Translate modern speculative fields to the pinned fork API, disable unsupported score and checkpoint features, and cover the compatibility transform with an idempotent regression test.
Assisted-by: Codex:gpt-5 [Codex]
The buun build copies the stock llama.cpp backend directory, including patches that target upstream. Remove that copied patch directory before invoking the shared build so only the explicit buun compatibility series is applied to the fork.
Assisted-by: Codex:gpt-5 [systematic-debugging]
The shared gRPC wrapper stores p_split under the draft sub-structure. Match that exact source spelling so the fork-specific patch stage reaches the build on every architecture.
Assisted-by: Codex:gpt-5 [Codex]
Two more hipblas-only build failures in buun's fattn.cu, fixed under the
same patches/ infrastructure:
1. cudaMemcpyToSymbol / cudaMemcpyFromSymbol — buun's Q² calibration +
TCQ codebook upload paths call the symbol variants of cudaMemcpy.
ggml/src/ggml-cuda/vendors/hip.h aliases every other cudaMemcpy*
name (cudaMemcpy, cudaMemcpyAsync, cudaMemcpy2DAsync, …) but the
symbol pair was never added. 15+ "use of undeclared identifier"
errors across fattn.cu lines 40, 54, 74-76, 94, 100-101, 371, 883,
905, 954, 976, 1449, 1463. Add the two missing aliases alongside
the existing memcpy block.
2. __shfl_xor_sync fwht128 calls — same 3-arg omission pattern as the
earlier argmax top-K fix. Lines 512 (ggml_cuda_fwht128 intra-warp
butterfly) and 536 (fwht128_store_half neighbor fetch) drop the
width argument that hip.h:33 requires. Add WARP_SIZE.
Assisted-by: Claude:claude-opus-4-7
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Two call sites in ggml/src/ggml-cuda/argmax.cu (the top-K intra-warp
merge added by buun) use the 3-arg CUDA form __shfl_xor_sync(mask, var,
laneMask), omitting the optional width parameter. The hipification shim
at ggml/src/ggml-cuda/vendors/hip.h:33 is a function-like macro that
requires all four arguments, so hipcc fails with:
argmax.cu:265: too few arguments provided to function-like macro
invocation
note: macro '__shfl_xor_sync' defined here:
#define __shfl_xor_sync(mask, var, laneMask, width) \
__shfl_xor(var, laneMask, width)
Every other call in the same file already passes WARP_SIZE explicitly;
aligning these two with that convention fixes the hipblas build without
changing CUDA codegen (warpSize is the CUDA default).
Assisted-by: Claude:claude-opus-4-7
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Buun's Q² calibration path in ggml/src/ggml-cuda/fattn.cu calls
atomicAdd with a double* destination. Native double atomicAdd is only
available on CUDA compute capability 6.0 and later — LocalAI's CUDA 12
Docker image builds for the full published arch range (which includes
sm_50/sm_52), so nvcc fails with:
fattn.cu:812: error: no instance of overloaded function "atomicAdd"
matches the argument list, argument types are: (double *, double)
Add the canonical CAS-loop shim from the CUDA C Programming Guide
(B.15 Atomic Functions) guarded on __CUDA_ARCH__ < 600. On sm_60+ the
guard is false and nvcc picks up the native intrinsic as before.
Patch file lives under backend/cpp/buun-llama-cpp/patches/ and is
applied to the cloned fork tree by apply-patches.sh (the infrastructure
already put in place for exactly this class of backport).
Assisted-by: Claude:claude-opus-4-7
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Adds the buun-llama-cpp backend to the same CI pipelines that turboquant
and sherpa-onnx already use:
- scripts/changed-backends.js: path resolution for Dockerfile.buun-llama-cpp,
plus fork-of-fork detection (changes under backend/cpp/llama-cpp/ also
retrigger the buun pipeline, mirroring how turboquant is handled).
- .github/workflows/test-extra.yml: detect-changes output and a new
tests-buun-llama-cpp-grpc job that runs make test-extra-backend-buun-llama-cpp
(turbo3 V-cache, same rationale as tests-turboquant-grpc).
- .github/workflows/backend.yml: 9 matrix entries (CUDA 12/13, L4T CUDA
13 ARM64, ROCm, SYCL f32/f16, CPU, L4T ARM64, Vulkan) paired with each
existing turboquant entry so image builds have platform parity.
Also updates .agents/ai-coding-assistants.md to clarify that AI agents
operating under the human submitter's git identity SHOULD emit
Signed-off-by via `git commit -s` (never inventing or guessing another
identity) — documents the workflow this PR is using.
Assisted-by: Claude:claude-opus-4-7
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Previous substitution kept the call as 5 args, but buun predates the
upstream refactor that also *added* the logit_bias_eog parameter to
params_from_json_cmpl — buun's signature is still the 4-arg form
(const llama_vocab*, const common_params&, int, const json&)
and it still derives logit_bias_eog internally from the common_params.
