an operation the admin cancelled while it was queued cannot reopen it as pending. ClearHistory returned nothing, so a failed delete logged a warning while the handler still answered 200. The admin watched the record clear and come back on the next fetch with nothing said about why. It now returns the error, the DELETE handler answers 500, and the store is cleared before the local ring so a failure leaves the fallback record intact rather than faking an empty one. Hydrate is the only reader that decides from op_type whether an operation is a removal, and it tested for "model_delete" exactly, so the backend_delete added in the previous commit hydrated as an install: a replica restarting during a backend removal rendered "Installing backend X". Both discriminations now go through IsDeleteOpType/IsBackendOpType so a fifth op_type cannot silently read as an install in whichever consumer was missed. Also: the backend channel now persists Cancellable as !op.Delete, matching the model channel; IsBackend falls back to the op_type prefix, since is_backend_op is only written by UpsertCacheKey and the rows needing the name fallback were reporting backend operations as models; and an unrecognized terminal status is logged rather than quietly filed as a success, which is what the comment already claimed. Assisted-by: Claude Code:claude-opus-5 [Read] [Edit] [Bash] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(activity): keep a reaped operation correctable by its real outcome The terminal-status freeze added in the previous commit was too wide. It froze "failed" alongside "completed" and "cancelled", and the stale reaper writes "failed" onto operations that are still going to run. The gallery worker is a single goroutine consuming both channels serially, so an operation queued behind a large download sits in "pending" with nothing bumping updated_at, and ReapStaleOperations gives up on it after 30 minutes. That used to be self-healing: the worker dequeued it, Create reset the row to "pending", and the operation reported its real outcome. With the freeze the row stayed "failed" forever while the install ran and succeeded underneath it: a red failure card offering Retry for a model that is installed, omitted from ListActive so no replica hydrates it, and no longer deduped cluster-wide by FindDuplicate. Freeze on ("completed", "cancelled") instead. That is all the cancelled-install fix ever needed, and it leaves a failure correctable by what actually happened. The set is separate from terminalStatuses, which ListTerminal, ClearTerminal and CleanOld all still want in full, because the two mean different things: a failure can be superseded by a real outcome, a completion or a cancellation is the real outcome. UpdateStatus now writes the error column unconditionally, so a corrected outcome drops the previous attempt's reason rather than being recorded as completed while still carrying "stale operation reaped" as its error. Also adds the route-level spec for the 500 branch of DELETE /api/operations/history, and trims a comment that credited the persisted cancellable column with more than it survives long enough to do. Assisted-by: Claude Code:claude-opus-5 [Read] [Edit] [Bash] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(activity): offer Cancel in the phase that can honour it The cancellable flag was set at both ends of an operation's life and was wrong at both, in opposite directions. A queued operation is cancellable whatever it is. EnqueueModelOp and EnqueueBackendOp select on the operation context, so cancelling one that is still waiting releases the delivery goroutine and abandonQueued retires it: the worker never sees it, nothing is downloaded, nothing is deleted. markQueued nevertheless wrote Cancellable: !deletion, so a queued removal reported cancellable: false and the UI hid the Cancel button in the one window where pressing it both works and leaves no trace. A removal queued behind a large install was stuck there until the install finished. A running removal is not cancellable at all. DeleteModel and DeleteBackend take no context, and modelHandler only checks the operation context after the call returns, so a "cancelled" verdict would land after the model was already gone. Both handlers nevertheless wrote Cancellable: true unconditionally at entry, ahead of the op.Delete branch, offering a Cancel button the server cannot honour. So the queued phase is more cancellable than the running phase, which is the reverse of the usual shape. markQueued now reports true unconditionally, and the handler-entry writes report !op.Delete. Both sites carry a comment saying why, because reading either one alone suggests the other is a bug. GalleryStore.Create keeps !op.Delete: it runs at dequeue, so its value already describes the running phase. Its comment now says so. Specs cover queued removal, queued install, running removal and running install through the handlers, plus the queued-removal case through /api/operations where the flag is consumed, plus the behaviour the whole asymmetry