xverse-ai / XVERSE-13B

XVERSE-13B: A multilingual large language model developed by XVERSE Technology Inc.
Apache License 2.0
649 stars 58 forks source link

MPS does not support cumsum op with int64 input #15

Closed xuwu125 closed 1 year ago

xuwu125 commented 1 year ago

To create a public link, set share=True in launch(). /Users/xxxx/.pyenv/versions/3.10.6/lib/python3.10/site-packages/transformers/generation/utils.py:742: UserWarning: MPS: no support for int64 repeats mask, casting it to int32 (Triggered internally at /Users/runner/work/pytorch/pytorch/pytorch/aten/src/ATen/native/mps/operations/Repeat.mm:236.) input_ids = input_ids.repeat_interleave(expand_size, dim=0) Exception in thread Thread-7 (generate): Traceback (most recent call last): File "/Users/xxxx/.pyenv/versions/3.10.6/lib/python3.10/threading.py", line 1016, in _bootstrap_inner self.run() File "/Users/xxxx/.pyenv/versions/3.10.6/lib/python3.10/threading.py", line 953, in run self._target(*self._args, *self._kwargs) File "/Users/xxxx/.pyenv/versions/3.10.6/lib/python3.10/site-packages/torch/utils/_contextlib.py", line 115, in decorate_context return func(args, kwargs) File "/Users/xxxx/.pyenv/versions/3.10.6/lib/python3.10/site-packages/transformers/generation/utils.py", line 1618, in generate return self.sample( File "/Users/xxxx/.pyenv/versions/3.10.6/lib/python3.10/site-packages/transformers/generation/utils.py", line 2737, in sample model_inputs = self.prepare_inputs_for_generation(input_ids, model_kwargs) File "/Users/xxxx/.cache/huggingface/modules/transformers_modules/modeling_xverse.py", line 800, in prepare_inputs_for_generation position_ids = attention_mask.long().cumsum(-1) - 1 RuntimeError: MPS does not support cumsum op with int64 input

请问如何解决,python ./chat_demo.py --port='8989' --model_path= 这种方式启动的,模型是下载的 https://huggingface.co/xverse/XVERSE-13B-Chat/tree/main

xuwu125 commented 1 year ago

电脑是m2 芯片的

underspirit commented 1 year ago

这是pytorch算子支持的问题, 可以去pytorch社区查找相关解决方案, 比如 https://github.com/pytorch/pytorch/issues/96610