335 lines
15 KiB
Python
335 lines
15 KiB
Python
from unittest.mock import Mock, patch
|
|
|
|
import regex as re
|
|
import torch
|
|
|
|
from tests.ut.base import TestBase
|
|
from vllm_ascend.ascend_forward_context import MoECommType
|
|
from vllm_ascend.quantization.methods.w4a16 import AscendW4A16FusedMoEMethod, pack_to_int32, unpack_from_int32
|
|
|
|
|
|
class TestUnpackFromInt32(TestBase):
|
|
def test_unpack_from_int32_restores_values_and_crops_padding(self):
|
|
weight = torch.tensor([[0x76543210]], dtype=torch.int32)
|
|
shape = torch.Size([1, 6])
|
|
|
|
result = unpack_from_int32(weight, shape, num_bits=4, packed_dim=1)
|
|
|
|
self.assertEqual(result.dtype, torch.int8)
|
|
self.assertEqual(result.shape, shape)
|
|
self.assertTrue(torch.equal(result, torch.tensor([[-8, -7, -6, -5, -4, -3]], dtype=torch.int8)))
|
|
|
|
def test_unpack_from_int32_packed_dim_1(self):
|
|
weight = torch.tensor([[305419896, -1420531520]], dtype=torch.int32)
|
|
shape = torch.Size([1, 8])
|
|
num_bits = 4
|
|
|
|
result = unpack_from_int32(weight, shape, num_bits, packed_dim=1)
|
|
|
|
self.assertEqual(result.dtype, torch.int8)
|
|
self.assertEqual(result.shape, shape)
|
|
|
|
def test_unpack_from_int32_packed_dim_0(self):
|
|
weight = torch.tensor([[305419896], [-1420531520]], dtype=torch.int32)
|
|
shape = torch.Size([8, 1])
|
|
num_bits = 4
|
|
|
|
result = unpack_from_int32(weight, shape, num_bits, packed_dim=0)
|
|
|
|
self.assertEqual(result.dtype, torch.int8)
|
|
self.assertEqual(result.shape, shape)
|
|
|
|
def test_unpack_from_int32_packed_dim_0_restores_values(self):
|
|
weight = torch.tensor([[0x76543210]], dtype=torch.int32)
|
|
shape = torch.Size([6, 1])
|
|
|
|
result = unpack_from_int32(weight, shape, num_bits=4, packed_dim=0)
|
|
|
|
self.assertEqual(result.dtype, torch.int8)
|
|
self.assertEqual(result.shape, shape)
|
|
expected = torch.tensor([[-8], [-7], [-6], [-5], [-4], [-3]], dtype=torch.int8)
|
|
self.assertTrue(torch.equal(result, expected))
|
|
|
|
def test_unpack_from_int32_assertion_dtype_message(self):
|
|
weight = torch.tensor([[1, 2]], dtype=torch.int64)
|
|
message = "Expecting `weight.dtype` is torch.int32 but got torch.int64."
|
|
|
|
with self.assertRaisesRegex(AssertionError, re.escape(message)):
|
|
unpack_from_int32(weight, torch.Size([8, 1]), 4)
|
|
|
|
def test_unpack_from_int32_assertion_num_bits_positive_message(self):
|
|
weight = torch.tensor([[1, 2]], dtype=torch.int32)
|
|
message = "Expecting `num_bits` should be positive but got 0."
|
|
|
|
with self.assertRaisesRegex(AssertionError, re.escape(message)):
|
|
unpack_from_int32(weight, torch.Size([8, 1]), 0)
|
|
|
|
def test_unpack_from_int32_assertion_num_bits_upper_bound_message(self):
|
|
weight = torch.tensor([[1, 2]], dtype=torch.int32)
|
|
message = "Expecting `num_bits` should not be larger than 8 but got 16."
|
|
|
|
with self.assertRaisesRegex(AssertionError, re.escape(message)):
|
|
unpack_from_int32(weight, torch.Size([8, 1]), 16)
|
|
|
|
def test_unpack_from_int32_assertion_num_bits_divides_int32_message(self):
|
|
weight = torch.tensor([[1, 2]], dtype=torch.int32)
|
|
message = "Expecting `num_bits` 3 to divide 32 exactly."
|
|
|
|
with self.assertRaisesRegex(AssertionError, re.escape(message)):
|
|
unpack_from_int32(weight, torch.Size([8, 1]), 3)
|
|
|
|
def test_unpack_from_int32_assertion_packed_dim_message(self):
|
|
weight = torch.tensor([[1, 2]], dtype=torch.int32)
|
|
message = "Expecting `packed_dim` is 0 or 1 but got 2."