rests on: a removal cancelled while queued never reaches the worker and deletes nothing. No existing spec asserted the old values. Assisted-by: Claude Code:claude-opus-5 [Read] [Edit] [Write] [Bash] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(activity): clamp the installer message, and add a real-binary e2e spec Running the page against a real local-ai showed the legacy installer message wrapping to three lines and dominating the card: it embeds an absolute file path, so it is both long and a single unbreakable token. One line, ellipsised, full text in the title, matching what the error string already does. The spec that found it runs with no route stubbing at all. Every other spec here stubs /api/operations, which is how a payload the server cannot emit (isDeletion true on a live operation) stayed green through a full review while the UI rendered a removal as an install. It is skipped unless LOCALAI_REAL_BINARY is set, so CI is unaffected. Assisted-by: Claude Code:claude-opus-5 [Read] [Edit] [Bash] 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> |
||
|
localai-org-maint-bot
|
823fc25bb7 |
fix(kokoro): add CPU backend fallback (#11161)
Publish the existing Kokoro CPU profile for amd64 and arm64 and use it as the default gallery capability so Vulkan-only and CPU hosts can install the backend. Assisted-by: Codex:gpt-5 Co-authored-by: localai-org-maint-bot <306269227+localai-org-maint-bot@users.noreply.github.com> |
||
|
mudler's LocalAI [bot]
|
366db11c59 |
chore: ⬆️ Update mudler/parakeet.cpp to 3e1ddd8455ceb9bfae564f84db24ba068b00c56e (#11150)
⬆️ Update mudler/parakeet.cpp Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: mudler <2420543+mudler@users.noreply.github.com> |
||
|
mudler's LocalAI [bot]
|
54012002fd |
chore: ⬆️ Update ikawrakow/ik_llama.cpp to 5f063b7bbae8f9a34dfc5c704aa77939e76494a9 (#11153)
⬆️ Update ikawrakow/ik_llama.cpp Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: mudler <2420543+mudler@users.noreply.github.com> |
||
|
mudler's LocalAI [bot]
|
176000190e |
chore: ⬆️ Update ggml-org/llama.cpp to 1cbfd1988311775425d36c0ce066590f7d3049cf (#11155)
⬆️ Update ggml-org/llama.cpp Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: mudler <2420543+mudler@users.noreply.github.com> |
||
|
mudler's LocalAI [bot]
|
c4d0c060ed |
chore: ⬆️ Update CrispStrobe/CrispASR to 7bb8be77a8c1677e32bba58514bb2d42f29a7a48 (#11156)
⬆️ Update CrispStrobe/CrispASR Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: mudler <2420543+mudler@users.noreply.github.com> |
||
|
mudler's LocalAI [bot]
|
cff31bbac0 |
chore: ⬆️ Update leejet/stable-diffusion.cpp to 22516991cbdf725e69b0b4a87e52ca16cce07c2d (#11157)
⬆️ Update leejet/stable-diffusion.cpp Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: mudler <2420543+mudler@users.noreply.github.com> |
||
|
mudler's LocalAI [bot]
|
e218c7f56a |
chore(model-gallery): ⬆️ update checksum (#11152)
⬆️ Checksum updates in gallery/index.yaml Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: mudler <2420543+mudler@users.noreply.github.com> |
||
|
mudler's LocalAI [bot]
|
12f2e1b99c |
feat(swagger): update swagger (#11149)
Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: mudler <2420543+mudler@users.noreply.github.com> |
||
|
mudler's LocalAI [bot]
|
2ecf893c6c |
fix(sherpa-onnx): install cuDNN in the CUDA builder so the package can bundle it (#11145)
sherpa-onnx links onnxruntime's CUDA execution provider, and libonnxruntime_providers_cuda.so carries cuDNN as a hard DT_NEEDED. The onnxruntime GPU tarball ships no cuDNN of its own, and Dockerfile.golang only installs libcudnn9 on the arm64 + CUDA 13 branch, so the amd64 CUDA builders have none at all. Since #10946 added the packaging guard, that combination is fatal rather than silent: package-gpu-libs.sh reports 'cuDNN: venv=absent system=absent -> bundle=detect', correctly detects the reference, finds nothing to copy and refuses to emit the package. Both -gpu-nvidia-cuda-12-sherpa-onnx and -gpu-nvidia-cuda-13-sherpa-onnx have failed to build since 2026-07-19, so neither image has been published. Before the guard existed they shipped without cuDNN and failed at load time instead. Install the runtime package for this backend only. The auto-detection bundles solely what a package references, so no other backend would grow, but every Go CUDA builder would pay ~1.1 GB of layer and registry cache for a library ggml never calls. Assisted-by: Claude Code:claude-opus-5 [Read] [Edit] [Bash] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Co-authored-by: Ettore Di Giacinto <mudler@localai.io> |
||
|
mudler's LocalAI [bot]
|
05ff401de8 |
fix(test): stop the backend-trace specs racing the lossy trace channel (#11146)
RecordBackendTrace does a non-blocking send onto a 100-slot channel and
drops when it is full, so tracing never stalls inference. The payload
bounding specs pushed all 200 traces in one tight loop, which overruns
that channel on a loaded machine: entries are dropped for good and the
Eventually waiting for 200 can never be satisfied, no matter the timeout.
