这个是因为显存不够还是什么？

│ /home/ubuntu/.cache/huggingface/modules/transformers_modules/fnlp/moss-moon- │
│ 003-sft/dc39155402a323c3ad85ac9c4cb02556c61594f1/modeling_moss.py:53 in │
│ apply_rotary_pos_emb │
│ │
│ 50 │
│ 51 # Copied from transformers.models.gptj.modeling_gptj.apply_rotary_pos_ │
│ 52 def apply_rotary_pos_emb(tensor: torch.Tensor, sin: torch.Tensor, cos: │
│ ❱ 53 │ sin = torch.repeat_interleave(sin[:, :, None, :], 2, 3) │
│ 54 │ cos = torch.repeat_interleave(cos[:, :, None, :], 2, 3) │
│ 55 │ return (tensor * cos) + (rotate_every_two(tensor) * sin) │
│ 56 │
╰──────────────────────────────────────────────────────────────────────────────╯
RuntimeError: CUDA error: device-side assert triggered
CUDA kernel errors might be asynchronously reported at some other API call, so
the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1.
Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.

Start with the apply_rotary_pos_emb entry point at the referenced modeling_moss.py:53 and rerun with CUDA_LAUNCH_BLOCKING=1, as the traceback suggests. Determine whether the failure is caused by GPU memory or a device-side assertion, then document a reproducible cause and confirm the corrected run completes without the CUDA error.