Type alias ModelSpec

ModelSpec: {
    contextWindowSize?: number;
    modelLibWasmUrl: string;
    modelWeightsConfigUrl: string;
}

Specifies where to retrieve model weights and configuration

Optional contextWindowSize?: number

Size of the context window, overrides the number given in config metadata
modelLibWasmUrl: string

URL of the compiled wasm for model. For example:

https://raw.githubusercontent.com/mlc-ai/binary-mlc-llm-libs/main/Llama-2-7b-chat-hf-q4f32_1-webgpu.wasm
modelWeightsConfigUrl: string

URL of root of repo containing ndarray-cache.json and mlc-chat-config.json For example:

https://huggingface.co/mlc-ai/mlc-chat-Llama-2-7b-chat-hf-q4f32_1/resolve/main/

Remarks
For hugging face, this root currently returns 404 -- that's fine, as long as the files themselves can be fetched. Do not provide /tree/main URL's.