CI hit this on master at
|
||
|
Tai An
|
a7fa678d83 |
fix(tts): forward the OpenAI speed field to the backend (#11097) (#11120)
* fix(tts): forward the OpenAI speed field to the backend (#11097) /v1/audio/speech accepted the documented OpenAI `speed` field and then dropped it: schema.TTSRequest had no Speed member, so the value never reached proto.TTSRequest and the request returned 200 with an unchanged playback rate. Accept speed and normalise it into the existing per-request params map, which core/backend forwards verbatim to the backend. An explicit params["speed"] still wins, and a value outside the documented 0.25-4.0 range is now rejected with 400 instead of being silently ignored. Signed-off-by: Anai-Guo <antai12232931@outlook.com> * fix(tts): distinguish explicit speed=0 from an omitted field Make TTSRequest.Speed a *float32 so an explicit `"speed": 0` (invalid, below the documented 0.25 minimum) is rejected with 400 instead of being treated as unset and silently defaulted. An omitted field stays nil and leaves the backend default untouched. Add a request-boundary regression that distinguishes an omitted speed from an explicit zero, addressing review feedback. Signed-off-by: Anai-Guo <antai12232931@outlook.com> * docs: drop the speed field from the TTS docs Per review: no backend consumes params.speed today, so documenting it would be misleading. The API-level plumbing and validation stay. Signed-off-by: Anai-Guo <antai12232931@outlook.com> --------- Signed-off-by: Anai-Guo <antai12232931@outlook.com> |
||
|
Anupam Mediratta
|
3698361510 |
fix: upgrade hono to 4.12.25 (CVE-2026-54290) (#11023)
* fix: CVE-2026-54290 security vulnerability Automated dependency upgrade by OrbisAI Security Signed-off-by: orbisai0security <mediratta@gmail.com> Signed-off-by: Anupam Mediratta <mediratta@gmail.com> * fix(deps): override hono transitive dep to eliminate CVE-2026-54290 Add package.json `overrides` field to force hono@4.12.25 across the entire dependency graph, including the transitive copy pulled in by @modelcontextprotocol/sdk. Previously bun.lock retained a scoped `@modelcontextprotocol/sdk/hono` entry resolved to the vulnerable hono@4.12.8; the override removes that entry so only the patched version ships. Assisted-by: Claude Code:claude-sonnet-4-6 Signed-off-by: Anupam Mediratta <mediratta@gmail.com> --------- Signed-off-by: orbisai0security <mediratta@gmail.com> Signed-off-by: Anupam Mediratta <mediratta@gmail.com> |
||
|
mudler's LocalAI [bot]
|
856b0ea951 |
fix(ci): build the CUDA 13 image on Ubuntu 24.04 (#11143)
* fix(ci): build the CUDA 13 image on Ubuntu 24.04 The amd64 `-gpu-nvidia-cuda-13` image is the only runtime image still built FROM ubuntu:22.04. The Ubuntu 24.04 migration (#7769) bumped its `ubuntu-version` to 2404 but left `base-image` on jammy, so the image ships glibc 2.35 while adding the noble CUDA apt repository, and every backend it unpacks is built on noble. Backends therefore cannot dlopen the libraries they bundle. The vLLM backend dies at import time with: OSError: /lib/x86_64-linux-gnu/libc.so.6: version `GLIBC_2.38' not found (required by /backends/cuda13-vllm/lib/libnuma.so.1) and torchcodec finds no usable libavutil because jammy ships ffmpeg 4.x (libavutil.so.56) while torchcodec looks for .so.57 through .so.60. Add a spec over the build matrices that fails when a base image and the `ubuntu-version`/`ubuntu-codename` it is paired with disagree, or when the runtime images are split across Ubuntu releases. Entries whose base image does not name a release (JetPack) are left alone. The `base-grpc-cuda-13-amd64` builder base stays on jammy: it only compiles backends, and a lower glibc floor in a builder is safe. Fixes #11059 Assisted-by: Claude:claude-opus-5 golangci-lint Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(ci): drop the build matrix invariant spec Per review, the CI matrix guard does not belong in the tree. Only the base image bump remains. Assisted-by: Claude:claude-opus-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> |
||
|
mudler's LocalAI [bot]
|
878a0d00a1 |
fix(distributed): reaper reaps live backends, ghost model stubs, in_flight leak, sidecar staging runaway (#11142)
* fix(distributed): stop the probe reaper from orphaning busy backends
The reconciler's liveness probe is a 1s gRPC HealthCheck, and a single
failed probe deleted the model's node_models row. A backend that is
merely busy cannot answer it: single-threaded Python backends (video and
avatar generation) block for minutes inside one request, so the reaper
was deleting registry rows for backends that were alive and mid-request.
The model then vanished from the nodes page while it was still
generating, and because the row was gone the in-flight decrement had
nothing to decrement ("DecrementInFlight: no matching row or already
zero"). Every subsequent request re-routed and re-staged the full model
from scratch.
Two guards:
- Replicas with in-flight requests are excluded in SQL. A row that is
actively serving is proof of life, and the running request is
exactly what stops the backend from answering the probe.
- Idle replicas must miss three CONSECUTIVE probes before removal, so
a transient blip cannot orphan a live replica. A successful probe
resets the streak.
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Assisted-by: Claude:claude-opus-5 [Claude Code]
* fix(distributed): drop the local model stub when its last replica goes
In distributed mode every routed model leaves an in-process stub in the
frontend's ModelLoader, and DistributedModelStore.Range reports local
stubs UNION the registry rows. Every registry removal path deletes only
the DB row, so the stub outlived the replica and the model was reported
as loaded forever.
That is the "loaded on the home page, absent from every node" ghost:
/system reads the union and still sees the stub, while /api/nodes/models
reads the registry and correctly sees nothing. It never self-healed,
and both frontend replicas showed it independently.
The replica-removed chokepoint could not fix this as it stood, because
it held a SINGLE hook that the prefix cache already owned, and it was
registered only when the prefix cache was enabled. Registering a second
listener would have silently displaced the first.
- Turn replicaRemovedHook into a list (AddReplicaRemovedHook), so
independent subsystems can each register without displacing others.
- Add NewLocalStubInvalidator, which drops the local stub once no
healthy replica of the model remains anywhere in the cluster, and
wire it unconditionally in startup.
The stub is kept while another node still serves the model: the
frontend is right to consider it loaded, and each request re-routes
through SmartRouter to pick a live replica anyway.
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Assisted-by: Claude:claude-opus-5 [Claude Code]
* fix(distributed): stop staging checksum sidecars back to workers
The file transfer server writes a "<file>.sha256" sidecar next to every
file it accepts. The sender walked the model directory with no filter,
so it staged those sidecars too, and the receiver duly wrote a sidecar
for each sidecar. Every staging pass multiplied the tree:
config.json -> config.json.sha256 -> config.json.sha256.sha256 -> ...
