Skip to content

vllm.models.qwen4_exp.amd.low_latency_gemm

Qwen4Exp low-latency GEMM hook for AMD ROCm.

Functions:

enable_qwen4_exp_low_latency_gemm(module, dtype)

Keep the standard vLLM linear methods on AMD ROCm.

Source code in vllm/models/qwen4_exp/amd/low_latency_gemm.py 
 9
10
11
12
13
14
15
def enable_qwen4_exp_low_latency_gemm(
    module: nn.Module,
    dtype: torch.dtype,
) -> None:
    """Keep the standard vLLM linear methods on AMD ROCm."""

    del module, dtype