The prebuilt vllm 0.14.1+cpu wheel requires SIMD instructions (AVX-512
VNNI/BF16) that stock ubuntu-latest GitHub runners don't support —
vllm.model_executor.models.registry SIGILLs on import during LoadModel.
Source compilation works but takes 30-40 minutes per CI run, which is
too slow for an e2e smoke test. Instead, switch tests-vllm-grpc to the
bigger-runner self-hosted label (already used by backend.yml for the
llama-cpp CUDA build) — that hardware has the required SIMD baseline
and the prebuilt wheel runs cleanly.
FROM_SOURCE=true is kept as an opt-in escape hatch:
- install.sh still has the CPU source-build path for hosts that need it
- backend/Dockerfile.python still declares the ARG + ENV
- Makefile docker-build-backend still forwards the build-arg when set
Default CI path uses the fast prebuilt wheel; source build can be
re-enabled by exporting FROM_SOURCE=true in the environment.
The prebuilt vllm 0.14.1+cpu wheel from GitHub releases is compiled with
SIMD instructions (AVX-512 VNNI/BF16 or AMX-BF16) that not every CPU
supports. GitHub Actions ubuntu-latest runners SIGILL when vllm spawns
the model_executor.models.registry subprocess for introspection, so
LoadModel never reaches the actual inference path.
- install.sh: when FROM_SOURCE=true on a CPU build, temporarily hide
requirements-cpu-after.txt so installRequirements installs the base
deps + torch CPU without pulling the prebuilt wheel, then clone vllm
and compile it with VLLM_TARGET_DEVICE=cpu. The resulting binaries
target the host's actual CPU.
- backend/Dockerfile.python: accept a FROM_SOURCE build-arg and expose
it as an ENV so install.sh sees it during `make`.
- Makefile docker-build-backend: forward FROM_SOURCE as --build-arg
when set, so backends that need source builds can opt in.
- Makefile test-extra-backend-vllm: call docker-build-vllm via a
recursive $(MAKE) invocation so FROM_SOURCE flows through.
- .github/workflows/test-extra.yml: set FROM_SOURCE=true on the
tests-vllm-grpc job. Slower but reliable — the prebuilt wheel only
works on hosts that share the build-time SIMD baseline.
Answers 'did you test locally?': yes, end-to-end on my local machine
with the prebuilt wheel (CPU supports AVX-512 VNNI). The CI runner CPU
gap was not covered locally — this commit plugs that gap.
Extends tests/e2e-backends to cover backends that:
- Resolve HuggingFace model ids natively (vllm, vllm-omni) instead of
loading a local file: BACKEND_TEST_MODEL_NAME is passed verbatim as
ModelOptions.Model with no download/ModelFile.
- Parse tool calls into ChatDelta.tool_calls: new "tools" capability
sends a Predict with a get_weather function definition and asserts
the Reply contains a matching ToolCallDelta. Uses UseTokenizerTemplate
with OpenAI-style Messages so the backend can wire tools into the
model's chat template.
- Need backend-specific Options[]: BACKEND_TEST_OPTIONS lets a test set
e.g. "tool_parser:hermes,reasoning_parser:qwen3" at LoadModel time.
Adds make target test-extra-backend-vllm that:
- docker-build-vllm
- loads Qwen/Qwen2.5-0.5B-Instruct
- runs health,load,predict,stream,tools with tool_parser:hermes
Drops backend/python/vllm/test_{cpu_inference,tool_calls}.py — those
standalone scripts were scaffolding used while bringing up the Python
backend; the e2e-backends harness now covers the same ground uniformly
alongside llama-cpp and ik-llama-cpp.
Concrete vLLM tool parsers override the abstract base's __init__ and
drop the tools kwarg (e.g. Hermes2ProToolParser only takes tokenizer).
Instantiating with tools= raised TypeError which was silently caught,
leaving chat_deltas.tool_calls empty.
Retry the constructor without the tools kwarg on TypeError — tools
aren't required by these parsers since extract_tool_calls finds tool
syntax in the raw model output directly.
Validated with Qwen/Qwen2.5-0.5B-Instruct + hermes parser on CPU:
the backend correctly returns ToolCallDelta{name='get_weather',
arguments='{"location": "Paris, France"}'} in ChatDelta.
test_tool_calls.py is a standalone smoke test that spawns the gRPC
backend, sends a chat completion with tools, and asserts the response
contains a structured tool call.
