About the checkpoint

[rentainhe]

Hi authors! Thanks so much for releasing such a great work! I was wondering would you like to provide the pytorch checkpoints (as .pth files) instead of the onnx checkpoints?

[Zailushang211]

�[0;93m2024-11-08 07:12:36.963576075 [W:onnxruntime:, transformer_memcpy.cc:74 ApplyImpl] 48 Memcpy nodes are added to the graph main_graph for CUDAExecutionProvider. It might have negative impact on performance (including unable to run CUDA graph). Set session_options.log_severity_level=1 to see the detail logs before this message.�[m
�[0;93m2024-11-08 07:12:36.974287246 [W:onnxruntime:, session_state.cc:1166 VerifyEachNodeIsAssignedToAnEp] Some nodes were not assigned to the preferred execution providers which may or may not have an negative impact on performance. e.g. ORT explicitly assigns shape related ops to CPU to improve perf.�[m
�[0;93m2024-11-08 07:12:36.974298274 [W:onnxruntime:, session_state.cc:1168 VerifyEachNodeIsAssignedToAnEp] Rerunning with verbose output on a non-minimal build will show node assignments.�[m
�[0;93m2024-11-08 07:12:37.634942805 [W:onnxruntime:, session_state.cc:1166 VerifyEachNodeIsAssignedToAnEp] Some nodes were not assigned to the preferred execution providers which may or may not have an negative impact on performance. e.g. ORT explicitly assigns shape related ops to CPU to improve perf.�[m
�[0;93m2024-11-08 07:12:37.634962319 [W:onnxruntime:, session_state.cc:1168 VerifyEachNodeIsAssignedToAnEp] Rerunning with verbose output on a non-minimal build will show node assignments.�[m
0.500493049621582
got warning as above with onnx, and the speed is even slower than on cpu, takes almost twice the time. is it a problem with onnx?

[qjadud1994]

Thank you for your interest in our work.

We are considering releasing Pytorch checkpoint (but it might take some time).

[qjadud1994]

�[0;93m2024-11-08 07:12:36.963576075 [W:onnxruntime:, transformer_memcpy.cc:74 ApplyImpl] 48 Memcpy nodes are added to the graph main_graph for CUDAExecutionProvider. It might have negative impact on performance (including unable to run CUDA graph). Set session_options.log_severity_level=1 to see the detail logs before this message.�[m �[0;93m2024-11-08 07:12:36.974287246 [W:onnxruntime:, session_state.cc:1166 VerifyEachNodeIsAssignedToAnEp] Some nodes were not assigned to the preferred execution providers which may or may not have an negative impact on performance. e.g. ORT explicitly assigns shape related ops to CPU to improve perf.�[m �[0;93m2024-11-08 07:12:36.974298274 [W:onnxruntime:, session_state.cc:1168 VerifyEachNodeIsAssignedToAnEp] Rerunning with verbose output on a non-minimal build will show node assignments.�[m �[0;93m2024-11-08 07:12:37.634942805 [W:onnxruntime:, session_state.cc:1166 VerifyEachNodeIsAssignedToAnEp] Some nodes were not assigned to the preferred execution providers which may or may not have an negative impact on performance. e.g. ORT explicitly assigns shape related ops to CPU to improve perf.�[m �[0;93m2024-11-08 07:12:37.634962319 [W:onnxruntime:, session_state.cc:1168 VerifyEachNodeIsAssignedToAnEp] Rerunning with verbose output on a non-minimal build will show node assignments.�[m 0.500493049621582 got warning as above with onnx, and the speed is even slower than on cpu, takes almost twice the time. is it a problem with onnx?

I guess it is a problem with onnx.
Please follow the instructions in the onnxruntime installation docs.

[rentainhe]

Thank you for your interest in our work.

We are considering releasing Pytorch checkpoint (but it might take some time).

Thank you so much, we are really looking forward to the PyTorch weight release.