内容
# vllm-mlx
**Continuous batching + OpenAI + Anthropic APIs in one server. Native Apple Silicon inference.**
**Read this in other languages:** [English](README.md) · [Español](README.es.md) · [Français](README.fr.md) · [中文](README.zh.md)
[](https://pypi.org/project/vllm-mlx/)
[](https://pepy.tech/projects/vllm-mlx)
[](https://www.python.org/downloads/)
[](LICENSE)
[](https://support.apple.com/en-us/HT211814)
[](https://github.com/waybarrios/vllm-mlx)
---
## What is vllm-mlx?
A vLLM-style inference server for Apple Silicon Macs. Unlike `Ollama` or `mlx-lm` used directly, it ships **continuous batching, paged KV cache, prefix caching, and SSD-tiered cache**, and exposes **both OpenAI `/v1/*` and Anthropic `/v1/messages`** from a single process. Run LLMs, vision models, audio, and embeddings on Metal with unified memory, no conversion step.
## Quick start
New users: follow the [isolated release install](docs/getting-started/installation.md#install-a-release)
and [first-response check](docs/getting-started/quickstart.md#first-response).
The example below assumes that environment is activated. Initial model
download/loading time depends on the machine and connection.
The version below is this walkthrough's pinned reference release, not an
automatically updated latest version.
```bash
python -m pip install 'vllm-mlx==0.4.1'
vllm-mlx serve mlx-community/Llama-3.2-3B-Instruct-4bit --host 127.0.0.1 --port 8000
```
After the first response, restart with `--continuous-batching` to try
[continuous batching and its cache options](docs/guides/continuous-batching.md).
The minimal command above uses the default engine without that flag.
**OpenAI SDK:**
Install `openai` in your Python client environment first; it is not installed
by the server package. See the [client setup](docs/getting-started/quickstart.md#option-1-openai-compatible-server).
```python
from openai import OpenAI
client = OpenAI(base_url="http://localhost:8000/v1", api_key="not-needed")
r = client.chat.completions.create(model="mlx-community/Llama-3.2-3B-Instruct-4bit", messages=[{"role": "user", "content": "Hi!"}])
print(r.choices[0].message.content)
```
**Anthropic SDK / Claude Code:**
```bash
export ANTHROPIC_BASE_URL=http://localhost:8000
export ANTHROPIC_API_KEY=not-needed
claude
```
## Coding CLI compatibility
**Validated with OpenCode, pi, Codex, Claude Code, GitHub Copilot CLI, Cline CLI,
and OpenClaw's embedded agent.** All seven completed a streamed tool interaction
and an exact file edit in one local run with Qwen3.8-27B-4bit on September 19,
2026. Results apply to the tested client versions and settings.
See the [validated CLI matrix and setup guide](docs/guides/client-acceptance.md#validated-cli-versions)
for versions, API transports, reproduction commands, and coverage limits.
## Features
### APIs
- **OpenAI-compatible**: `/v1/chat/completions`, `/v1/completions`, `/v1/embeddings`, `/v1/rerank`, `/v1/responses`
- **Anthropic-compatible**: `/v1/messages` (streaming, tool use, system prompts)
- **MCP Tool Calling**: 19 parsers (OpenAI, Anthropic, Gemini, Qwen, DeepSeek, Gemma, and more)
- **Structured output**: JSON Schema via `response_format` (lm-format-enforcer)
### Throughput & memory
- **Continuous batching**: high throughput for concurrent requests
- **Paged KV cache**: memory-efficient with prefix sharing
- **SSD-tiered KV cache**: spill prefix cache to disk for long-context agents (`--ssd-cache-dir`)
- **Warm prompts**: preload popular prefixes at startup (`--warm-prompts`) for 1.3-2.25x TTFT
- **Prefix cache**: trie-based, shared across requests
### Multimodal
- **Text + image + video + audio** from one server
- Vision models: Gemma 3, Gemma 4, Qwen3-VL, Pixtral, Llama vision
- **Audio input** in chat (`audio_url` content blocks)
- **Native TTS**: 11 voices, 15+ languages (Kokoro, Chatterbox, VibeVoice, VoxCPM)
- **STT**: Whisper family with RTF up to 197x on M4 Max
### Reasoning & advanced
- **Reasoning extraction**: Qwen3 and DeepSeek-R1 parsers (`--reasoning-parser`)
- **MoE expert reduction**: `--moe-top-k` for +7-16% on Qwen3-30B-A3B
- **Speculative decoding**: `--enable-mtp` for supported models
- **Sparse prefill**: attention-based `--specprefill` for supported model/draft combinations
### Observability
- **Prometheus metrics**: `/metrics` endpoint with `--enable-metrics`
- **Built-in benchmarker**: `vllm-mlx bench-serve` for prompt sweeps with CSV/JSON output
### Native GPU acceleration
- Apple Silicon only (M1, M2, M3, M4, M5) with Metal kernels via MLX
- Unified memory, no model conversion
## Performance
**LLM decode (M4 Max, 128 GB, greedy, single stream):**
| Model | Tok/s | Memory |
|-------|------:|-------:|
| Qwen3-0.6B-8bit | 417.9 | 0.7 GB |
| Llama-3.2-3B-Instruct-4bit | 205.6 | 1.8 GB |
| Qwen3-30B-A3B-4bit | 127.7 | ~18 GB |
**Audio speech-to-text (M4 Max, RTF = real-time factor):**
| Model | RTF | Use case |
|-------|----:|----------|
| whisper-tiny | 197x | Real-time / low latency |
| whisper-large-v3-turbo | 55x | Quality + speed |
| whisper-large-v3 | 24x | Highest accuracy |
See [docs/benchmarks/](docs/benchmarks/) for continuous-batching results, KV-cache quantization (4-bit / 8-bit / fp16), and MoE top-k sweeps.