|
|
|
|
with self.assertRaisesRegex(AssertionError, re.escape(message)):
|
|
unpack_from_int32(weight, torch.Size([8, 1]), 4, packed_dim=2)
|
|
|
|
|
|
class TestPackToInt32(TestBase):
|
|
@patch("vllm_ascend.quantization.methods.w4a16.torch_npu.npu_convert_weight_to_int4pack")
|
|
def test_pack_to_int32_int8(self, mock_npu_convert_weight_to_int4pack):
|
|
mock_npu_convert_weight_to_int4pack.return_value = torch.zeros((2, 4), dtype=torch.int32)
|
|
|
|
weight = torch.zeros((2, 8, 16), dtype=torch.int8)
|
|
result = pack_to_int32(weight)
|
|
|
|
self.assertEqual(result.dtype, torch.int32)
|
|
mock_npu_convert_weight_to_int4pack.assert_not_called()
|
|
|
|
self.assertEqual(result.shape, torch.Size([2, 8, 4]))
|
|
|
|
@patch("vllm_ascend.quantization.methods.w4a16.torch_npu.npu_convert_weight_to_int4pack")
|
|
def test_pack_to_int32_int32(self, mock_npu_convert_weight_to_int4pack):
|
|
def mock_convert_weight(weight):
|
|
return weight
|
|
|
|
mock_npu_convert_weight_to_int4pack.side_effect = mock_convert_weight
|
|
weight = torch.zeros((2, 8, 8), dtype=torch.int32)
|
|
result = pack_to_int32(weight)
|
|
|
|
self.assertEqual(result.dtype, torch.int32)
|
|
self.assertEqual(result.shape, weight.shape)
|
|
|
|
def test_pack_to_int32_assertion_dim(self):
|
|
weight = torch.zeros((8, 8), dtype=torch.int8)
|
|
message = (
|
|
"Expecting `weight.dim()` is 3 ([expert, output_channel, input_channel] or "
|
|
"[expert, input_channel, output_channel]) but got 2."
|
|
)
|
|
|
|
with self.assertRaisesRegex(AssertionError, re.escape(message)):
|
|
pack_to_int32(weight)
|
|
|
|
def test_pack_to_int32_assertion_dtype(self):
|
|
weight = torch.zeros((2, 8, 8), dtype=torch.float32)
|
|
message = "Expecting `weight.dtype` is torch.int8 or torch.int32 but got torch.float32."
|
|
|
|
with self.assertRaisesRegex(AssertionError, re.escape(message)):
|
|
pack_to_int32(weight)
|
|
|
|
def test_pack_to_int32_assertion_int32_divisible_message(self):
|
|
weight = torch.zeros((2, 8, 7), dtype=torch.int32)
|
|
message = "the last dim of weight needs to be divided by 8."
|
|
|
|
with self.assertRaisesRegex(AssertionError, re.escape(message)):
|
|
pack_to_int32(weight)
|
|
|
|
def test_pack_to_int32_assertion_int8_divisible_message(self):
|
|
weight = torch.zeros((2, 8, 7), dtype=torch.int8)
|
|
message = "the last dim of weight needs to be divided by 4."
|
|
|
|
with self.assertRaisesRegex(AssertionError, re.escape(message)):
|
|
pack_to_int32(weight)
|
|
|
|
|
|
class TestAscendW4A16FusedMoEMethod(TestBase):
|
|
experts = 8
|
|
input_size = 32
|
|
output_size = 128
|
|
group_size = 32
|
|
|
|
@patch("vllm_ascend.quantization.methods.w4a16.get_ascend_config")
|
|
@patch("vllm_ascend.quantization.methods.w4a16.get_current_vllm_config")
|
|
def setUp(self, mock_get_current_vllm_config, mock_get_ascend_config):
|
|
mock_ascend_config = Mock()
|
|
mock_ascend_config.eplb_config.dynamic_eplb = False
|
|
mock_ascend_config.eplb_config.expert_map_record_path = None
|
|
mock_get_ascend_config.return_value = mock_ascend_config
|
|
|
|
mock_vllm_config = Mock()
|
|
mock_vllm_config.quant_config = Mock(
|
|
quant_description={
|
|
"group_size": self.group_size,
|
|
}
|
|
)
|
|
mock_get_current_vllm_config.return_value = mock_vllm_config
|
|
|
|
self.quant_method = AscendW4A16FusedMoEMethod()
|
|
|
|
def test_get_weight(self):
|
|
param_dict = self.quant_method.get_weight(self.experts, self.input_size, self.output_size, torch.bfloat16)
|
|
|
|
self.assertEqual(param_dict["w13_weight_packed"].dtype, torch.int32)
|
|
expected_w13_shape = (self.experts, 2 * self.input_size, self.output_size // self.quant_method.pack_factor)
|
|
self.assertEqual(param_dict["w13_weight_packed"].shape, expected_w13_shape)
|
|
|
|