Validated end-to-end on CPU with Qwen2.5-0.5B-Instruct (LoadModel, Predict,
TokenizeString, Free all working).
- requirements-cpu-after.txt: pin vllm to 0.14.1+cpu (pre-built wheel from
GitHub releases) for x86_64 and aarch64. vllm 0.14.1 is the newest CPU
wheel whose torch dependency resolves against published PyTorch builds
(torch==2.9.1+cpu). Later vllm CPU wheels currently require
torch==2.10.0+cpu which is only available on the PyTorch test channel
with incompatible torchvision.
- requirements-cpu.txt: bump torch to 2.9.1+cpu, add torchvision/torchaudio
so uv resolves them consistently from the PyTorch CPU index.
- install.sh: add --index-strategy=unsafe-best-match for CPU builds so uv
can mix the PyTorch index and PyPI for transitive deps (matches the
existing intel profile behaviour).
- backend.py LoadModel: vllm >= 0.14 removed AsyncLLMEngine.get_model_config
so the old code path errored out with AttributeError on model load.
Switch to the new get_tokenizer()/tokenizer accessor with a fallback
to building the tokenizer directly from request.Model.
- Split vllm install per acceleration: move generic `vllm` out of
requirements-after.txt into per-profile after files (cublas12, hipblas,
intel) and add CPU wheel URL for cpu-after.txt
- requirements-cpu.txt now pulls torch==2.7.0+cpu from PyTorch CPU index
- backend/index.yaml: register cpu-vllm / cpu-vllm-development variants
- New backend/python/common/vllm_utils.py: shared parse_options,
messages_to_dicts, setup_parsers helpers (used by both vllm backends)
- vllm-omni: replace hardcoded chat template with tokenizer.apply_chat_template,
wire native parsers via shared utils, emit ChatDelta with token counts,
add TokenizeString and Free RPCs, detect CPU and set VLLM_TARGET_DEVICE
- Add test_cpu_inference.py: standalone script to validate CPU build with
a small model (Qwen2.5-0.5B-Instruct)
- Use vLLM's ToolParserManager/ReasoningParserManager to extract structured
output (tool calls, reasoning content) instead of reimplementing parsing
- Convert proto Messages to dicts and pass tools to apply_chat_template
- Emit ChatDelta with content/reasoning_content/tool_calls in Reply
- Extract prompt_tokens, completion_tokens, and logprobs from output
- Replace boolean GuidedDecoding with proper GuidedDecodingParams from Grammar
- Add TokenizeString and Free RPC methods
- Fix missing `time` import used by load_video()
When TASK_RESPONSE_TYPE_OAI_CHAT is used, the first streaming token
produces a JSON array with two elements: a role-init chunk and the
actual content chunk. The grpc-server loop called attach_chat_deltas
for both elements with the same raw_result pointer, stamping the first
token's ChatDelta.Content on both replies. The Go side accumulated both,
emitting the first content token twice to SSE clients.
Fix: in the array iteration loops in PredictStream, detect role-init
elements (delta has "role" key) and skip attach_chat_deltas for them.
Only content/reasoning elements get chat deltas attached.
Reasoning models are unaffected because their first token goes into
reasoning_content, not content.
* fix(chat): do not retry if we had chatdeltas or tooldeltas from backend
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* fix: use oai compat for llama.cpp
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* fix: apply to non-streaming path too
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* map also other fields
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
---------
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
The C++ PEG parser needs a few tokens to identify the reasoning format
(e.g. "<|channel>thought\n" for Gemma 4). During this warm-up, the gRPC
layer was sending raw partial tag tokens to Go, which leaked into the
reasoning field.
- Clear reply.message in gRPC when autoparser is active but has no diffs
yet, matching llama.cpp server behavior of only emitting classified output
- Prefer C++ autoparser chat deltas for reasoning/content in all streaming
paths, falling back to Go-side extraction for backends without autoparser
(e.g. vLLM)
- Override non-streaming no-tools result with chat delta content when available
- Guard PrependThinkingTokenIfNeeded against partial tag prefixes during
streaming accumulation
- Reorder default thinking tokens so <|channel>thought is checked before
<|think|> (Gemma 4 templates contain both)