One LongCat snapshot had grown to 498 files, 466 of them chained, up to
29 levels deep, and the staged file count climbed on every pass. This
inflates each transfer and grows disk without bound on both ends.
Skip hash sidecars in stageDirectory, and mirror the skip in
countStageableFiles so the progress bar still reaches 100%. The check is
"a sidecar sitting next to a real file" rather than a blanket suffix
ban, so a model that genuinely ships a .sha256 payload with no
corresponding base file is still transferred.
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Assisted-by: Claude:claude-opus-5 [Claude Code]
* fix(distributed): classify the liveness probe instead of gating on in_flight
The previous commit excluded replicas with in-flight requests from the probe
reaper. That was the wrong guard, and could invert the bug it fixed.
in_flight has no decrement guarantee: track() balances its increment with a
defer, but a frontend killed mid-request never runs it, and the load-time
reservation is released only when the first inference completes. Nothing
resets a leaked counter. Gating the reaper on it therefore meant a leaked
counter would shield a genuinely dead replica from ever being reaped.
Nor was patience alone a fix: three misses at the default interval is ~90s of
silence, while the generation that triggered this blocks for 15+ minutes.
The real conflation was in the probe itself. A gRPC HealthCheck against the
backend's serving port measures "is it idle enough to answer", not "does the
process exist", and probeLoadedModels discarded the error that tells them
apart. Because the gRPC client is lazy, the status code is decisive:
- DeadlineExceeded: transport fine, nothing serviced the RPC. Busy.
- Unavailable: nothing is listening. Gone.
ModelProber now returns a ProbeOutcome, and only ProbeUnreachable counts
toward the reap threshold. ProbeBusy clears the streak: it is evidence of
life. A blackholed network reads as busy too, deliberately, since whole-node
failure is the health monitor's job.
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Assisted-by: Claude:claude-opus-5 [Claude Code]
* feat(distributed): reconcile replicas against worker-reported processes
Probing a backend's own serving port cannot distinguish "busy" from "gone"
without inferring it from an error code. The worker can answer directly: it
spawned the process, holds the handle, and its reply is not blocked by
whatever that backend is doing.
Adds a models.running request-reply subject. The worker answers out of its
in-memory process table, reporting each live process as (modelID,
replicaIndex, address) — the supervisor's process keys are `modelID#replica`,
which is isomorphic to a NodeModel row, so the reconciler can diff the two
directly.
reconcileNodeProcesses runs before the port probe and reaps rows for models
the worker is not running. Models the worker vouches for get updated_at
bumped, which takes them out of the port prober's stale set entirely: that is
what keeps a backend deep in a long generation away from the probe in the
first place, rather than relying on classifying its silence after the fact.
A worker that does not answer is skipped, not assumed empty. A messaging
failure says nothing about the processes, and assuming the worst would delete
a node's rows on a transient NATS blip; the port probe stays as the fallback
for those nodes. Rows younger than probeStaleAfter are ignored so a freshly
created row is never judged against a process table that has not caught up.
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Assisted-by: Claude:claude-opus-5 [Claude Code]
* fix(distributed): stop in_flight leaking and pin replicas against eviction
A leaked in_flight counter is not cosmetic. FindLRUModel,
FindGlobalLRUModelWithZeroInFlight and the router's eviction query all require
in_flight = 0, so a replica whose counter never came back is pinned and its
VRAM is unreclaimable for the lifetime of the process.
Two halves.
The source: routing reserves in_flight = 1 at load time so a freshly loaded
replica is not evicted out from under the request that caused the load. That
reservation was released ONLY by the first inference completing, so a route
torn down before any inference ran (client disconnect, handler error, failure
between load and the backend call) stranded it. newRouteResult now wires the
reservation to a sync.Once fired by whichever comes first, the first inference
or route teardown, and replaces three copies of the old wiring.
The backstop: a sweeper for counters leaked by paths that cannot run a defer
at all, such as a frontend killed mid-request.
Identifying a leak by elapsed time alone is unsafe. IncrementInFlight stamps
last_used at request START and nothing moves it while the request runs, so a
long generation is indistinguishable from a leak by age, and resetting there
would expose a serving model to eviction. The probe supplies the missing bit:
a backend that answers a health check promptly is not inside a request,
because that is precisely what a busy one cannot do. Requiring the row to also
be idle for 30 minutes covers backends that serve in parallel and can answer
while working, since those keep last_used fresh through each new increment.
Two existing tests asserted the old behaviour ("No decrement on Release").
That assertion was the leak, so both now pin the release instead.
