Weijie Su comments

Results 5 comments of


                                            Weijie Su

caffe make fails on Debian 9 VM [CXX src/caffe/layers/beam_search_node_layer.cpp]

> Change (Remove explicit Argument Types): > **make_pair(partial_sum[src], ix + end_of_sequence_)** > to > **make_pair(partial_sum[src], ix + end_of_sequence_)** > and similarly at other places in the file. > > It...

Input type (torch.cuda.FloatTensor) and weight type (torch.cuda.HalfTensor) should be the same

Facing the same error while using 'O2', have you found the solution?@BenQLange I found this error only appeared when I run code with multiple GPUs, it is ok while using...

Input type (torch.cuda.FloatTensor) and weight type (torch.cuda.HalfTensor) should be the same

In my case, the reason is that I used a non-official SyncBatchNorm implementation and it seems apex couldn't deal with it.

The problem of multitasking parallel processing

@Pter61 What config do you run? And what is the full command you ran? And please provide your pytorch version and cuda version also.

More details about OSWorld running script.

> Hi, the argument values are here: https://github.com/xlang-ai/OSWorld/blob/main/mm_agents/uitars_agent.py#L404 https://github.com/xlang-ai/OSWorld/blob/884676cebc4300983cf1285177b30dcc8ef25746/mm_agents/uitars_agent.py#L413 The observation is screenshot + a11y tree? It is confusing since UI-TARS is trained on screenshot-only observation.