## Examples
### Anthropic API (Claude Code, OpenCode)
```bash
vllm-mlx serve mlx-community/Qwen3-8B-4bit --port 8000
export ANTHROPIC_BASE_URL=http://localhost:8000
export ANTHROPIC_API_KEY=not-needed
claude
```
### Reasoning models (Qwen3, DeepSeek-R1)
```bash
vllm-mlx serve mlx-community/Qwen3-8B-4bit --reasoning-parser qwen3
```
```python
r = client.chat.completions.create(
model="mlx-community/Qwen3-8B-4bit",
messages=[{"role": "user", "content": "What is 17 * 23?"}],
)
print("Thinking:", r.choices[0].message.reasoning)
print("Answer:", r.choices[0].message.content)
```
### Multimodal (image + text)
```bash
vllm-mlx serve mlx-community/Qwen3-VL-4B-Instruct-3bit --port 8000
```
```python
r = client.chat.completions.create(
model="mlx-community/Qwen3-VL-4B-Instruct-3bit",
messages=[{"role": "user", "content": [
{"type": "text", "text": "What is in this image?"},
{"type": "image_url", "image_url": {"url": "https://example.com/cat.jpg"}},
]}],
)
```
### Structured output (JSON Schema)
With the text server from the quick start:
```python
r = client.chat.completions.create(
model="mlx-community/Llama-3.2-3B-Instruct-4bit",
messages=[{"role": "user", "content": "List 3 colors."}],
response_format={
"type": "json_schema",
"json_schema": {
"schema": {"type": "object", "properties": {"colors": {"type": "array", "items": {"type": "string"}}}}
},
},
)
```
### Reranking (`/v1/rerank`)
```bash
curl http://localhost:8000/v1/rerank -H 'Content-Type: application/json' -d '{
"model": "default",
"query": "apple silicon inference",
"documents": ["MLX is Apples framework", "Metal kernels on M-series", "CUDA on NVIDIA"]
}'
```
The built-in MLX reranker forward path supports standard BERT/XLM-RoBERTa
sequence-classification weights with `gelu`, `gelu_new`/`gelu_fast`, `relu`, or
`silu`/`swish` `hidden_act` values. Other activations fail explicitly so custom
reranker architectures can add a dedicated adapter instead of silently using the
wrong activation.
### Embeddings
```bash
vllm-mlx serve <llm-model> --embedding-model mlx-community/all-MiniLM-L6-v2-4bit
```
```python
emb = client.embeddings.create(model="mlx-community/all-MiniLM-L6-v2-4bit", input=["Hello", "World"])
```
### Audio (TTS / STT)
```bash
pip install vllm-mlx[audio]
brew install espeak-ng # macOS, needed for non-English TTS
python examples/tts_example.py "Hello, how are you?" --play
python examples/tts_multilingual.py "Hola mundo" --lang es --play
```
### Built-in benchmarking
```bash
vllm-mlx bench-serve --url http://localhost:8000 --concurrency 5 --prompts prompts.txt --output results.csv
# Product-style workload with quality checks and metrics deltas
vllm-mlx bench-serve --url http://localhost:8000 --workload workload.json --repetitions 5 --output results.json
# Append workload rows into SQLite for longitudinal comparisons
vllm-mlx bench-serve --url http://localhost:8000 --workload workload.json --repetitions 5 --format sqlite --output bench.db
```
### Model acquisition and conversion
```bash
# Inspect repo metadata, file sizes, config, and rough fit before downloading weights
vllm-mlx model inspect mlx-community/Llama-3.2-3B-Instruct-4bit
# Acquire with resumable Hugging Face transfer and write a local artifact manifest
vllm-mlx model acquire mlx-community/Llama-3.2-3B-Instruct-4bit --target-dir ./models/llama-3b-4bit
# Wrap mlx-lm conversion and record the exact recipe in the converted artifact
vllm-mlx model convert meta-llama/Llama-3.2-3B-Instruct --output ./models/llama-3b-mlx-q4 --quantize --q-bits 4 --q-group-size 64 --q-mode affine
```
### Prometheus metrics
```bash
vllm-mlx serve <model> --enable-metrics
curl http://localhost:8000/metrics
```
## Installation
Use the [release walkthrough](docs/getting-started/installation.md#install-a-release)
for an isolated environment and a pinned server version. If you already manage
CLI tools with uv, its isolated equivalent is:
```bash
uv tool install 'vllm-mlx==0.4.1'
```
Keep [development checkouts](docs/getting-started/installation.md#development-checkout)
separate. See the [Installation Guide](docs/getting-started/installation.md) for
optional extras and [Audio Guide](docs/guides/audio.md) for audio setup.
