Unverified Commit 6534d2fc authored by zifeitong's avatar zifeitong Committed by GitHub
Browse files

Fix torch version check for SM100 mxfp4 (#22535)


Signed-off-by: default avatarZifei Tong <zifeitong@gmail.com>
Signed-off-by: default avatarmgoin <mgoin64@gmail.com>
Co-authored-by: default avatarmgoin <mgoin64@gmail.com>
parent 422f22e0
......@@ -741,12 +741,14 @@ class FusedMoE(torch.nn.Module):
# we padding globally so EP buffer allocation works
if quant_config and quant_config.get_name() == "mxfp4":
if not current_platform.is_device_capability(100):
if not is_torch_equal_or_newer("2.8.0"):
raise RuntimeError("Mxfp4 on hopper requires torch >= 2.8.0")
if current_platform.is_device_capability(
90) and not has_triton_kernels():
raise RuntimeError(
"Mxfp4 on non-blackwell requires torch >= 2.8.0")
if not has_triton_kernels():
raise NotImplementedError(
"Triton kernels must be installed for mxfp4 on hopper")
"triton_kernels must be installed for "
"mxfp4 on non-blackwell")
if (current_platform.is_rocm()
or envs.VLLM_USE_FLASHINFER_MOE_MXFP4_MXFP8
or envs.VLLM_USE_FLASHINFER_MOE_MXFP4_BF16):
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment