apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-05 00:33:03 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

refactor rl get_name_mappings_to_training (#2847)
Some checks failed
Deploy GitHub Pages / deploy (push) Has been cancelled
Details

Browse Source 
* refactor rl get_name_mappings_to_training

* fix tp>1

* change variable name(ffn1->up_gate_proj/ffn2->down_proj)

* change variable name(linear_weight->weight/linear_bias->bias)

* add rl names mapping for vl

* fix ernie 0.3B error

* fix develop code

* fix
This commit is contained in:
Yuanle Liu
2025-07-15 22:31:42 +08:00
committed by GitHub
parent e7bcbbab52
commit 61b3997b85
47 changed files with 1591 additions and 1629 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
fastdeploy/model_executor/layers/quantization/w4afp8.py
View File

@@ -63,8 +63,8 @@ class W4AFP8LinearMethod(QuantMethodBase):
self.quant_config = quant_config
def create_weights(self, layer):
layer.linear_weight_shape.reverse()
layer.linear_weight_shape[0] //= 2
layer.weight_shape.reverse()
layer.weight_shape[0] //= 2
layer.weight_dtype = "int8"
pass
@@ -77,16 +77,16 @@ class W4AFP8LinearMethod(QuantMethodBase):
scale_dtype="float16",
))
weight_scale_tensor = paddle.view(weight_scale_tensor, layer._dtype)
layer.linear_weight.set_value(quanted_weight_tensor)
layer.linear_weight_scale.set_value(weight_scale_tensor)
layer.weight.set_value(quanted_weight_tensor)
layer.weight_scale.set_value(weight_scale_tensor)
def apply(self, layer, x):
linear_out = fastdeploy.model_executor.ops.gpu.scaled_gemm_f8_i4_f16(
x,
layer.linear_weight,
layer.linear_weight_scale,
layer.weight,
layer.weight_scale,
zero_points=None,
bias=layer.linear_bias if layer.add_bias else 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 +