self.assertEqual(param_dict["w2_weight_packed"].dtype, torch.int32)
|
|
expected_w2_shape = (self.experts, self.output_size, self.input_size // self.quant_method.pack_factor)
|
|
self.assertEqual(param_dict["w2_weight_packed"].shape, expected_w2_shape)
|
|
|
|
def test_get_weight_assertion_intermediate_size_message(self):
|
|
message = "Expecting `intermediate_size_per_partition` 33 can be divided by `pack_factor` 8"
|
|
|
|
with self.assertRaisesRegex(AssertionError, re.escape(message)):
|
|
self.quant_method.get_weight(self.experts, self.input_size + 1, self.output_size, torch.bfloat16)
|
|
|
|
def test_get_weight_assertion_hidden_sizes_message(self):
|
|
message = "Expecting `hidden_sizes` 129 can be divided by `pack_factor` 8"
|
|
|
|
with self.assertRaisesRegex(AssertionError, re.escape(message)):
|
|
self.quant_method.get_weight(self.experts, self.input_size, self.output_size + 1, torch.bfloat16)
|
|
|
|
def test_get_dynamic_quant_param(self):
|
|
param_dict = self.quant_method.get_dynamic_quant_param(
|
|
self.experts, self.input_size, self.output_size, torch.bfloat16
|
|
)
|
|
|
|
self.assertEqual(param_dict["w13_weight_scale"].dtype, torch.bfloat16)
|
|
expected_w13_scale_shape = (self.experts, 2 * self.input_size, self.output_size // self.group_size)
|
|
self.assertEqual(param_dict["w13_weight_scale"].shape, expected_w13_scale_shape)
|
|
|
|
self.assertEqual(param_dict["w2_weight_shape"].dtype, torch.int32)
|
|
self.assertEqual(param_dict["w2_weight_shape"].shape, (self.experts, 2))
|
|
|
|
self.assertEqual(param_dict["w13_weight_offset"].dtype, torch.bfloat16)
|
|
self.assertEqual(param_dict["w13_weight_offset"].shape, expected_w13_scale_shape)
|
|
|
|
def test_get_dynamic_quant_param_assertion_intermediate_size_message(self):
|
|
message = "Expecting `intermediate_size_per_partition` 33 can be divided by `group_size` 32"
|
|
|
|
with self.assertRaisesRegex(AssertionError, re.escape(message)):
|
|
self.quant_method.get_dynamic_quant_param(
|
|
self.experts, self.input_size + 1, self.output_size, torch.bfloat16
|
|
)
|
|
|
|
def test_get_dynamic_quant_param_assertion_hidden_sizes_message(self):
|
|
message = "Expecting `hidden_sizes` 129 can be divided by `group_size` 32"
|
|
|
|
with self.assertRaisesRegex(AssertionError, re.escape(message)):
|
|
self.quant_method.get_dynamic_quant_param(
|
|
self.experts, self.input_size, self.output_size + 1, torch.bfloat16
|
|
)
|
|
|
|
def build_layer(self):
|
|
"""Build a mock layer for testing"""
|
|
layer = torch.nn.Module()
|
|
|
|
w13_shape = (self.experts, 2 * self.input_size, self.output_size // self.quant_method.pack_factor)
|
|
w2_shape = (self.experts, self.output_size, self.input_size // self.quant_method.pack_factor)
|
|
|
|
layer.w13_weight_packed = torch.nn.Parameter(
|
|
torch.randint(-100, 100, w13_shape, dtype=torch.int32), requires_grad=False
|
|
)
|
|
layer.w2_weight_packed = torch.nn.Parameter(
|
|
torch.randint(-100, 100, w2_shape, dtype=torch.int32), requires_grad=False
|
|
)
|
|
|
|
w13_scale_shape = (self.experts, 2 * self.input_size, self.output_size // self.group_size)
|
|
w2_scale_shape = (self.experts, self.output_size, self.input_size // self.group_size)
|
|
|
|
layer.w13_weight_scale = torch.nn.Parameter(
|
|
torch.ones(w13_scale_shape, dtype=torch.bfloat16), requires_grad=False
|
|
)
|
|
layer.w2_weight_scale = torch.nn.Parameter(
|
|
torch.ones(w2_scale_shape, dtype=torch.bfloat16), requires_grad=False
|
|
)
|
|
|
|
layer.w13_weight_offset = torch.nn.Parameter(
|
|
torch.zeros(w13_scale_shape, dtype=torch.bfloat16), requires_grad=False
|
|
)
|
|
layer.w2_weight_offset = torch.nn.Parameter(
|
|
torch.zeros(w2_scale_shape, dtype=torch.bfloat16), requires_grad=False
|
|
)
|
|
|
|
