【Hackathon 9th No.71】add test_weight_only.py #4109
Open
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Thanks for your contribution! |
Sunny-bot1
reviewed
Sep 23, 2025
Comment on lines
+90
to
+110
| def test_apply_numerical_precision(self): | ||
| """Test numerical precision of quantized output""" | ||
| x = paddle.to_tensor(np.random.randn(2, self.in_features).astype("float16")) | ||
|
|
||
| # Reference FP32 output | ||
| ref_out = paddle.matmul( | ||
| x.astype("float32"), | ||
| (self.layer.weight.astype("float32") * self.layer.weight_scale.astype("float32")).transpose([1, 0]), | ||
| ) | ||
| if self.layer.bias is not None: | ||
| ref_out += self.layer.bias.astype("float32") | ||
|
|
||
| # Manual quantized output | ||
| weight_f32 = self.layer.weight.astype("float32") | ||
| x_f32 = x.astype("float32") | ||
| quant_out = paddle.matmul(x_f32, weight_f32 * self.layer.weight_scale.astype("float32"), transpose_y=True) | ||
| if self.layer.bias is not None: | ||
| quant_out += self.layer.bias.astype("float32") | ||
| quant_out = quant_out.astype("float16") | ||
|
|
||
| np.testing.assert_allclose(ref_out.numpy(), quant_out.numpy(), rtol=1e-2, atol=1e-2) |
There was a problem hiding this comment.
这里好像并没有检查self.method.apply的结果
Comment on lines
+20
to
+21
| from fastdeploy.model_executor.layers.quantization.weight_only import ( | ||
| GPUWeightOnlyLinearMethod, |
There was a problem hiding this comment.
weight_only.py 除了 GPUWeightOnlyLinearMethod，还有 MacheteWeightOnlyLinearMethod，使用的kernel实现方式不同，如果是H卡的环境也可以加一下MacheteWeightOnlyLinearMethod的case
| self.fd_config.load_config = type("load_config", (), {"load_choices": "default_v1"})() | ||
|
|
||
| # float32 weights | ||
| weight_fp32 = np.random.randn(*self.weight_shape).astype("float32") |
Comment on lines
+44
to
+49
| # Per-channel scale | ||
| max_abs = np.max(np.abs(weight_fp32), axis=1, keepdims=True) + 1e-6 | ||
| scale = (max_abs / 127.0).astype("float32") | ||
|
|
||
| # Int8 quantization | ||
| weight_int8 = np.clip(np.round(weight_fp32 / scale), -128, 127).astype("int8") |
There was a problem hiding this comment.
这里建议调用GPUWeightOnlyLinearMethod里的process_loaded_weights进行权重的量化
| return paddle.create_parameter(shape=shape, dtype=dtype, default_initializer=default_initializer) | ||
|
|
||
|
|
||
| class TestGPUWeightOnlyLinearMethod(unittest.TestCase): |
There was a problem hiding this comment.
除了对apply方法的检查，也希望增加对process_prequanted_weights、process_loaded_weights方法的检查
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
add test_weight_only.py