EngineX-Ascend/enginex-ascend-910-vllm
10
0
Fork 1
Code Issues Pull Requests Actions Projects Releases Wiki Activity

init v0.11.0rc0

Browse Source
This commit is contained in:
luopingyi
2025-10-14 10:38:28 +08:00
parent 67afd0ea78
commit 66dc16f966
278 changed files with 28130 additions and 11708 deletions
Download Patch File Download Diff File Expand all files Collapse all files

45
tests/ut/quantization/test_quant_config.py
View File

@@ -73,9 +73,12 @@ class TestAscendQuantConfig(TestBase):
self.assertIsNone(result)
def test_get_quant_method_for_linear(self):
mock_config = MagicMock()
mock_config.model_config.hf_config.model_type = None
linear_layer = MagicMock(spec=LinearBase)
# Test skipped layer
with patch.object(self.ascend_config,
with patch("vllm_ascend.quantization.quant_config.get_current_vllm_config", return_value=mock_config), \
patch.object(self.ascend_config, \
'is_layer_skipped_ascend',
return_value=True):
method = self.ascend_config.get_quant_method(linear_layer, ".attn")
@@ -83,6 +86,7 @@ class TestAscendQuantConfig(TestBase):
# Test quantized layer
with patch.object(self.ascend_config, 'is_layer_skipped_ascend', return_value=False), \
patch("vllm_ascend.quantization.quant_config.get_current_vllm_config", return_value=mock_config), \
patch('vllm_ascend.quantization.quant_config.AscendLinearMethod', return_value=MagicMock()) as mock_ascend_linear:
method = self.ascend_config.get_quant_method(linear_layer, ".attn")
@@ -93,14 +97,18 @@ class TestAscendQuantConfig(TestBase):
def test_get_quant_method_for_attention(self):
attention_layer = MagicMock(spec=Attention)
with patch('vllm_ascend.quantization.quant_config.AscendKVCacheMethod',
mock_config = MagicMock()
mock_config.model_config.hf_config.model_type = None
with patch("vllm_ascend.quantization.quant_config.get_current_vllm_config", return_value=mock_config), \
patch('vllm_ascend.quantization.quant_config.AscendKVCacheMethod', \
return_value=MagicMock()) as mock_ascend_kvcache:
# Test with fa_quant_type
method = self.ascend_config.get_quant_method(
attention_layer, ".attn")
self.assertIs(method, mock_ascend_kvcache.return_value)
with patch('vllm_ascend.quantization.quant_config.AscendKVCacheMethod',
with patch("vllm_ascend.quantization.quant_config.get_current_vllm_config", return_value=mock_config), \
patch('vllm_ascend.quantization.quant_config.AscendKVCacheMethod', \
return_value=MagicMock()) as mock_ascend_kvcache:
# Test with kv_quant_type
modified_config = {"kv_quant_type": "C8"}
@@ -113,9 +121,12 @@ class TestAscendQuantConfig(TestBase):
fused_moe_layer = MagicMock(spec=FusedMoE)
fused_moe_layer.moe = MagicMock(spec=FusedMoEConfig)
fused_moe_layer.moe_config = MagicMock(spec=FusedMoEConfig)
mock_config = MagicMock()
mock_config.model_config.hf_config.model_type = None
# Test skipped layer
with patch.object(self.ascend_config, 'is_layer_skipped_ascend', return_value=True), \
patch("vllm_ascend.quantization.quant_config.get_current_vllm_config", return_value=mock_config), \
patch('vllm_ascend.quantization.quant_config.AscendUnquantizedFusedMoEMethod', return_value=MagicMock()) as mock_ascend_moe:
method = self.ascend_config.get_quant_method(
fused_moe_layer, "moe_layer")
@@ -123,6 +134,7 @@ class TestAscendQuantConfig(TestBase):
# Test quantized layer
with patch.object(self.ascend_config, 'is_layer_skipped_ascend', return_value=False), \
patch("vllm_ascend.quantization.quant_config.get_current_vllm_config", return_value=mock_config), \
patch('vllm_ascend.quantization.quant_config.AscendFusedMoEMethod', return_value=MagicMock()) as mock_ascend_moe:
method = self.ascend_config.get_quant_method(
fused_moe_layer, "moe_layer")
@@ -156,33 +168,22 @@ class TestAscendKVCacheMethod(TestBase):
def setUp(self):
# Setup common test fixtures
self.mock_quant_config = MagicMock(spec=AscendQuantConfig)
self.mock_quant_config.quant_description = {"some_config": "value"}
self.prefix = "attention_layer"
self.mock_quant_config.quant_description = {"kv_quant_type": "C8"}
self.prefix = "layer.attn"
# Mock the quantizer and quant_method
self.mock_quantizer = MagicMock()
# Mock quant_method
self.mock_quant_method = MagicMock()
# Patch the AscendQuantizer
self.quantizer_patcher = patch(
'vllm_ascend.quantization.quant_config.AscendQuantizer.get_quantizer',
return_value=self.mock_quantizer)
self.mock_get_quantizer = self.quantizer_patcher.start()
self.mock_quantizer.build_attention_method.return_value = self.mock_quant_method
self.patcher = patch(
'vllm_ascend.quantization.quant_config.get_quant_method')
self.mock_get_quant_method = self.patcher.start()
self.mock_get_quant_method.return_value = self.mock_quant_method
# Create instance
self.kv_cache_method = AscendKVCacheMethod(self.mock_quant_config,
self.prefix)
def tearDown(self):
self.quantizer_patcher.stop()
def test_init(self):
"""Test initialization with proper quantizer setup."""
self.mock_get_quantizer.assert_called_once_with(
self.mock_quant_config.quant_description, self.prefix)
self.mock_quantizer.build_attention_method.assert_called_once()
self.patcher.stop()
def test_create_weights(self):
"""Test create_weights delegates to quant_method."""