Skip to content

Commit 6e8601d

Browse files
authored
Update README.md
1 parent 4c0a909 commit 6e8601d

File tree

1 file changed

+0
-8
lines changed

1 file changed

+0
-8
lines changed

example/auto_compression/nlp/README.md

-8
Original file line numberDiff line numberDiff line change
@@ -30,12 +30,6 @@
3030
| ERNIE 3.0-Medium | Base模型| 75.35 | 57.45 | 60.17 | 81.16 | 77.19 | 80.59 | 79.70 | 73.09 |
3131
| ERNIE 3.0-Medium | 剪枝+量化训练| 74.17 | 56.84 | 59.75 | 80.54 | 76.03 | 76.97 | 80.80 | 72.16 |
3232

33-
| 模型 | 策略 | 报销工单数据 |
34-
|:------:|:------:|:------:|
35-
| UIE-base | Base模型 | [91.83](https://bj.bcebos.com/v1/paddle-slim-models/act/uie_base.tar) |
36-
| UIE-base | 量化训练 | [95.80](https://bj.bcebos.com/v1/paddle-slim-models/act/uie_base_qat_model.tar) |
37-
38-
注:UIE模型精度为在5-shot(每个类别包含5条标注数据)数据集上进行模型微调的结果,压缩后精度更高可能原因是过拟合在当前数据集。
3933

4034
模型在不同任务上平均精度以及加速对比如下:
4135
| 模型 |策略| Accuracy(avg) | 预测时延<sup><small>FP32</small><sup><br><sup> | 预测时延<sup><small>FP16</small><sup><br><sup> | 预测时延<sup><small>INT8</small><sup><br><sup> | 加速比 |
@@ -44,8 +38,6 @@
4438
|PP-MiniLM| 剪枝+离线量化 | 71.85 | - | - | 15.76ms | 5.99x |
4539
|ERNIE 3.0-Medium| Base模型| 73.09 | 89.71ms | 20.76ms | - | - |
4640
|ERNIE 3.0-Medium| 剪枝+量化训练 | 72.16 | - | - | 14.08ms | 6.37x |
47-
|UIE-base| Base模型| 91.83 | 42.66ms | 14.23ms | - | - |
48-
|UIE-base| 量化训练 | 95.80 | - | - | 10.94ms | 3.90x |
4941

5042
性能测试的环境为
5143
- 硬件:NVIDIA Tesla T4 单卡

0 commit comments

Comments
 (0)