wrong number of dimensions

This commit is contained in:
Francesco Bertolotti 2026-01-12 16:19:43 +01:00 committed by GitHub
commit eaf3f932e0

View file

@ -128,7 +128,7 @@ def _DWf_DW_dfg_kernel(
def swiglu_DWf_DW_dfg_kernel(DW, e, g):
batch_seq_len, hd = e.shape
batch, seq_len, hd = e.shape
n_elements = e.numel()
grid = lambda meta: (triton.cdiv(n_elements, meta["BLOCK_SIZE"]),)
with torch_gpu_device(e.device):