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Assisted-by: Claude:claude-opus-5 [Claude Code]
---------
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Co-authored-by: Ettore Di Giacinto <mudler@localai.io>
|
||
|
mudler's LocalAI [bot]
|
e9c2754fc2 |
chore(model-gallery): propose variant groupings for review (#11139)
chore(model-gallery): propose variant groupings Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: mudler <2420543+mudler@users.noreply.github.com> |
||
|
mudler's LocalAI [bot]
|
0a8a7fbbb4 |
chore(llama-cpp): bump llama.cpp and adapt to the load-mode refactor (#11140)
Bump LLAMA_VERSION to 0d47ea7427463093e69128bf2c2f9cd06b3ee5b3 (73 commits touching common/, src/ and tools/server/). Two upstream changes break the backend: * ggml-org/llama.cpp#20834 folded common_params::use_mmap / use_mlock / use_direct_io into a single `load_mode` enum. LocalAI still exposes the three as independent settings (`mmap`, `mmlock`, and the `direct_io` option), so params_parse folds them once all three have been read, keeping the precedence the separate booleans had: direct I/O bypasses the page cache, mlock implies mmap, everything off is a plain buffered read. turboquant and bonsai compile this same grpc-server.cpp against forks that predate the refactor, so prepare.sh probes the checkout for LLAMA_LOAD_MODE_MMAP and generates llama_compat.h with LOCALAI_LEGACY_LOAD_MODE set accordingly. Probing beats a per-fork build flag here because the fork flavor targets disagree on whether they forward CMAKE_ARGS or EXTRA_CMAKE_ARGS, and it heals itself once a fork rebases past the refactor. * The MiniMax M3 patch no longer applies. Upstream merged the model half of llama.cpp#24523 (LLM_ARCH_MINIMAX_M3, src/models/minimax-m3.cpp, the gguf-py constants and conversion/minimax.py) but not the chat half, so the patch is re-cut to carry only the common/chat.cpp template detection and PEG parser, rebased onto the new pin and onto the thinking_end_tag -> thinking_end_tags rename. Dropping it wholesale (as #11008 did, reverted in #11136) would have silently regressed MiniMax M3 tool calling and thinking. Verified with a CPU docker build of the backend plus LoadModel and Predict against a real GGUF over gRPC in all four load modes. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Co-authored-by: Ettore Di Giacinto <mudler@localai.io> |
||
|
Ettore Di Giacinto
|
d9f3007876 |
Revert "chore: ⬆️ Update ggml-org/llama.cpp to d2a818231effb12b7b20b80b3b8c7756a9a33a04" (#11136)
Revert "chore: ⬆️ Update ggml-org/llama.cpp to `d2a818231effb12b7b20b…"
This reverts commit
|
||
|
mudler's LocalAI [bot]
|
6e69dbd617 |
chore: ⬆️ Update ggml-org/llama.cpp to d2a818231effb12b7b20b80b3b8c7756a9a33a04 (#11008)
* ⬆️ Update ggml-org/llama.cpp Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> * fix(llama-cpp): drop upstreamed MiniMax M3 patch The pinned llama.cpp revision already contains MiniMax M3 support, so the downstream patch rejects during backend preparation on every platform. Assisted-by: Codex:gpt-5 --------- Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: mudler <2420543+mudler@users.noreply.github.com> Co-authored-by: localai-org-maint-bot <306269227+localai-org-maint-bot@users.noreply.github.com> |
||
|
mudler's LocalAI [bot]
|
19ffc33011 |
chore: ⬆️ Update leejet/stable-diffusion.cpp to 2d0385ba85af358f7115dda608a63eafd9de7ffd (#11132)
⬆️ Update leejet/stable-diffusion.cpp Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: mudler <2420543+mudler@users.noreply.github.com> |
||
|
mudler's LocalAI [bot]
|
76ce4f59b6 |
chore: ⬆️ Update vllm-metal (darwin) to v0.3.0.dev20260726174827 (#11133)
⬆️ Update vllm-project/vllm-metal (darwin) Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: mudler <2420543+mudler@users.noreply.github.com> |
||
|
dependabot[bot]
|
4f883c86a9 |
chore(deps): bump postcss from 8.5.15 to 8.5.23 in /core/http/react-ui in the npm_and_yarn group across 1 directory (#11106)
chore(deps): bump postcss Bumps the npm_and_yarn group with 1 update in the /core/http/react-ui directory: [postcss](https://github.com/postcss/postcss). Updates `postcss` from 8.5.15 to 8.5.23 - [Release notes](https://github.com/postcss/postcss/releases) - [Changelog](https://github.com/postcss/postcss/blob/main/CHANGELOG.md) - [Commits](https://github.com/postcss/postcss/compare/8.5.15...8.5.23) --- updated-dependencies: - dependency-name: postcss dependency-version: 8.5.23 dependency-type: indirect dependency-group: npm_and_yarn ... Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> |
||
|
mudler's LocalAI [bot]
|
c56373d772 |
chore(model-gallery): ⬆️ update checksum (#11134)
⬆️ Checksum updates in gallery/index.yaml Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: mudler <2420543+mudler@users.noreply.github.com> |
||
|
Tai An
|
53006bb8e1 |
fix(realtime): accept legacy 'modalities' alias for output_modalities (fixes #11103) (#11104)
