EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

support qwen25 vl w8a8 quantization (#2778)

Browse Source 
### What this PR does / why we need it?
support qwen25 vl w8a8 quantization
### Does this PR introduce _any_ user-facing change?
N/A
### How was this patch tested?

- vLLM version: v0.10.1.1
- vLLM main:
62f66be1f7

---------

Signed-off-by: lijiaojiao <lijiaojiao990304@163.com>
Co-authored-by: lijiaojiao <lijiaojiao990304@163.com>
This commit is contained in:
6lazijiamo
2025-09-11 16:40:51 +08:00
committed by GitHub
parent 2b9269b581
commit bd3dedea61
3 changed files with 103 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

40
tests/ut/models/test_qwen2_5_vl.py
View File

@@ -353,6 +353,46 @@ class TestAscendQwen2_5_VisionTransformer(PytestBase):
cos_new, _ = vision_transformer.cal_cos_sin(self.input_data)
assert cos_new.shape == (1, 32, 1, 2)
def test_pad_qkv_bias(self, mocker: MockerFixture):
attention = self.init_vision_transformer(mocker)
mocker.patch("torch.nn.Module.__setattr__")
mocker.patch("torch.nn.Module.__getattr__")
mocker.patch("torch.nn.Module.__delattr__")
res = attention.pad_qkv_bias(torch.rand((300)))
assert res.shape[0] == 384
def test_pad_qkv_weight(self, mocker: MockerFixture):
attention = self.init_vision_transformer(mocker)
mocker.patch("torch.nn.Module.__setattr__")
mocker.patch("torch.nn.Module.__getattr__")
mocker.patch("torch.nn.Module.__delattr__")
res = attention.pad_qkv_weight(torch.rand((300, 300)))
assert res.shape == (384, 300)
def test_pad_proj_weight(self, mocker: MockerFixture):
attention = self.init_vision_transformer(mocker)
mocker.patch("torch.nn.Module.__setattr__")
mocker.patch("torch.nn.Module.__getattr__")
mocker.patch("torch.nn.Module.__delattr__")
res = attention.pad_proj_weight(torch.rand((300, 300)))
assert res.shape == (300, 384)
def test_pad_qkv_weight_scale_offset(self, mocker: MockerFixture):
attention = self.init_vision_transformer(mocker)
mocker.patch("torch.nn.Module.__setattr__")
mocker.patch("torch.nn.Module.__getattr__")
mocker.patch("torch.nn.Module.__delattr__")
res = attention.pad_qkv_weight_scale_offset(torch.rand((300, 1)))
assert res.shape == (384, 1)
def test_pad_qkv_deq_scale_quant_bias(self, mocker: MockerFixture):
attention = self.init_vision_transformer(mocker)
mocker.patch("torch.nn.Module.__setattr__")
mocker.patch("torch.nn.Module.__getattr__")
mocker.patch("torch.nn.Module.__delattr__")
res = attention.pad_qkv_deq_scale_quant_bias(torch.rand((300)))
assert res.shape[0] == 384
def test_forward(self, mocker: MockerFixture):
vision_transformer = self.init_vision_transformer(mocker)
mocker.patch("torch.nn.Module.__setattr__")