apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-05 08:37:06 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

polish code with new pre-commit rule (#2923)

Browse Source
This commit is contained in:
Zero Rains
2025-07-19 23:19:27 +08:00
committed by GitHub
parent b8676d71a8
commit 25698d56d1
424 changed files with 14307 additions and 13518 deletions
Download Patch File Download Diff File Expand all files Collapse all files

27
fastdeploy/model_executor/layers/quantization/w4afp8.py
View File

@@ -13,6 +13,7 @@
# See the License for the specific language governing permissions and
# limitations under the License.
"""
from typing import Optional
import paddle
@@ -69,13 +70,14 @@ class W4AFP8LinearMethod(QuantMethodBase):
pass
def process_loaded_weights(self, layer, weights) -> None:
quanted_weight_tensor, weight_scale_tensor = (
fastdeploy.model_executor.ops.gpu.
scaled_gemm_f8_i4_f16_weight_quantize(
paddle.cast(weights, "float32").cpu(),
groupsize=-1,
scale_dtype="float16",
))
(
quanted_weight_tensor,
weight_scale_tensor,
) = fastdeploy.model_executor.ops.gpu.scaled_gemm_f8_i4_f16_weight_quantize(
paddle.cast(weights, "float32").cpu(),
groupsize=-1,
scale_dtype="float16",
)
weight_scale_tensor = paddle.view(weight_scale_tensor, layer._dtype)
layer.weight.set_value(quanted_weight_tensor)
layer.weight_scale.set_value(weight_scale_tensor)
@@ -87,11 +89,12 @@ class W4AFP8LinearMethod(QuantMethodBase):
layer.weight_scale,
zero_points=None,
bias=layer.bias if layer.add_bias else None,
out_scale=self.quant_config.weight_scale_dict.get(layer.prefix +
".weight_scale")
/ (self.quant_config.act_scale_dict.get(layer.prefix +
".activation_scale") *
QUANT_SCALING_FACTOR * QUANT_SCALING_FACTOR),
out_scale=self.quant_config.weight_scale_dict.get(layer.prefix + ".weight_scale")
/ (
self.quant_config.act_scale_dict.get(layer.prefix + ".activation_scale")
* QUANT_SCALING_FACTOR
* QUANT_SCALING_FACTOR
),
groupsize=0,
out_dtype=layer._dtype,
)