Zihao Ye comments

Results 332 comments of


                                            Zihao Ye

[Bug] mlc-llm/cpp/conv_templates.cc:156: Unknown conversation template: dolly

You can follow this tutorial: https://mlc.ai/mlc-llm/docs/tutorials/compile-models.html

[Question] Is Chatglm currently not supported?

Not yet, we will have tutorials on how to add new model architectures like GLM, and you are welcomed to contribute.

[Bug] Relax repository should be a submodule

We had discussions on this before and part of the reason we didn't make it a submodule is to avoid duplication (people have already cloned relax and in that case,...

[Bug] Relax repository should be a submodule

Hi @tobrun already merged, thanks for your suggestions.

[Question] TVM version used in tvm wheels?

Hi @ziyu-guo , I think we have prebuilt wheels for M1 Mac: ``` pip install --pre mlc-ai-nightly -f https://mlc.ai/wheels ``` should work for you. If you really need to install...

[Feature Request] Is there any plan for stable diffusion running under this project?

Thanks for suggesting the paper, we'll read it and see how can we benefit from SnapFusion. Yes, TVM Unity could be used to deploy Stable-Diffusion models, we already have a...

[Question] Can i use customized transformer llm model?

@spectrometerHBH @jinhongyii are writing tutorials on adding new model architectures in MLC-LLM.

use end-to-end DGL scripts run featGraph

You might checkout this branch of DGL: https://github.com/kira-lin/dgl/tree/tvm_integration

use end-to-end DGL scripts run featGraph

I don't think Featgraph has better performance against cusparse for GCN on GPU, see table IV in the paper, since DGL uses cusparse, it's normal that you don't observe any...

use end-to-end DGL scripts run featGraph

Sorry I just noticed that you were using `dgl.sparse._CAPI_FG_LoadModule("../build/featgraph/libfeatgraph_kernels.so")` to use `featgraph` as backend, actually the integration was abandoned because TVM do not have native sparse support and we might...