diff --git a/unsloth/kernels/swiglu.py b/unsloth/kernels/swiglu.py index b321f5179e..9e2680e862 100644 --- a/unsloth/kernels/swiglu.py +++ b/unsloth/kernels/swiglu.py @@ -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):