search

Alpha-VLLM / LLaMA2-Accessory

An Open-source Toolkit for LLM Development
https://llama2-accessory.readthedocs.io/
Other
2.72k stars 176 forks source link

Tensor must be cuda and dense #199

Open bibibabibo26 opened 5 months ago

bibibabibo26 commented 5 months ago

hello, when run main_finetune.py till 238th row: for param in fsdp_ignored_parameters: dist.broadcast(param.data, src=dist.get_global_rank(fs_init.get_data_parallel_group(), 0), group=fs_init.get_data_parallel_group()) it throws an runtime error: 发生异常: RuntimeError Tensors must be CUDA and dense File "/amax/yt26/.conda/envs/accessory/lib/python3.10/site-packages/torch/distributed/distributed_c10d.py", line 1570, in broadcast work = group.broadcast([tensor], opts) File "/amax/yt26/.conda/envs/accessory/lib/python3.10/site-packages/torch/distributed/distributed_c10d.py", line 1451, in wrapper return func(*args, **kwargs) File "/amax/yt26/VCM/LLaMA2-Accessory/accessory/main_finetune.py", line 238, in main dist.broadcast(param.data, src=dist.get_global_rank(fs_init.get_data_parallel_group(), 0), File "/amax/yt26/VCM/LLaMA2-Accessory/accessory/main_finetune.py", line 369, in main(args) RuntimeError: Tensors must be CUDA and dense how can I deal with that? thank you.

bibibabibo26 commented 5 months ago

I find the reason, I didn't config the -quat