debug: disable kernel cache to test fresh compilation
This commit is contained in:
@@ -355,7 +355,7 @@ def run_nvfp4_grouped_gemm(
|
||||
N_packed = mat_b.shape[2]
|
||||
cache_key = (num_experts, str(device), mma_tiler_mn, cluster_shape_mn, K_packed, N_packed)
|
||||
|
||||
if cache_key not in _compiled_kernel_cache:
|
||||
if False and cache_key not in _compiled_kernel_cache: # TEMP: always recompile
|
||||
kernel = ScaledGroupedGemmKernel(
|
||||
scenario="2Dx3D",
|
||||
sf_vec_size=SF_VEC_SIZE,
|
||||
|
||||
Reference in New Issue
Block a user