## Documentation
Browse the complete documentation at [vllm-mlx.is-a.dev](https://vllm-mlx.is-a.dev/).
- **Getting started**: [Installation](docs/getting-started/installation.md) · [Quick Start](docs/getting-started/quickstart.md)
- **Servers & APIs**: [OpenAI server](docs/guides/server.md) · [Anthropic Messages API](docs/guides/server.md#anthropic-messages-api) · [Python API](docs/guides/python-api.md)
- **Client compatibility**: [Run coding-client acceptance checks](docs/guides/client-acceptance.md)
- **Features**: [Multimodal](docs/guides/multimodal.md) · [Audio](docs/guides/audio.md) · [Embeddings](docs/guides/embeddings.md) · [Reasoning](docs/guides/reasoning.md) · [MCP & Tool Calling](docs/guides/mcp-tools.md) · [Tool Parsers](docs/guides/tool-calling.md)
- **Performance**: [Continuous Batching](docs/guides/continuous-batching.md) · [Multi-Model Serving](docs/guides/model-registry.md) · [Warm Prompts](docs/guides/warm-prompts.md) · [MoE Top-K](docs/guides/moe-top-k.md)
- **Reference**: [CLI](docs/reference/cli.md) · [Models](docs/reference/models.md) · [Configuration](docs/reference/configuration.md)
- **Benchmarks**: [LLM](docs/benchmarks/llm.md) · [Image](docs/benchmarks/image.md) · [Video](docs/benchmarks/video.md) · [Audio](docs/benchmarks/audio.md)
## Architecture
```
┌─────────────────────────────────────────────────────────────────────────┐
│ vllm-mlx Server │
│ OpenAI /v1/* · Anthropic /v1/messages · /v1/rerank · /metrics │
└─────────────────────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────────────┐
│ Continuous batching · Paged KV cache · Prefix cache · SSD tiering │
└─────────────────────────────────────────────────────────────────────────┘
│
┌─────────────┬────────────┴────────────┬─────────────┐
▼ ▼ ▼ ▼
┌───────────────┐ ┌───────────────┐ ┌───────────────┐ ┌───────────────┐
│ mlx-lm │ │ mlx-vlm │ │ mlx-audio │ │mlx-embeddings │
│ (LLMs) │ │ (Vision) │ │ (TTS + STT) │ │ (Embeddings) │
└───────────────┘ └───────────────┘ └───────────────┘ └───────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────────────┐
│ MLX · Metal kernels · Unified memory │
└─────────────────────────────────────────────────────────────────────────┘
```
## Contributing
Bug fixes, perf work, docs, and benchmarks on different Apple Silicon chips all welcome. See the [Contributing Guide](docs/development/contributing.md).
## License
Apache 2.0. See [LICENSE](LICENSE).
## Citation
```bibtex
@software{vllm_mlx2025,
author = {Barrios, Wayner},
title = {vllm-mlx: Apple Silicon MLX Backend for vLLM},
year = {2025},
url = {https://github.com/waybarrios/vllm-mlx},
note = {Native GPU-accelerated LLM and vision-language model inference on Apple Silicon}
}
```
## Acknowledgments
- [MLX](https://github.com/ml-explore/mlx). Apple's ML framework.
- [mlx-lm](https://github.com/ml-explore/mlx-lm). LLM inference library.
- [mlx-vlm](https://github.com/Blaizzy/mlx-vlm). Vision-language models.
- [mlx-audio](https://github.com/Blaizzy/mlx-audio). Text-to-Speech and Speech-to-Text.
- [mlx-embeddings](https://github.com/Blaizzy/mlx-embeddings). Text embeddings.
- [Rapid-MLX](https://github.com/raullenchai/Rapid-MLX). Community fork of vllm-mlx.
- [vLLM](https://github.com/vllm-project/vllm). High-throughput LLM serving. vllm-mlx is inspired by vLLM and adopts its continuous-batching and paged KV-cache design for Apple Silicon via MLX.
## Star history
[](https://star-history.dera.page/#waybarrios/vllm-mlx&Date)
---
**If vllm-mlx helped you, please star the repo. It helps more Apple Silicon devs find it.**
连接信息
猜你喜欢
暂无推荐服务器