Replace the substitution with a line-delete. Guard matches both the
original call (ctx_server.get_meta().logit_bias_eog) and the previously
substituted form (params_base.sampling.logit_bias_eog) so the script
stays safe across re-runs and whatever state the tree was left in.
Assisted-by: Claude:Opus-4.7 [Read] [Edit] [Bash]
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
LocalAI's shared grpc-server.cpp reaches
ctx_server.get_meta().logit_bias_eog twice (the twin params_from_json_cmpl
callsites). That accessor was added to server_context_meta upstream after
buun's 2026-04-05 fork-point, so compiling against buun errors with
'struct server_context_meta' has no member named 'logit_bias_eog'.
Rewrite the call sites — only in the buun grpc-server.cpp copy — to source
the vector from params_base.sampling.logit_bias_eog instead. That vector is
the underlying data the upstream meta accessor eventually returns (buun
still carries common_params_sampling::logit_bias_eog at common.h:280), so
the substitution yields identical behavior on both trees.
The sed is guarded by a grep for the call site, so this patch is
self-disabling once buun rebases past the upstream refactor.
Assisted-by: Claude:Opus-4.7 [Read] [Edit] [Bash] [WebFetch]
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Two additions that pair with the new backend:
- An Import()-side case that asserts preference buun-llama-cpp produces
backend: buun-llama-cpp in the emitted YAML (mirrors the existing
ik-llama-cpp and turboquant cases).
- AdditionalBackends() spec now asserts all three drop-in replacements
are advertised, and verifies buun-llama-cpp's Modality/Description
alongside the other two.
Assisted-by: Claude:Opus-4.7 [Read] [Edit] [Bash]
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
spiritbuun/buun-llama-cpp is a fork of TheTom/llama-cpp-turboquant that adds
two independent features on top: DFlash block-diffusion speculative decoding
(via a dedicated DFlashDraftModel GGUF arch) and two extra TCQ KV-cache
variants (turbo2_tcq, turbo3_tcq) on top of TurboQuant's turbo2/turbo3/turbo4.
Follows the turboquant thin-wrapper pattern — reuses backend/cpp/llama-cpp
grpc-server sources verbatim, patches only the build copy to extend the KV
allow-list and wire up buun-exclusive tree_budget / draft_topk options.
DraftModel is already wired end-to-end (proto field 39 → params.speculative),
so DFlash activation only needs the existing options passthrough
(spec_type:dflash) plus the drafter path in draft_model.
CacheTypeOptions now surfaces the five turbo* values so the React UI dropdown
shows them — benefits turboquant too (previously users had to type them in
YAML manually).
Assisted-by: Claude:Opus-4.7 [Read] [Edit] [Bash] [WebFetch]
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
The LocalAGI bump in #11985 changed state.NewAgentPool to take a
SkillsProvider and a PoolLimits value. The call site here was not
updated, so master stopped compiling and every Go job went red.
Pass the limits explicitly, mirroring LocalAGI's own defaults, so the
pool prunes conversation dumps and scheduler run history instead of
growing without end.
Assisted-by: Claude Code:claude-opus-5 [Bash] [Edit]
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
The CUDA 13 FlashAttention build still exhausts hosted-runner memory with a single ninja worker because nvcc can compile multiple threads internally. Limit nvcc to one thread for that profile and guard the setting in the backend test script.
Assisted-by: Codex:gpt-5
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
grpc::ServerWriter::Write() returns false once the peer is gone, and
PredictStream ignored that result at every call site. The handler kept
pulling decoded tokens and writing them into a dead stream, so the
llama.cpp slot stayed busy until the generation ended on its own terms.
A model configured with max_tokens 0 and a large context ends on its own
terms only at the context limit. On a 35B model at ~41 t/s a 120k context
is about fifty minutes, and a slot held that long is a slot every other
request for that model queues behind. Two abandoned requests were enough
to make a node with free VRAM and a healthy control plane serve nothing:
new requests timed out waiting for a slot, each timeout abandoned another
generation, and the node fell further behind the longer it ran.
Track the peer instead. The first failed write retires it for good, since
a stream never recovers, and the RPC's own cancellation flag folds into
the same predicate so the loop has one condition to test. Returning early
is what frees the slot: ~server_response_reader() posts
SERVER_TASK_TYPE_CANCEL for whatever is still decoding.
TTSStream already checked Write(); this brings PredictStream in line.
Cancellation stays cooperative and is checked between decoded results, so
a batch already in flight may finish before the request stops.
