crowwork

The goal of the project is to bring open ABI and FFI for machine learning systems.

- Stable, minimal C ABI designed for kernels, DSLs, and runtime extensibility. - Zero-copy interop across PyTorch, JAX, and CuPy using DLPack protocol. - Compact value and call convention covering common data types for ultra low-overhead ML applications. - Multi-language support out of the box: Python, C++, and Rust (with a path towards more languages).

Scale LLM serving with programmable cross-engine serving patterns, all in a few lines of Python

XGrammar is an open-source library for efficient, flexible, and portable structured generation. Bring 2x-10x speedup in grammar grammar-guided(JSON and CFG) LLM serving.

Comes with ability to do full structured generation with json schema

also a in-browser demo https://chat.webllm.ai/

runs on qwen2 on iphone with 26 tok/sec and a OpenAI style swift API

2b model running at 20tok/sec on iphone, nice potential for future applications

Runs Phi-2 on Samsung S23 with pretty decent speed on Google Chrome browser.

LLM on browser on a phone

You can also try out the vulkan backend, which we know should work for windows, although speed might be slower than rocm

Yes, it works out of box and the blog contains a prebuilt python package that you can try out

There is also vulkan support which should be more universal(also included in the post), for example, the post also shows running LLM on a steamdeck APU.

Checkout the latest docs https://mlc.ai/mlc-llm/docs/ MLC started with demos and it evolved lately, with API integrations, documentations into an inference solution that everyone can reuse for universal deployments

It certainly also involves generating code(e.g. WebGPU, vulkan) that are more akin to traditionally compiler, and more like graph and memory optimization. So indeed more than packaging.

Please checkout the course if you are interested

There is a conda app that can be installed on macos

You can try out the demo and benchmark yourself

Supported platforms include:

- Metal GPUs on iPhone and Intel/ARM MacBooks

- AMD and NVIDIA GPUs via Vulkan on Windows and Linux

- NVIDIA GPUs via CUDA on Windows and Linux

- WebGPU on browsers (through companion project WebLLM).

tvm runtime is pretty decent(~700k-2M level depending on dependency included), you can checkout tvm community and bring up the question there, i think there might be some common interest. There are impl of runtime for vulkan, metal that can be used as reference.

I think instead what would be needed is a wgpu native runtime support for TVM. Like the implementations in tvm vulkan, then it will be naturally link to any runtime that provides webgpu.h

Then yah the llm_chat.js would be high-level logic that targets the tvm runtime, and can be implemented in any language that tvm runtime support(that includes, js, java, c++ rust etc).

Support webgpu native is an interesting direction. Feel free to open a thread in tvm discuss forum and perhaps there would be fun things to collaborate in OSS