* fix(realtime): accept legacy 'modalities' alias for output_modalities OpenAI's Realtime *beta* used the field name `modalities`; the GA field is `output_modalities`. LocalAI only binds `output_modalities`, so a client sending the still-common beta field `modalities: ["text"]` has it silently dropped by encoding/json and the session falls back to audio: TTS runs and the client receives large response.output_audio.* frames even though it asked for text-only. Accept `modalities` as an alias on both session.update (RealtimeSession) and response.create (ResponseCreateParams). The GA `output_modalities` wins when both are present, so GA clients are unaffected. Applied at the two existing resolution points via a small modalitiesWithAlias helper. Fixes #11103 Signed-off-by: Anai-Guo <antai12232931@anaiguo.com> * test(realtime): add JSON-boundary regression for modalities alias Decode representative session.update and response.create payloads that carry only the legacy beta `modalities` key and assert the effective output modality resolves to text (not audio), reproducing the exact expressions used in updateSession and triggerResponseAtTurn. This guards against a wrong JSON tag or a missed call site letting encoding/json drop the alias silently. Also document output_modalities (and the accepted legacy modalities alias) for text-only sessions in the realtime feature docs. Signed-off-by: Tai An <antai12232931@outlook.com> --------- Signed-off-by: Anai-Guo <antai12232931@anaiguo.com> Signed-off-by: Tai An <antai12232931@outlook.com> Co-authored-by: Anai-Guo <antai12232931@anaiguo.com> |
||
|
mudler's LocalAI [bot]
|
62e3b8304e |
chore: ⬆️ Update CrispStrobe/CrispASR to 306faee45fab641d54f9f941f075de1e9c0d3278 (#11131)
⬆️ Update CrispStrobe/CrispASR Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: mudler <2420543+mudler@users.noreply.github.com> |
||
|
mudler's LocalAI [bot]
|
2476509321 |
chore: ⬆️ Update ikawrakow/ik_llama.cpp to 0a4e10c7fb65d2dd5a4afb78339c7d373a8cdfaa (#11128)
⬆️ Update ikawrakow/ik_llama.cpp Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: mudler <2420543+mudler@users.noreply.github.com> |
||
|
dependabot[bot]
|
ac9352ef54 |
chore(deps): bump actions/setup-node from 4 to 7 (#11080)
Bumps [actions/setup-node](https://github.com/actions/setup-node) from 4 to 7. - [Release notes](https://github.com/actions/setup-node/releases) - [Commits](https://github.com/actions/setup-node/compare/v4...v7) --- updated-dependencies: - dependency-name: actions/setup-node dependency-version: '7' dependency-type: direct:production update-type: version-update:semver-major ... Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> |
||
|
mudler's LocalAI [bot]
|
4baa36ddd8 |
feat(backend): vllm-cpp - text-generation backend for vllm.cpp with llama.cpp-parity tool calling (#11100)
* feat(backend): add vllm-cpp text-generation backend (vllm.cpp) Wrap https://github.com/mudler/vllm.cpp - the LocalAI-team from-scratch C++20 port of vLLM (paged KV cache, continuous batching, prefix caching, safetensors + GGUF loading, no Python at inference) - as a Go gRPC backend over its stable C ABI (ABI v2) via purego. Backend (backend/go/vllm-cpp): - Load -> vllm_engine_load: accepts a .gguf file or a config.json model dir (anything else is refused, satisfying the greedy-probe rule); context_size maps to max_model_len, options block_size/num_blocks/max_num_seqs size the KV cache and scheduler admission. - Predict -> vllm_complete (blocking); PredictStream -> vllm_complete_stream with the per-delta C callback bridged into the gRPC stream. The backend embeds base.Base (not SingleThread): concurrent requests batch continuously in the engine's shared AsyncLLM scheduler. - PredictOptions.Grammar -> the ABI's structured_grammar (GBNF), giving grammar-constrained tool calling at parity with llama-cpp; the ABI also exposes JSON-schema/regex/choice constraints. - Hand-mirrored POD structs with layout locked by unit tests (unsafe.Offsetof vs the C offsets) and a runtime vllm_abi_version gate. - One portable library per platform (vllm.cpp uses per-file SIMD tiers with runtime dispatch), so no avx/avx2/avx512 variant builds. Wiring: - backend-matrix: CPU amd64+arm64 (per-arch + manifest merge), CUDA 12/13 amd64 (120a;121a Blackwell fat binary), L4T arm64 (121a, GB10/DGX Spark - the runtime-proven GPU target), Vulkan amd64, and Darwin arm64 Metal. - backend/index.yaml meta + 12 image entries (latest/development x cpu, cuda12, cuda13, l4t, vulkan, metal); bump_deps registration for the VLLM_CPP_VERSION pin; root Makefile registration; test-extra runs the unit specs (pure Go, no engine build). - Importers: preference-only swaps - llama-cpp (GGUF) and vllm (safetensors) advertise vllm-cpp via AdditionalBackends and emit backend: vllm-cpp without tokenizer templating (the C ABI takes the FINAL prompt; templating and tool parsing stay LocalAI-side). No auto-detect importer. - Docs: backends list, top-level README maintained-engines table, compatibility table. Verified: 20/20 Ginkgo specs against the real pinned engine and Qwen3.5-2B-UD-Q8_K_XL.gguf on CPU - blocking + streaming parity, greedy determinism, stop words, GBNF-constrained generation, and 4 concurrent streams; plus a dlopen/ABI-gate smoke of the built gRPC server binary. Upstream ABI v2 + production structured-output wiring landed as mudler/vllm.cpp@86013f3. Assisted-by: Claude Code:claude-fable-5 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(vllm-cpp): ride the autoparser code path - engine-side chat templating and tool engagement (ABI v3) The backend now implements AIModelRich (PredictRich / PredictStreamRich) over vllm.cpp's ABI v3 chat entry points, so chat and tool calling ride the SAME code path as the llama.cpp autoparser: the ENGINE renders the model's chat template, decides when a tool call engages, and parses it - LocalAI receives pre-parsed ChatDelta / ToolCallDelta protos exactly as it does from llama-cpp. - With use_tokenizer_template + structured Messages, PredictOptions lowers to ONE OpenAI chat request JSON (messages, tools, tool_choice, sampling, stream_options.include_usage) for vllm_chat / vllm_chat_stream. tool_choice auto lowers engine-side to a LAZY structural-tag decode constraint - free text until the model emits the tool trigger, then the call is grammar-constrained; required/named force a call. Tool output is parsed by the engine's streaming Hermes-style parser; each chat.completion.chunk maps onto ChatDeltas (content / reasoning_content / tool_calls) which the host already prefers over Go-side tag extraction. Without structured messages the plain path (LocalAI templating + optional GBNF grammar) applies unchanged. - The engine resolves the chat template from the GGUF tokenizer.chat_template metadata (or tokenizer_config.json); templates beyond its minja subset - e.g. the full Qwen3.5 namespace()/macro template - degrade engine-side to a Hermes-aware fallback prompt (tools schemas + <tool_call> instruction) with a stderr witness, so structural-tag engagement keeps working. - Importers now emit the same config shape as llama-cpp for vllm-cpp (use_tokenizer_template: true, no-grammar autoparser flow); only the llama-cpp-specific use_jinja option and the vllm-python parser options are dropped. - Pin bumped to mudler/vllm.cpp@aaed7ec (ABI v3 + chat-prompt resolution). Verified against the real engine and Qwen3.5-2B-UD-Q8_K_XL.gguf on CPU: full suite green - blocking chat, streaming deltas concatenating byte-equal to the blocking answer, a REQUIRED tool call returning schema-valid arguments JSON, and an AUTO run where the engine itself engages get_weather and streams parsed tool deltas; plus unit specs for the request lowering, chunk->ChatDelta mapping, and the C struct mirrors (ABI gate now v3). Assisted-by: Claude Code:claude-fable-5 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(vllm-cpp): ABI v5 - engine-side parser selection for 30 tool dialects + reasoning Bump the vllm.cpp pin to the autoparser-parity engine: 30 tool-call dialects (every pure-text parser in the pinned vLLM registry, each ported 1:1 with its upstream tests), 7 reasoning parsers, google/minja as the template renderer (the full Qwen3.5 template now renders engine-side), per-family structural tags (tool_choice required/named compiles the model's NATIVE syntax where expressible), and template auto-detection for both parser axes. Backend changes: - cModelParams mirrors ABI v5 (tool_parser + reasoning_parser fields, layout-locked by the offset tests; ABI gate now v5). - New model options tool_parser:<name> / reasoning_parser:<name> pass through to the engine; unset means template auto-detection (18-row tool marker table; [THINK]->mistral, <think>->think_auto for reasoning); "none" disables the reasoning split; unknown names fail the first chat call. - Chat chunks parse the `reasoning` field (the pin renamed reasoning_content), flowing into ChatDelta.ReasoningContent which the host already prefers. Live e2e against Qwen3.5-2B-UD-Q8_K_XL.gguf on CPU, full suite green: the real chat template renders (no more fallback), reasoning auto-detection picks think_auto so markerless answers stay pure content (the live run caught the deepseek_r1 content-swallow upstream and drove the think_auto fix), required tool_choice returns schema-valid arguments, auto tool_choice engages engine-side and streams parsed deltas, and blocking/streaming stay byte-identical. Turn latency also dropped (proper template EOS behavior). Upstream program landed as mudler/vllm.cpp 86013f3..5fffe7e (ABI v2-v5, minja, parser waves B1/B2/B4, reasoning seam, structural-tag registry, think_auto). Assisted-by: Claude Code:claude-fable-5 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * chore(vllm-cpp): bump the engine pin to the ENG-wave close-out mudler/vllm.cpp@df8909b: the six engine-backed vLLM tool-parser families (qwen3-coder/xml/mimo, kimi_k2, glm45/47, minimax_m2, gemma4, seed_oss) text-reimplemented from their wire formats and held to the upstream test suites - 39 registered dialects; the pinned vLLM registry is now covered except the three Rust/Harmony-backed families, descoped by decision. kimi_k2 also gains a full native structural-tag builder; four new template auto-detection rows land with test-pinned ordering. Full backend e2e re-run green against Qwen3.5-2B-UD-Q8_K_XL.gguf on CPU. Assisted-by: Claude Code:claude-fable-5 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(vllm-cpp): add the vllm-cpp-development gallery meta The gallery grew the twelve latest/development image entries but was missing the separate vllm-cpp-development meta (own capabilities map targeting the -development image names), which every backend ships so the development gallery resolves per-platform. Validated: all capability targets in both metas resolve to existing entries, and every image URI's tag suffix matches a backend-matrix build. Also full-stack verified in this change's context (single-node local-ai from this branch, locally-built backend under --backends-path, Qwen3.5-2B GGUF): /v1/chat/completions non-stream (clean content + usage), streaming (SSE deltas), tool_choice auto engaging get_weather engine-side with schema-valid arguments and finish_reason=tool_calls, and streamed tool-call deltas in the standard name-first cadence. Assisted-by: Claude Code:claude-fable-5 