Support 45t fp8 8 GPU (#3659)

2025-10-30 11:26:39 +08:00 · 2025-08-28 10:52:53 +08:00
parent 7afcd4b776
commit 76513f6416
3 changed files with 23 additions and 12 deletions
--- a/fastdeploy/model_executor/layers/moe/triton_moe_kernels.py
+++ b/fastdeploy/model_executor/layers/moe/triton_moe_kernels.py
@@ -14,14 +14,11 @@
 # limitations under the License.
 """

+import triton
 import triton.language as tl

-from fastdeploy.model_executor.ops.triton_ops.triton_utils_v2 import (
-    paddle_use_triton_v2,
-)

-
-@paddle_use_triton_v2()
+@triton.jit()
 def fused_moe_kernel_paddle(
    a_ptr,
    b_ptr,