webtorch-models / README.md
9bow's picture
Add WebTorch bundle documentation
9dbb66e verified
|
Raw History Blame Contribute Delete
1.53 kB
---
license: other
library_name: webtorch
tags:
- webgpu
- webtorch
- browser
- text-generation
---
# WebTorch browser bundles
This repository hosts exported, hash-verified WebTorch bundles used by the
[WebTorch browser playground](https://9bow.github.io/webtorch/). They are static
runtime artifacts, not unmodified Transformers checkpoints. Each directory keeps
its manifest, graphs, WGSL kernels, tokenizer files, and weight shards together so
the browser can load every resource relative to `manifest.json`.
The bundles are experimental and demonstrate reproducibility on tested WebGPU
adapters. They do not certify model quality, safety, production readiness, or broad
hardware compatibility. Quantized profiles are optional diagnostics unless their
directory status says otherwise.
## Licenses
Each bundle remains subject to its upstream model license. Qwen-derived bundles use
the relevant Qwen model license. EXAONE-derived bundles remain subject to the EXAONE
Model License Agreement, including its noncommercial restrictions where applicable.
The WebTorch runtime source is licensed separately in the
[9bow/webtorch](https://github.com/9bow/webtorch) repository.
## Loading
Use a public manifest URL such as:
```text
https://huggingface.co/9bow/webtorch-models/resolve/main/qwen35-08b-fp32-8k-token-major-home/manifest.json
```
The browser downloads the manifest and all manifest-relative assets directly from
this public repository. A raw `model.safetensors` file cannot replace the exported
bundle.