Commit
·
01e2dd7
1
Parent(s):
ec9cbc8
update int8 onnx model and readme
Browse filesSigned-off-by: dujun <[email protected]>
- README.md +5 -3
- model.onnx +2 -2
README.md
CHANGED
|
@@ -12,7 +12,7 @@ metrics:
|
|
| 12 |
- f1
|
| 13 |
---
|
| 14 |
|
| 15 |
-
# INT8 MiniLM finetuned MRPC
|
| 16 |
|
| 17 |
## Post-training static quantization
|
| 18 |
|
|
@@ -47,12 +47,14 @@ This is an INT8 ONNX model quantized with [Intel® Neural Compressor](https://gi
|
|
| 47 |
|
| 48 |
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).
|
| 49 |
|
|
|
|
|
|
|
| 50 |
#### Test result
|
| 51 |
|
| 52 |
| |INT8|FP32|
|
| 53 |
|---|:---:|:---:|
|
| 54 |
-
| **Accuracy (eval-f1)** |0.
|
| 55 |
-
| **Model size (MB)** |
|
| 56 |
|
| 57 |
|
| 58 |
#### Load ONNX model:
|
|
|
|
| 12 |
- f1
|
| 13 |
---
|
| 14 |
|
| 15 |
+
# INT8 MiniLM-L12-H384 finetuned MRPC
|
| 16 |
|
| 17 |
## Post-training static quantization
|
| 18 |
|
|
|
|
| 47 |
|
| 48 |
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).
|
| 49 |
|
| 50 |
+
The calibration dataloader is the eval dataloader. The calibration sampling size is 100.
|
| 51 |
+
|
| 52 |
#### Test result
|
| 53 |
|
| 54 |
| |INT8|FP32|
|
| 55 |
|---|:---:|:---:|
|
| 56 |
+
| **Accuracy (eval-f1)** |0.9013|0.9097|
|
| 57 |
+
| **Model size (MB)** |33|128|
|
| 58 |
|
| 59 |
|
| 60 |
#### Load ONNX model:
|
model.onnx
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:a6e1b600142cfb7374432a1fcdcfe1b2903b04ccca0fe5e41fec47c74639ef80
|
| 3 |
+
size 34012804
|