[taskflow] Fix taskflow bug (PaddlePaddle#9930)

* fix * add doc * update docs * add experiment * update readme.md --------- Co-authored-by: DrownFish19 <[email protected]>
DrownFish19 · Feb 27, 2025 · 07ca81c · 07ca81c
1 parent ebcdc4b
commit 07ca81c
Show file tree

Hide file tree

Showing 5 changed files with 64 additions and 54 deletions.
diff --git a/docs/llm/application/information_extraction/README.md b/docs/llm/application/information_extraction/README.md
@@ -0,0 +1 @@
+../../../../llm/application/information_extraction/README.md
diff --git a/docs/llm/application/information_extraction/doccano.md b/docs/llm/application/information_extraction/doccano.md
@@ -0,0 +1 @@
+../../../../llm/application/information_extraction/doccano.md
diff --git a/llm/application/information_extraction/README.md b/llm/application/information_extraction/README.md
diff --git a/llm/application/information_extraction/doccano.md b/llm/application/information_extraction/doccano.md
@@ -28,7 +28,7 @@ PP-UIE 支持抽取类型的任务，根据实际需要创建一个新的项目
 
 #### 2.1 抽取式任务项目创建
 
-创建项目时选择**序列标注**任务，并勾选**Allow overlapping entity**及**Use relation Labeling**。适配**命名实体识别、关系抽取、事件抽取、评价观点抽取**等任务。
+创建项目时选择**序列标注**任务，并勾选**Allow overlapping entity**及**Use relation Labeling**。适配**命名实体识别、关系抽取、事件抽取**等任务。
 
 <div align="center">
     <img src=https://user-images.githubusercontent.com/40840292/167249142-44885510-51dc-4359-8054-9c89c9633700.png height=230 hspace='15'/>
@@ -236,14 +236,14 @@ schema = {
 python doccano.py \
     --doccano_file ./data/doccano_ext.json \
     --save_dir ./data \
-    --negative_ratio 5
+    --negative_ratio 1
 ```
 
 可配置参数说明：
 
 - ``doccano_file``: 从 doccano 导出的数据标注文件。
 - ``save_dir``: 训练数据的保存目录，默认存储在``data``目录下。
-- ``negative_ratio``: 最大负例比例，该参数只对抽取类型任务有效，适当构造负例可提升模型效果。负例数量和实际的标签数量有关，最大负例数量 = negative_ratio * 正例数量。该参数只对训练集有效，默认为5。为了保证评估指标的准确性，验证集和测试集默认构造全正例。
+- ``negative_ratio``: 最大负例比例，该参数只对抽取类型任务有效，适当构造负例可提升模型效果。负例数量和实际的标签数量有关，最大负例数量 = negative_ratio * 正例数量。
 - ``splits``: 划分数据集时训练集、验证集所占的比例。默认为[0.8, 0.1, 0.1]表示按照``8:1:1``的比例将数据划分为训练集、验证集和测试集。
 - ``task_type``: 选择任务类型，目前只有信息抽取这一种任务。
 - ``is_shuffle``: 是否对数据集进行随机打散，默认为 True。

diff --git a/paddlenlp/taskflow/task.py b/paddlenlp/taskflow/task.py
@@ -20,7 +20,6 @@
 from multiprocessing import cpu_count
 
 import paddle
-from paddle.base.framework import use_pir_api
 from paddle.dataset.common import md5file
 
 from ..utils.env import (
@@ -371,7 +370,6 @@ def _get_inference_model(self):
                     self._construct_input_spec()
                     self._convert_dygraph_to_static()
 
-
         self._static_model_file = self.inference_model_path + PADDLE_INFERENCE_MODEL_SUFFIX
         self._static_params_file = self.inference_model_path + PADDLE_INFERENCE_WEIGHTS_SUFFIX
 
@@ -398,10 +396,7 @@ def _get_inference_model(self):
             self._static_model_file = self._static_fp16_model_file
             self._static_params_file = self._static_fp16_params_file
         if self._predictor_type == "paddle-inference":
-            if use_pir_api():
-                self._config = paddle.inference.Config(self._static_json_file, self._static_params_file)
-            else:
-                self._config = paddle.inference.Config(self._static_model_file, self._static_params_file)
+            self._config = paddle.inference.Config(self._static_model_file, self._static_params_file)
             self._prepare_static_mode()
         else:
             self._prepare_onnx_mode()
Original file line number	Diff line number	Diff line change
		@@ -0,0 +1 @@
		../../../../llm/application/information_extraction/README.md