xinhe commited on
Commit
37ee185
1 Parent(s): 0ddbe50

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +44 -0
README.md CHANGED
@@ -1,3 +1,47 @@
1
  ---
 
2
  license: mit
 
 
 
 
 
 
 
 
 
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ language: en
3
  license: mit
4
+ tags:
5
+ - text-classfication
6
+ - int8
7
+ - Intel® Neural Compressor
8
+ - QuantizationAwareTraining
9
+ datasets:
10
+ - mrpc
11
+ metrics:
12
+ - f1
13
  ---
14
+
15
+ # INT8 BERT base uncased finetuned MRPC
16
+
17
+ ### QuantizationAwareTraining
18
+
19
+ This is an INT8 PyTorch model quantized with [Intel® Neural Compressor](https://github.com/intel/neural-compressor).
20
+
21
+ The original fp32 model comes from the fine-tuned model [Intel/MiniLM-L12-H384-uncased-mrpc](https://huggingface.co/Intel/MiniLM-L12-H384-uncased-mrpc).
22
+
23
+ ### Test result
24
+
25
+ | |INT8|FP32|
26
+ |---|:---:|:---:|
27
+ | **Accuracy (eval-f1)** |0.9068|0.9097|
28
+ | **Model size (MB)** |33.1|127|
29
+
30
+ ### Load with Intel® Neural Compressor:
31
+
32
+ ```python
33
+ from neural_compressor.utils.load_huggingface import OptimizedModel
34
+ int8_model = OptimizedModel.from_pretrained(
35
+ 'Intel/MiniLM-L12-H384-uncased-mrpc-int8-qat',
36
+ )
37
+ ```
38
+
39
+ ### Training hyperparameters
40
+
41
+ The following hyperparameters were used during training:
42
+ - learning_rate: 1e-05
43
+ - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
44
+ - lr_scheduler_type: linear
45
+ - num_epochs: 1.0
46
+ - train_batch_size: 16
47
+ - eval_batch_size: 8