Local syntax highlighting with WebGPU and a micromodel 🧠⚡️
gpu-lexer uses a local WebGPU micromodel to tokenize code and highlight syntax without server-side computation.
Following up on the post about parsing dates with the gpu-time micro neural network, another SLM has landed in the browser runtime: the gpu-lexer project.
The library handles code syntax highlighting, but instead of conventional lexers and AST parsers, a micromodel runs under the hood directly on your WebGPU.
What makes it interesting, and the concept behind it:
- An SLM in the browser: More proof that narrow tasks don’t require heavy APIs from OpenAI/Claude or monstrous LLMs.
- All the magic happens on the client: The model tokenizes code locally and highlights syntax on the graphics card, without clogging the CPU’s main thread.
- Zero server overhead: All computation takes place on the user’s hardware, with no token charges.
It looks like we’re entering an era when developers will start putting micro neural networks into practically every utility tool, from regular expressions to Markdown parsers.