Jinjing Zhou

We had some changes in the latest version. VBASE is recommended instead of prefilter. We'll have a new PR for this later.

@oocococo You can copy the function content directly to your file as a temporary test. ``` def tensorrt(os="20.04", cuda="11.6.2", trt="8.4.3.1"): """Install tensorrt Args: os (Optional[str]): os version cuda (Optional[str]): cuda...

It's just python function. You can copy that as declaration of function and call `tensorrt(....)` directly in your build function.

How to merge different different stages like llb.Merge? Also I think [bramble](https://github.com/maxmcd/bramble)'s syntax can be an option

Some examples from bramble https://github.com/maxmcd/bramble/blob/eea4aee51e6ad881166412d61190012fb0d97c56/internal/project/testdata/project/default.bramble

I think we should have an idea about what the llb graph will look like after more dependency is set by the user (such as gcc and pypi packages), that...

I'm a bit concerned about the current proposal. The current design is detail-oriented, which is more complicated than original design. Also current design looks similar to llb, we can also...

Thanks. I've applied to join PyTorch slack channel through the form at PyTorch's README. My email address is zhoujinjing09 (gmail). Could you invite me if possible? I have experiences implementing...

Some differences between ibv found so far: - EFA doesn't require the memory registration process. The bounce buffer/cached memory registration is managed by libfabric, which can simplify the whole process...

@lw Thanks a lot. Actually I'm not so familiar with the underlying implementation of EFA. Here's something I found at https://ofiwg.github.io/libfabric/v1.13.0/man/fi_efa.7.html about the configuration, which partially shows how the underlying...