layer.w13_weight_shape = torch.nn.Parameter(
|
|
torch.tensor([[2 * self.input_size, self.output_size]] * self.experts, dtype=torch.int32),
|
|
requires_grad=False,
|
|
)
|
|
layer.w2_weight_shape = torch.nn.Parameter(
|
|
torch.tensor([[self.output_size, self.input_size]] * self.experts, dtype=torch.int32), requires_grad=False
|
|
)
|
|
|
|
return layer
|
|
|
|
@patch("vllm_ascend.quantization.methods.w4a16.torch_npu.npu_convert_weight_to_int4pack")
|
|
def test_process_weights_after_loading_with_transpose(self, mock_npu_convert_weight_to_int4pack):
|
|
def mock_convert_weight(weight):
|
|
new_shape = list(weight.shape)
|
|
new_shape[-1] = new_shape[-1] // 8
|
|
return torch.zeros(new_shape, dtype=torch.int32)
|
|
|
|
mock_npu_convert_weight_to_int4pack.side_effect = mock_convert_weight
|
|
layer = self.build_layer()
|
|
self.quant_method.process_weights_after_loading(layer)
|
|
|
|
self.assertEqual(layer.w13_weight_packed.data.shape, torch.Size([8, 128, 8]))
|
|
self.assertEqual(layer.w2_weight_packed.data.shape, torch.Size([8, 32, 16]))
|
|
self.assertEqual(layer.w13_weight_scale.data.shape, torch.Size([8, 4, 64]))
|
|
self.assertEqual(layer.w2_weight_offset.data.shape, torch.Size([8, 1, 128]))
|
|
self.assertTrue(layer.w13_weight_scale.data.is_contiguous())
|
|
|
|
@patch("vllm_ascend.quantization.methods.w4a16._EXTRA_CTX")
|
|
@patch("vllm_ascend.quantization.methods.w4a16.select_experts")
|
|
def test_apply_uses_explicit_dispatch_and_mlp_args(self, mock_select_experts, mock_extra_ctx):
|
|
tokens = 3
|
|
hidden_size = self.output_size
|
|
layer = self.build_layer()
|
|
x = torch.randn(tokens, hidden_size, dtype=torch.float32)
|
|
router_logits = torch.randn(tokens, self.experts, dtype=torch.float32)
|
|
topk_weights = torch.randn(tokens, 2, dtype=torch.float32)
|
|
topk_ids = torch.randint(0, self.experts, (tokens, 2), dtype=torch.int64)
|
|
mc2_mask = torch.tensor([1, 0, 1], dtype=torch.bool)
|
|
pertoken_scale = torch.randn(tokens, dtype=torch.float32)
|
|
layer.swiglu_limit = 1000000
|
|
|
|
mock_select_experts.return_value = (topk_weights, topk_ids)
|
|
mock_comm = Mock()
|
|
mock_comm.fused_experts.return_value = torch.randn(tokens, hidden_size, dtype=torch.float32)
|
|
mock_extra_ctx.moe_comm_method = mock_comm
|
|
mock_extra_ctx.moe_comm_type = MoECommType.ALLGATHER
|
|
|
|
self.quant_method.apply(
|
|
layer=layer,
|
|
x=x,
|
|
router_logits=router_logits,
|
|
top_k=2,
|
|
renormalize=True,
|
|
num_experts=self.experts,
|
|
activation="gelu",
|
|
apply_router_weight_on_input=True,
|
|
mc2_mask=mc2_mask,
|
|
pertoken_scale=pertoken_scale,
|
|
)
|
|
|
|
mock_select_experts.assert_called_once()
|
|
fused_experts_input = mock_comm.fused_experts.call_args.kwargs["fused_experts_input"]
|
|
self.assertEqual(fused_experts_input.activation, "gelu")
|
|
self.assertTrue(fused_experts_input.routing.apply_router_weight_on_input)
|
|
self.assertIs(fused_experts_input.routing.mc2_mask, mc2_mask)
|
|
self.assertIs(fused_experts_input.routing.pertoken_scale, pertoken_scale)
|
|
|
|
@patch("vllm_ascend.quantization.methods.w4a16._EXTRA_CTX")
|
|
@patch("vllm_ascend.quantization.methods.w4a16.select_experts")
|
|
def test_apply_router_logits_mismatch_raises(self, mock_select, mock_ctx):
|
|
layer = self.build_layer()
|
|
x = torch.randn(4, self.output_size, dtype=torch.float32)
|
|
router_logits = torch.randn(4, self.experts + 1, dtype=torch.float32)
|
|
message = (
|
|
"Number of global experts mismatch (excluding redundancy): router_logits.shape[1]=9, num_logical_experts=8"
|
|
)
|
|
|
|
with self.assertRaisesRegex(AssertionError, re.escape(message)):
|
|
self.quant_method.apply(layer, x, router_logits, top_k=2, renormalize=True, num_experts=self.experts)
|
|
mock_select.assert_not_called()
|