{"id":35759140,"url":"https://github.com/waybarrios/vllm-mlx","last_synced_at":"2026-01-23T02:02:29.619Z","repository":{"id":330057862,"uuid":"1111335968","full_name":"waybarrios/vllm-mlx","owner":"waybarrios","description":"OpenAI-compatible server for Apple Silicon. Run LLMs and vision-language models (Llama, Qwen-VL, LLaVA) with continuous batching, MCP tool calling, and multimodal support. Native MLX backend, 400+ tok/s.","archived":false,"fork":false,"pushed_at":"2026-01-21T03:27:11.000Z","size":10163,"stargazers_count":180,"open_issues_count":3,"forks_count":17,"subscribers_count":6,"default_branch":"main","last_synced_at":"2026-01-21T13:51:59.442Z","etag":null,"topics":["apple-silicon","audio-processing","computer-vision","image-understanding","inference","llm","machine-learning","macos","mllm","mlx","multimodal-ai","speech-to-text","stt","text-to-speech","tts","video-understanding","vision-language-model","vllm"],"latest_commit_sha":null,"homepage":null,"language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":null,"status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/waybarrios.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":null,"code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2025-12-06T18:26:08.000Z","updated_at":"2026-01-21T13:50:41.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/waybarrios/vllm-mlx","commit_stats":null,"previous_names":["waybarrios/vllm-mlx"],"tags_count":3,"template":false,"template_full_name":null,"purl":"pkg:github/waybarrios/vllm-mlx","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/waybarrios%2Fvllm-mlx","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/waybarrios%2Fvllm-mlx/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/waybarrios%2Fvllm-mlx/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/waybarrios%2Fvllm-mlx/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/waybarrios","download_url":"https://codeload.github.com/waybarrios/vllm-mlx/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/waybarrios%2Fvllm-mlx/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":28677715,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-01-23T01:00:35.747Z","status":"online","status_checked_at":"2026-01-23T02:00:08.296Z","response_time":59,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["apple-silicon","audio-processing","computer-vision","image-understanding","inference","llm","machine-learning","macos","mllm","mlx","multimodal-ai","speech-to-text","stt","text-to-speech","tts","video-understanding","vision-language-model","vllm"],"created_at":"2026-01-06T22:28:22.361Z","updated_at":"2026-01-23T02:02:29.584Z","avatar_url":"https://github.com/waybarrios.png","language":"Python","funding_links":[],"categories":["Langchain","LLM \u0026 Inference"],"sub_categories":[],"readme":"# vLLM-MLX\n\n**vLLM-like inference for Apple Silicon** - GPU-accelerated Text, Image, Video \u0026 Audio on Mac\n\n[![License](https://img.shields.io/badge/License-Apache_2.0-blue.svg)](LICENSE)\n[![Python 3.10+](https://img.shields.io/badge/python-3.10+-blue.svg)](https://www.python.org/downloads/)\n[![Apple Silicon](https://img.shields.io/badge/Apple-Silicon-black.svg)](https://support.apple.com/en-us/HT211814)\n[![GitHub](https://img.shields.io/badge/GitHub-waybarrios%2Fvllm--mlx-blue?logo=github)](https://github.com/waybarrios/vllm-mlx)\n\n## Overview\n\nvllm-mlx brings native Apple Silicon GPU acceleration to vLLM by integrating:\n\n- **[MLX](https://github.com/ml-explore/mlx)**: Apple's ML framework with unified memory and Metal kernels\n- **[mlx-lm](https://github.com/ml-explore/mlx-lm)**: Optimized LLM inference with KV cache and quantization\n- **[mlx-vlm](https://github.com/Blaizzy/mlx-vlm)**: Vision-language models for multimodal inference\n- **[mlx-audio](https://github.com/Blaizzy/mlx-audio)**: Speech-to-Text and Text-to-Speech with native voices\n\n## Features\n\n- **Multimodal** - Text, Image, Video \u0026 Audio in one platform\n- **Native GPU acceleration** on Apple Silicon (M1, M2, M3, M4)\n- **Native TTS voices** - Spanish, French, Chinese, Japanese + 5 more languages\n- **OpenAI API compatible** - drop-in replacement for OpenAI client\n- **MCP Tool Calling** - integrate external tools via Model Context Protocol\n- **Paged KV Cache** - memory-efficient caching with prefix sharing\n- **Continuous Batching** - high throughput for multiple concurrent users\n\n## Quick Start\n\n### Installation\n\n```bash\ngit clone https://github.com/waybarrios/vllm-mlx.git\ncd vllm-mlx\npip install -e .\n```\n\n### Start Server\n\n```bash\n# Simple mode (single user, max throughput)\nvllm-mlx serve mlx-community/Llama-3.2-3B-Instruct-4bit --port 8000\n\n# Continuous batching (multiple users)\nvllm-mlx serve mlx-community/Llama-3.2-3B-Instruct-4bit --port 8000 --continuous-batching\n\n# With API key authentication\nvllm-mlx serve mlx-community/Llama-3.2-3B-Instruct-4bit --port 8000 --api-key your-secret-key\n```\n\n### Use with OpenAI SDK\n\n```python\nfrom openai import OpenAI\n\n# Without API key (local development)\nclient = OpenAI(base_url=\"http://localhost:8000/v1\", api_key=\"not-needed\")\n\n# With API key (production)\nclient = OpenAI(base_url=\"http://localhost:8000/v1\", api_key=\"your-secret-key\")\n\nresponse = client.chat.completions.create(\n    model=\"default\",\n    messages=[{\"role\": \"user\", \"content\": \"Hello!\"}],\n)\nprint(response.choices[0].message.content)\n```\n\n### Multimodal (Images \u0026 Video)\n\n```bash\nvllm-mlx serve mlx-community/Qwen3-VL-4B-Instruct-3bit --port 8000\n```\n\n```python\nresponse = client.chat.completions.create(\n    model=\"default\",\n    messages=[{\n        \"role\": \"user\",\n        \"content\": [\n            {\"type\": \"text\", \"text\": \"What's in this image?\"},\n            {\"type\": \"image_url\", \"image_url\": {\"url\": \"https://example.com/image.jpg\"}}\n        ]\n    }]\n)\n```\n\n### Audio (TTS/STT)\n\n```bash\n# Install audio dependencies\npip install vllm-mlx[audio]\npython -m spacy download en_core_web_sm\nbrew install espeak-ng  # macOS, for non-English languages\n```\n\n```bash\n# Text-to-Speech (English)\npython examples/tts_example.py \"Hello, how are you?\" --play\n\n# Text-to-Speech (Spanish)\npython examples/tts_multilingual.py \"Hola mundo\" --lang es --play\n\n# List available models and languages\npython examples/tts_multilingual.py --list-models\npython examples/tts_multilingual.py --list-languages\n```\n\n**Supported TTS Models:**\n| Model | Languages | Description |\n|-------|-----------|-------------|\n| Kokoro | EN, ES, FR, JA, ZH, IT, PT, HI | Fast, 82M params, 11 voices |\n| Chatterbox | 15+ languages | Expressive, voice cloning |\n| VibeVoice | EN | Realtime, low latency |\n| VoxCPM | ZH, EN | High quality Chinese/English |\n\n## Documentation\n\nFor full documentation, see the [docs](docs/) directory:\n\n- **Getting Started**\n  - [Installation](docs/getting-started/installation.md)\n  - [Quick Start](docs/getting-started/quickstart.md)\n\n- **User Guides**\n  - [OpenAI-Compatible Server](docs/guides/server.md)\n  - [Python API](docs/guides/python-api.md)\n  - [Multimodal (Images \u0026 Video)](docs/guides/multimodal.md)\n  - [Audio (STT/TTS)](docs/guides/audio.md)\n  - [MCP \u0026 Tool Calling](docs/guides/mcp-tools.md)\n  - [Continuous Batching](docs/guides/continuous-batching.md)\n\n- **Reference**\n  - [CLI Commands](docs/reference/cli.md)\n  - [Supported Models](docs/reference/models.md)\n  - [Configuration](docs/reference/configuration.md)\n\n- **Benchmarks**\n  - [LLM Benchmarks](docs/benchmarks/llm.md)\n  - [Image Benchmarks](docs/benchmarks/image.md)\n  - [Video Benchmarks](docs/benchmarks/video.md)\n  - [Audio Benchmarks](docs/benchmarks/audio.md)\n\n## Architecture\n\n```\n┌─────────────────────────────────────────────────────────┐\n│                    vLLM API Layer                       │\n│           (OpenAI-compatible interface)                 │\n└─────────────────────────────────────────────────────────┘\n                           │\n                           ▼\n┌─────────────────────────────────────────────────────────┐\n│                    MLXPlatform                          │\n│       (vLLM platform plugin for Apple Silicon)          │\n└─────────────────────────────────────────────────────────┘\n                           │\n          ┌────────────────┼────────────────┐\n          ▼                ▼                ▼\n┌──────────────────┐ ┌──────────────┐ ┌──────────────────┐\n│     mlx-lm       │ │   mlx-vlm    │ │    mlx-audio     │\n│  (LLM inference) │ │ (Vision+LLM) │ │   (TTS + STT)    │\n└──────────────────┘ └──────────────┘ └──────────────────┘\n          │                │                  │\n          └────────────────┴──────────────────┘\n                           │\n                           ▼\n┌─────────────────────────────────────────────────────────┐\n│                        MLX                              │\n│         (Apple ML Framework - Metal kernels)            │\n└─────────────────────────────────────────────────────────┘\n```\n\n## Performance\n\n**LLM Performance (M4 Max, 128GB):**\n\n| Model | Speed | Memory |\n|-------|-------|--------|\n| Qwen3-0.6B-8bit | 402 tok/s | 0.7 GB |\n| Llama-3.2-1B-4bit | 464 tok/s | 0.7 GB |\n| Llama-3.2-3B-4bit | 200 tok/s | 1.8 GB |\n\n**Continuous Batching (5 concurrent requests):**\n\n| Model | Single | Batched | Speedup |\n|-------|--------|---------|---------|\n| Qwen3-0.6B-8bit | 328 tok/s | 1112 tok/s | **3.4x** |\n| Llama-3.2-1B-4bit | 299 tok/s | 613 tok/s | **2.0x** |\n\n**Audio - Speech-to-Text (M4 Max, 128GB):**\n\n| Model | RTF* | Use Case |\n|-------|------|----------|\n| whisper-tiny | **197x** | Real-time, low latency |\n| whisper-large-v3-turbo | **55x** | Best quality/speed balance |\n| whisper-large-v3 | **24x** | Highest accuracy |\n\n*RTF = Real-Time Factor. RTF of 100x means 1 minute transcribes in ~0.6 seconds.\n\nSee [benchmarks](docs/benchmarks/) for detailed results.\n\n## Contributing\n\nWe welcome contributions! See [Contributing Guide](docs/development/contributing.md) for details.\n\n- Bug fixes and improvements\n- Performance optimizations\n- Documentation improvements\n- Benchmarks on different Apple Silicon chips\n\nSubmit PRs to: [https://github.com/waybarrios/vllm-mlx](https://github.com/waybarrios/vllm-mlx)\n\n## License\n\nApache 2.0 - see [LICENSE](LICENSE) for details.\n\n## Citation\n\nIf you use vLLM-MLX in your research or project, please cite:\n\n```bibtex\n@software{vllm_mlx2025,\n  author = {Barrios, Wayner},\n  title = {vLLM-MLX: Apple Silicon MLX Backend for vLLM},\n  year = {2025},\n  url = {https://github.com/waybarrios/vllm-mlx},\n  note = {Native GPU-accelerated LLM and vision-language model inference on Apple Silicon}\n}\n```\n\n## Acknowledgments\n\n- [MLX](https://github.com/ml-explore/mlx) - Apple's ML framework\n- [mlx-lm](https://github.com/ml-explore/mlx-lm) - LLM inference library\n- [mlx-vlm](https://github.com/Blaizzy/mlx-vlm) - Vision-language models\n- [mlx-audio](https://github.com/Blaizzy/mlx-audio) - Text-to-Speech and Speech-to-Text\n- [vLLM](https://github.com/vllm-project/vllm) - High-throughput LLM serving\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fwaybarrios%2Fvllm-mlx","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fwaybarrios%2Fvllm-mlx","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fwaybarrios%2Fvllm-mlx/lists"}