[ROCm][CI] fix test_rocm_quick_reduce.py (#49234)
Signed-off-by: charlifu <charlifu@amd.com>
Charlie Fu committed
Jul 20, 2026 at 16:39 UTC
5feb3950e529278b1f9dbe0a4071470d3d382a3d
1 file changed
+13
-3
tests/distributed/test_rocm_quick_reduce.py
+13
-3
@@ -60,6 +60,8 @@ def _make_quick_allreduce(
60
qar.use_fp16_kernels = use_fp16_kernels
61
qar.qr_quant_level = QuickReduceRegime[quant_level]
62
qar.qr_max_size = qr_max_size
63
+ qar.qr_min_size = None
64
+ qar.qr_quantization_min_size = None
65
return qar
66
67
@@ -511,13 +513,21 @@ def test_quick_reduce_regime_values():
513
assert QuickReduceRegime.INT8.value == 1
514
assert QuickReduceRegime.INT6.value == 2
515
assert QuickReduceRegime.INT4.value == 3
514
- assert QuickReduceRegime.NONE.value == 4
516
+ assert QuickReduceRegime.INT3.value == 4
517
+ assert QuickReduceRegime.NONE.value == 5
518
519
520
def test_quick_reduce_regime_names():
521
from vllm.distributed.device_communicators.quick_all_reduce import QuickReduceRegime
522
520
- assert set(QuickReduceRegime.__members__) == {"FP", "INT8", "INT6", "INT4", "NONE"}
523
+ assert set(QuickReduceRegime.__members__) == {
524
+ "FP",
525
+ "INT8",
526
+ "INT6",
527
+ "INT4",
528
+ "INT3",
529
+ "NONE",
530
+ }
531
532
533
@pytest.mark.parametrize("quant_level", QUANT_LEVELS + ["NONE"])
@@ -693,7 +703,7 @@ def test_quick_allreduce_min_size_table():
703
for dtype in [torch.float16, torch.bfloat16]:
704
for world_size in QuickAllReduce._SUPPORTED_WORLD_SIZES:
705
min_sizes = QuickAllReduce._QR_MIN_SIZE[(dtype, world_size)]
696
- assert len(min_sizes) == 4
706
+ assert len(min_sizes) == 5
707
assert all(size > 0 for size in min_sizes)
708
709