Ting Yu

icon location I am a Computer Science student at Beijing University of Posts and Telecommunications. I work on deep learning and computer vision.

Results 3 comments of Ting Yu

Support for TorchScript export

In ops/modules/ms_deform_attn.py， replace # output = MSDeformAttnFunction.apply( # value, input_spatial_shapes, input_level_start_index, sampling_locations, attention_weights, self.im2col_step) by: output = ms_deform_attn_core_pytorch(value, input_spatial_shapes, sampling_locations, attention_weights) It seems work for me.

post-training InternVL with RL objective using vllm engine and FSDP results in RuntimeError: The tensor has a non-zero number of elements, but its data is not allocated yet.

I get the same problem, don't know how to solve the problem. I'm trying to freeze the internvl vision-tower，as the repo mentions "https://github.com/Qsingle/verl/blob/e057c06067090d1ad058ccbee47c961a79a26453/verl/workers/fsdp_workers.py#L3" line 300.

post-training InternVL with RL objective using vllm engine and FSDP results in RuntimeError: The tensor has a non-zero number of elements, but its data is not allocated yet.

After > I get the same problem, don't know how to solve the problem. I'm trying to freeze the internvl vision-tower，as the repo mentions "https://github.com/Qsingle/verl/blob/e057c06067090d1ad058ccbee47c961a79a26453/verl/workers/fsdp_workers.py#L3" line 300. After freeze the...