Disable mypy arg-type and assignment checks in tests (#59428)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
co-authored by
Claude Opus 5.5
parent
17e9295dd5
commit
fc2c801ad9
@@ -259,7 +259,7 @@ def test_w8a8_block_fp8_torch_scaled_mm_matmul():
|
||||
stub = BlockWiseTorchFP8ScaledMMLinearKernel.__new__(
|
||||
BlockWiseTorchFP8ScaledMMLinearKernel
|
||||
)
|
||||
stub.config = types.SimpleNamespace(out_dtype=out_dtype) # type: ignore[assignment]
|
||||
stub.config = types.SimpleNamespace(out_dtype=out_dtype)
|
||||
out = stub.apply_block_scaled_mm(
|
||||
A_fp8_cuda.cuda(), B_fp8.cuda(), As_cuda.cuda(), Bs.cuda()
|
||||
)
|
||||
|
||||
@@ -208,7 +208,7 @@ def mm_test_helper(
|
||||
a=tensors.a,
|
||||
b_q=tensors.w_q,
|
||||
b_group_scales=tensors.w_g_s,
|
||||
b_group_size=group_size, # type: ignore[arg-type]
|
||||
b_group_size=group_size,
|
||||
b_channel_scales=tensors.w_ch_s,
|
||||
a_token_scales=tensors.w_tok_s,
|
||||
)
|
||||
|
||||
@@ -525,15 +525,13 @@ def test_check_marlin_supports_layer_allow_tile_padding():
|
||||
)
|
||||
|
||||
# Tile-misaligned but group-aligned: rejected strictly, allowed w/ padding
|
||||
layer: LinearBase = _FakeLinear( # type: ignore[assignment]
|
||||
4640, 512, input_size=2048
|
||||
)
|
||||
layer: LinearBase = _FakeLinear(4640, 512, input_size=2048)
|
||||
assert not check_marlin_supports_layer(layer, 128)
|
||||
assert check_marlin_supports_layer(layer, 128, allow_tile_padding=True)
|
||||
assert check_marlin_supports_layer(layer, -1, allow_tile_padding=True)
|
||||
|
||||
# A group straddling the TP shard cannot be fixed by padding
|
||||
layer = _FakeLinear(4608, 4672, input_size=18688) # type: ignore[assignment]
|
||||
layer = _FakeLinear(4608, 4672, input_size=18688)
|
||||
assert not check_marlin_supports_layer(layer, 128, allow_tile_padding=True)
|
||||
|
||||
|
||||
|
||||
@@ -206,7 +206,7 @@ def test_nvfp4_emulation_support_check_rejects_bias_and_lora(
|
||||
in_dtype=torch.bfloat16,
|
||||
device="cuda",
|
||||
routing_method=RoutingMethodType.TopK,
|
||||
**config_kwargs, # type: ignore[arg-type]
|
||||
**config_kwargs,
|
||||
)
|
||||
|
||||
supported, reason = Nvfp4QuantizationEmulationTritonExperts.is_supported_config(
|
||||
|
||||
Reference in New Issue
Block a user