WebLLM 0.2.87, MLC AI, Apache-2.0 (LICENSE).
Official runtime: https://github.com/mlc-ai/web-llm
Model: Qwen3-0.6B-q4f16_1-MLC, Qwen Team / MLC AI, Apache-2.0.
Pinned model revision: 8c14ce481d4c692769976ad52afea453a102df19.
https://huggingface.co/mlc-ai/Qwen3-0.6B-q4f16_1-MLC
Browser model library: Qwen3-0.6B-q4f16_1_cs1k-webgpu.wasm, WebLLM v0_2_87/base.
Original source: https://raw.githubusercontent.com/mlc-ai/binary-mlc-llm-libs/main/web-llm-models/v0_2_87/base/Qwen3-0.6B-q4f16_1_cs1k-webgpu.wasm
The model library is served locally as an immutable release asset; model weights remain on Hugging Face.
Weights: 335,372,288 bytes across nine shards. Tokenizer, model config, wasm and JS add overhead (~360 MB total, depending on transport compression).
No paid API, server inference, credentials, or connection to the owner's computer. Inference runs in each visitor's browser.
Legacy CPU fallback: see ../ai/NOTICE.txt. Fallback does not silently download a second model after GPU failure; the visitor chooses compatibility retry.
