A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 9, 18:15 UTC
[Hardware][AMD][CI][Bugfix] Fix Kernels Attention Cache test (#32904)
Signed-off-by: Matthew Wong <Matthew.Wong2@amd.com>
Matt committed
Jan 23, 2026 at 10:24 UTC
305e53ade8f7cbdbece6fd211eb45a612800e657
1 file changed
+1
-1
tests/kernels/attention/test_cache.py
+1
-1
@@ -242,7 +242,7 @@ def test_reshape_and_cache_flash(
242
value_cache_compact = permute_and_compact(value_cache)
243
244
def convert_fp8_local(output, input, scale, kv_dtype):
245
- fp8_input = input.view(torch.float8_e4m3fn)
245
+ fp8_input = input.view(current_platform.fp8_dtype())
246
if scale.numel() == 1: # per-tensor
247
result = scaled_dequantize(
248
fp8_input.flatten(0, 2), scale, group_shape=None, out_dtype=output.dtype