LukeLin

replicated with https://github.com/open-compass/opencompass/pull/1985

Same problem,

I have another problem, Can scripts/demo/turbo.py support f16? When I use f32 model it will OOM. I only have 12G VRAM

here is one have https://huggingface.co/datasets/bhatvineet/modified_libero_rlds_cotdep/tree/main/libero_10_cotdep/1.0.0 . libero spatial/goal/object has depth But libero 10 cannot use . ``` ValueError: Cannot infer argument `num` ERROR | >> [*] depth_wrist: shape=, dtype= ```

I use `python3 benchmarks/image_to_video.py --model ~/.cache/huggingface/hub/models--siliconflow--stable-video-diffusion-img2vid-xt-int8/snapshots/37b3b3943ce72c5c0b2ea827072aa6a22c1f8cc3/ --deepcache --input-image output_image.png --output-video output_image.mp4` It has Loading pipeline components...: 40%|█████████████████████▏ | 2/5 [00:00

And I want to know how to install `from onediffx import compile_pipe, OneflowCompileOptions` `Traceback (most recent call last): File "/root/share/onediff/benchmarks/image_to_video.py", line 46, in from onediffx import compile_pipe, OneflowCompileOptions ModuleNotFoundError: No...

I use https://github.com/GeWu-Lab/DepthHelps-IROS2024/tree/master reproduce it and add depth, it v1 has 199GB, 199G ./v1 94G ./v0 https://huggingface.co/juyil/libero_depth_hdf5/tree/main

It seems not affect evaluation.

> [@jpli02](https://github.com/jpli02) Thanks a lot, this works for me. But may I sincerely ask why this operation works? It looks like cuda 12 is install in conda But flashattention cannot...

smolvla said it has libero and metaworld, does these eval code open source? https://arxiv.org/abs/2506.01844