Inference failed with "RuntimeError: cutlassF: no kernel found to launch!"

[NuanBaobao]

Issue Title: RuntimeError: cutlassF: No kernel found to launch in sample_t2v.py

Description:

I encountered an error when running the code in opensora/sample/sample_t2v.py. The specific segment of code is as follows:

with torch.backends.cuda.sdp_kernel(enable_math=False, enable_flash=False, enable_mem_efficient=True):
    hidden_states = F.scaled_dot_product_attention(
        query, key, value, attn_mask=attention_mask, dropout_p=0.0, is_causal=False
    )

The error message I receive is:

RuntimeError: cutlassF: no kernel found to launch!

Environment Details:

• GPU: NVIDIA V100
• PyTorch version: 2.1.0
• CUDA version: 12.1

Request for Assistance: Could you please help me troubleshoot this issue? Any guidance on resolving this error would be greatly appreciated.

Thank you!