---
format: "aidr-story-markdown/v1"
id: "40e32c06480c876c58c30f307caebfbf022a3bf3a4f5952c5d07a7e6a23ef834"
canonical_url: "https://aidr.today/40e32c06?lang=en"
title: "Developer Proves LLM Fine Tuning Possible in Browser Using WebGPU"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-09-06T11:35:52.000Z"
category: "Research"
topics: ["fine-tuning","open-source","llm","inference"]
source_urls: ["https://huggingnews.com/ai/developer-proves-llm-fine-tuning-possible-in-browser-using-webgpu-48af825b","https://x.com/ngxson/status/2096379210939994573","https://x.com/maximelabonne/status/2096456605076476239"]
summary: "Developer ngxson created a software prototype that allows for the training of AI models directly on a user's machine without an external server. The proof of concept demonstrates that large language models can be fine-tuned in a web browser using WebGPU, backed by the llama.cpp and wllama libraries. This approach eliminates the need for expensive cloud infrastructure during the model adaptation process. This development shifts the capability of browser-based AI beyond inference, which only runs pre-existing models to generate output. Current efforts are focused on integrating Low-Rank Adaptation, known as LoRA, to further optimize the memory and power required for client-side training."
---

# Developer Proves LLM Fine Tuning Possible in Browser Using WebGPU

> [Open the canonical story](<https://aidr.today/40e32c06?lang=en>)

**Published:** 2026-09-06T11:35:52.000Z
**Category:** Research
**Topics:** fine\-tuning, open\-source, llm, inference

## Summary

Developer ngxson created a software prototype that allows for the training of AI models directly on a user's machine without an external server\. The proof of concept demonstrates that large language models can be fine\-tuned in a web browser using WebGPU, backed by the llama\.cpp and wllama libraries\. This approach eliminates the need for expensive cloud infrastructure during the model adaptation process\. This development shifts the capability of browser\-based AI beyond inference, which only runs pre\-existing models to generate output\. Current efforts are focused on integrating Low\-Rank Adaptation, known as LoRA, to further optimize the memory and power required for client\-side training\.

## Sources

- [Story source](<https://huggingnews.com/ai/developer-proves-llm-fine-tuning-possible-in-browser-using-webgpu-48af825b>)
- [Story source](<https://x.com/ngxson/status/2096379210939994573>)
- [Supporting source](<https://x.com/maximelabonne/status/2096456605076476239>)