Li Ruixiao comments

Results 6 comments of


                                            Li Ruixiao

[Feature Request] Add additional debug information for `Triton Error [CUDA]: device kernel image is invalid`

This seems to be related to the version of the CUDA driver. Have you tried upgrading the CUDA driver to a higher version?

[Feature Request] Add additional debug information for `Triton Error [CUDA]: device kernel image is invalid`

For my configuration (triton 2.2.0, cuda 12.2, nvcc 11.8 on A100), this error is resolved by upgrading the nvidia driver from 470.57.02 to 535.104.05. the minimum version required by *cuda*...

Incorrect result of tl.max() in Triton2.2.0

Could you please provide some code snippets that can reliably reproduce the issue？ 🙂

Optimizing Shared Memory Usage

Any progress? I am quite interested in this `Shared Memory Usage` issue. Would it be possible for you to share your code?

[BUG] Is mcore-0.12.0 checkpoint resume correct?

maybe this could help? i'm not sure🤔 https://github.com/alibaba/Pai-Megatron-Patch/tree/main/megatron_patch/fixes/optimizer_offloading

when I use async vllm rollout in dapo get ERROR

same issues, using verl commit id `c5b189a1af496d0bc68320cd1d5bd7a1f1e3638a`