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(vllm-cpp): repair the CI backend builds - gcc-14 -Werror + fat-arch Triton Two distinct failures took down all five vllm-cpp backend builds on the PR: 1. gcc-14 (ubuntu:24.04 CI images; the local toolchain is gcc-13) fails the engine build with -Werror=maybe-uninitialized in InputBatch::condense - a false positive through a staging std::optional's raw storage. Fixed upstream (mudler/vllm.cpp@61f3e85) by moving slot-to-slot directly; verified BOTH ways under dockerized g++-14.2 (unfixed reproduces CI's two diagnostics exactly, fixed compiles clean) with the engine's behavior suites green. Pin bumped to that sha. 2. The amd64 CUDA builds died at CMake configure: the vendored Triton-AOT cubin trees are per-arch and the engine refuses -DVLLM_CPP_TRITON=ON on a multi-arch (120a;121a) fat build unless pinned to one tree, which would be unsound for the other arch. Triton is now enabled only on the single-arch arm64/GB10 build (where the cubins matter); the fat amd64 binary uses the engine's non-AOT GDN path. Backend e2e re-run green at the new pin (Qwen3.5-2B on CPU, full suite). Assisted-by: Claude Code:claude-fable-5 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(vllm-cpp): cuda-12 images cannot compile compute_121a - target 120a only The second CI round surfaced a CUDA-version constraint: the cuda-12 (12.8) image's nvcc rejects 'compute_121a' (GB10 arch support landed with CUDA 13), killing the amd64 cuda-12 build at nvcc. Gate the architecture list on CUDA_MAJOR_VERSION (exported by Dockerfile.golang): cuda-12 builds consumer Blackwell 120a only, cuda-13 keeps the 120a;121a fat binary, arm64/l4t (cuda-13) keeps single-arch 121a with the Triton cubins. GB10 is arm64, so the amd64 cuda-12 image never served it - no capability change. Verified by Makefile dry-run variable dumps for all three combinations (cuda12 -> 120a; cuda13 -> 120a;121a; cpu -> CUDA off). Assisted-by: Claude Code:claude-fable-5 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(vllm-cpp): drop the cuda-12 variant - the engine needs the CUDA 13 toolchain Third CI round, third layer: with the arch list already narrowed to 120a, the cuda-12 (12.8) build still dies in ptxas compiling the sm_120a NVFP4 MMA kernels ("Vector type too large, exceeds 128 bit limit") - the Blackwell fp4 path genuinely requires the CUDA 13 toolchain, and vllm.cpp supports Blackwell-family GPUs only. Shipping a cuda-12 image without the fp4 kernels would be a crippled build of an engine whose whole GPU story is fp4, so the variant is dropped instead: - backend-matrix: cuda-12 vllm-cpp entry removed (cuda-13 amd64, l4t arm64, cpu, vulkan, metal remain). - gallery: cuda12 image entries removed; the nvidia capability now resolves to the cuda13 image in both metas; the nvidia-cuda-12 key is dropped so older-driver hosts fall back to the CPU image instead of an unrunnable one. - backend Makefile: BUILD_TYPE=cublas under CUDA_MAJOR_VERSION=12 now fails fast with a clear message; cuda-13 keeps the 120a;121a fat binary and arm64/l4t keeps 121a with the Triton cubins. Verified: Makefile branch dumps for all four combinations (cuda12 loud error, cuda13 fat, arm64 121a+Triton, cpu off), YAML parses, matrix filter tests green, gallery capability targets all resolve. Assisted-by: Claude Code:claude-fable-5 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(vllm-cpp): forward multi-turn tool identity and reasoning to the engine chatRequestJSON dropped Message.ToolCallId and Message.Name on role="tool" replies and Message.ReasoningContent on assistant history, so a second turn after tool execution reached the engine's chat template without the fields that bind a tool result to the call it answers. Forward all three (present-only, matching the OpenAI wire shape) and pin vllm.cpp to 6a0bd3e7, where ChatMessage parses/round-trips tool_calls, tool_call_id, name and reasoning and the minja adapter exposes them to the template context. Adds the round-trip request-lowering spec (user -> assistant tool_call -> tool reply -> lowered request) and re-ran the gated e2e suite against the new engine pin with a real Qwen3.5 GGUF: chat, reasoning split, streaming parity, required-tool and auto-tool cases all green. Assisted-by: Claude Code:claude-fable-5 [Bash] [Edit] [Read] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(vllm-cpp): bump vllm.cpp for the darwin arm64 i8mm build fix The darwin-metal CI job was the first build to compile the engine's arm CPU-quant files on macOS and hit their Linux-only <asm/hwcap.h> / <sys/auxv.h> includes. vllm.cpp 9e1c9025 detects i8mm per-OS (auxv on Linux, sysctl on Apple Silicon) with kernels untouched. Gated e2e suite re-run green against the new pin with a real Qwen3.5 GGUF. Assisted-by: Claude Code:claude-fable-5 [Bash] [Read] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(vllm-cpp): darwin build - bound cmake parallelism when nproc is absent The macOS runners have no nproc, so JOBS evaluated empty and `cmake --build -j$(JOBS)` became bare `-j`: unlimited clang jobs on a 3-core/7GB Mac, which swap-thrashed until the 6h GHA timeout (the log shows "nproc: Command not found" and 7+ concurrent clang processes being reaped at the cutoff). Use the same portable fallback chain as the other darwin backends: nproc, then sysctl hw.ncpu, then 4. Assisted-by: Claude Code:claude-fable-5 [Bash] [Edit] [Read] 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> |