Skip to content

flashinfer.decode.single_decode_with_kv_cache: Floating point exception (core dumped) #1027

Open
@MenHimChan

Description

@MenHimChan

Hi, i met an error when i run flashinfer.decode.single_decode_with_kv_cache example program, which is :

import torch
import flashinfer
kv_len = 4096
num_qo_heads = 32
num_kv_heads = 32
head_dim = 128
q = torch.randn(num_qo_heads, head_dim).half().to("cuda:0")
k = torch.randn(kv_len, num_kv_heads, head_dim).half().to("cuda:0")
v = torch.randn(kv_len, num_kv_heads, head_dim).half().to("cuda:0")
o = flashinfer.single_decode_with_kv_cache(q, k, v)
o.shape

The error message is: Floating point exception (core dumped).
What reason casues this error?

My environment is:

  1. GPU: RTX 3060
  2. torch vesrsion : Torch 2.6
  3. cuda version : CUDA 12.4
  4. Flashinfer version: 0.2.5+cu124torch2.6

Looking forward to your relpy!

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions