No need to contiguous

andrewor14 · andrewor14 · commit a2c07ef8cc0c · 2025-10-29T13:12:13.000-07:00
diff --git a/torchao/quantization/quantize_/workflows/float8/float8_tensor.py b/torchao/quantization/quantize_/workflows/float8/float8_tensor.py
@@ -341,7 +341,6 @@ def _float8_linear_impl(
                 assert _is_rowwise_scaled(input_tensor), (
                     "Input tensor must be rowwise block size"
                 )
-                wq = wq.contiguous()
                 res = torch.ops.fbgemm.f8f8bf16_rowwise(
                     xq,
                     wq,

Original file line number	Diff line number	Diff line change
`@@ -341,7 +341,6 @@ def _float8_linear_impl(`
`341`	`341`	`assert _is_rowwise_scaled(input_tensor), (`
`342`	`342`	`"Input tensor must be rowwise block size"`
`343`	`343`	`)`
`344`		`- wq = wq.contiguous()`
`345`	`344`	`res = torch.ops.fbgemm.f8f8bf16_rowwise(`
`346`	`345`	`xq,`
`347`	`346`	`wq,`