Assisted-by: Claude:claude-opus-5
pinned: true was only honoured by the per-node watchdog. Every distributed
eviction path was pinned-blind: the router's LRU eviction (EvictLRU,
evictLRUAndFreeNode) and the replica reconciler's idle scale-down would
happily unload a pinned model — and since eviction is gated on
in_flight = 0, a pinned model became eviction-eligible the instant each
response completed. Under capacity pressure that surfaces as the backend
being freed immediately after every request (#11101).
Wire the model config loader into the router and reconciler through a new
PinnedModelResolver seam (mirroring ConcurrencyConflictResolver):
- EvictLRU passes the pinned set into FindLRUModel's query so the
next-oldest unpinned model is selected instead of the attempt failing
- evictLRUAndFreeNode filters pinned models inside its locked selection
- scaleDownIdle skips pinned models entirely: trimming to the floor still
means requests beyond the survivor's capacity pay a cold reload
Deliberate teardown (admin unload, model delete, node drain) intentionally
still applies to pinned models, as does dead-row reaping (state correction,
not eviction).
Regression specs verified to fail with the exclusion disabled.
Addresses the cluster-side eviction gap in #11101
Assisted-by: Claude Code:claude-fable-5 [Claude Code]
Signed-off-by: Adira Denis Muhando <dennisadira@gmail.com>
SciPy 1.18 requires Python 3.12 or newer. Keep this backend on a
compatible portable Python for Linux and macOS builds.
Assisted-by: Codex:gpt-5.6 [Codex]
(cherry picked from commit 70bf6d4a3a)
* chore(deps): bump github.com/mudler/localrecall to v0.6.5
Picks up two Postgres engine fixes: the RRF fusion no longer scores
every hybrid-search candidate 0 through integer division, and the
search_vector text config is no longer pinned to 'simple' for the life
of the process after one transient lookup failure.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ANmTgdYikmzBq67jVJ1CgX
* chore(deps): bump github.com/mudler/LocalAGI to d93d478
Picks up mudler/LocalAGI#493, which bumps localrecall to v0.6.5 there
too, so the direct pin in this module and the version arriving through
LocalAGI agree.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ANmTgdYikmzBq67jVJ1CgX
---------
Co-authored-by: Ettore Di Giacinto <mudler@localai.io>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Let the model-provided tokenizer template format Gemma conversations instead of maintaining a shared inline prompt template.
Assisted-by: Codex:gpt-5
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Do not seed streaming reasoning state when the latest prompt thinking marker is already followed by its matching closing marker. This keeps direct Gemma 4 output in content when its template disables thinking with a preclosed channel.
Assisted-by: Codex:gpt-5
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
backend.stop was the one lifecycle subject a worker never answered. The
controller published and returned nil as soon as the local publish
succeeded, so a stop that killed nothing, and a stop that failed
outright, were indistinguishable from one that worked.
The unload endpoint calls model.unload and then StopBackend. Only the
first is acknowledged, so the endpoint answered 200 while the backend
kept running and held its VRAM, and its own "backend stop failed" branch
could never run. The worker logged the failure and nobody saw it.
Give the subject a reply. The worker now enumerates the process keys it
terminated and reports any per-process error, so StopBackend fails when
the stop failed. Resolving to nothing stays a success: stopping a backend
that is not running leaves the caller in the state it asked for, and
eviction paths stop already-gone models routinely. The empty list is what
says nothing matched, and ReportsStoppedProcesses is what makes that
emptiness trustworthy, the same way BackendDeleteReply handles it.
A worker built before this reply still receives the request and still
stops the backend, it only stays silent, so a timeout degrades to the old
assumption rather than failing every stop on a fleet mid-upgrade. Only
silence degrades: a transport error is still reported, because
UnloadRemoteModel skips its registry cleanup for a node it could not
reach and needs to keep hearing about that.
Assisted-by: Claude:claude-opus-5 golangci-lint
Gallery installs merged family defaults at the YAML root and only re-marshaled them on the artifact path. Persist the defaults in the loader-visible parameters map for every install path while preserving authored overrides.
Assisted-by: Codex:gpt-5
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Adds FunASR/SenseVoice as a Python backend for speech-to-text with
support for CPU, CUDA 12/13, ROCm, Intel SYCL, L4T, and Apple MPS.
Co-authored-by: xingyifeng <xingyifeng@users.noreply.github.com>
Agent Status replaced the chat route and unmounted its EventSource. Any response still in flight could then disappear from the conversation.\n\nOpen status in a separate tab so the chat keeps its live connection until the response completes.\n\nAssisted-by: Codex:gpt-5 [eslint]
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Saved profiles previously resolved to one audio path and transcript, so
cloning backends could not use several examples of one personality.
Store ordered audio and transcript pairs while preserving the legacy
first-reference fields. Fish Speech and audio.cpp receive all pairs,
including on distributed workers. Other backends retain their
single-reference behavior.
Assisted-by: Codex:gpt-5
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
localai-org-maint-bot