remove qwen2.py llama.py fix llama output

2025-12-31 11:31:26 +08:00
parent b3c30a3cb9
commit b015bb76fd
11 changed files with 65 additions and 1263 deletions
--- a/vllm_kunlun/ops/activation.py
+++ b/vllm_kunlun/ops/activation.py
@@ -98,7 +98,7 @@ class SiluAndMul(CustomOp):
        d = x.shape[-1] // 2
        output_shape = (x.shape[:-1] + (d, ))
        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
-        torch.ops._C.swiglu(x, out)
+        torch.ops._C.silu_and_mul(out, x)
        return out

    def forward_kunlun(self, x: torch.Tensor) -> torch.Tensor: