Tianqi Chen

Thanks @slyubomirsky . To dig a bit further on the FuncType and FuncStructInfo. One interesting direction is to explore possibility to not go with the function subtyping rule (that is...

Agree it is important to preserve call result(and we can afford to do so because erasure of call result goes with the same direction of subtyping). The main challenge lies...

Agreed, the goal is to NOT simplify rules for determining call results -- so results are precise. But relax the rules for checking call parameters -- mainly because of the...

Just want to provide some context about the potential design confliction of erasing and subfunction typing. We relies on erasure when we unify the values from different branches, or when...

Great thoughts for discussions. The main reason why we had a `PackedFuncType` is that the original FuncType does not support opaque function. Ideally we should refactor `FuncType` to be able...

agree, for private functions, the renaming can happen through adding numbering suffix. By default we should not rename external functions. Because the users only interact with external functions, in theory...

make a high-level remark here. The emscripten generate module have some conditional blocks for nodejs (this is default behavior of emscripten) and it does not work well with some of...

I think webllm would needs its own mechanism. There are a few things. - First of all, check whether webgpu buffer creation can throw an error, or caught by error...

Sorry for chiming in late. For the caching layer, ideally we would like something that comes with minimal dependency. Spefically, we should: - Make ArtifactCache https://github.com/apache/tvm/blob/main/web/src/runtime.ts#L991 an interface `ArtifactCache` in...

If you would only like to use webllm, you do not need to run `npm run build`, instead, follow instructions in examples to directly install WebLLM from npm. See https://github.com/mlc-ai/web-llm/tree/main/examples/get-started...