Update README.md
Browse files
README.md
CHANGED
@@ -34,16 +34,14 @@ This is an INT8 PyTorch model quantized with [Intel® Neural Compressor](https:
|
|
34 |
|
35 |
The original fp32 model comes from the fine-tuned model [roberta-base-mrpc](https://huggingface.co/Intel/roberta-base-mrpc).
|
36 |
|
37 |
-
The calibration dataloader is the train dataloader. The default calibration sampling size
|
38 |
-
|
39 |
-
The embedding module **roberta.embeddings.token_type_embeddings** falls back to fp32 due to *RuntimeError('Expect weight, indices, and offsets to be contiguous.')*
|
40 |
|
41 |
### Test result
|
42 |
|
43 |
| |INT8|FP32|
|
44 |
|---|:---:|:---:|
|
45 |
-
| **Accuracy (eval-f1)** |0.
|
46 |
-
| **Model size (MB)** |
|
47 |
|
48 |
### Load with Intel® Neural Compressor:
|
49 |
|
|
|
34 |
|
35 |
The original fp32 model comes from the fine-tuned model [roberta-base-mrpc](https://huggingface.co/Intel/roberta-base-mrpc).
|
36 |
|
37 |
+
The calibration dataloader is the train dataloader. The default calibration sampling size 100 isn't divisible exactly by batch size 8, so the real sampling size is 104.
|
|
|
|
|
38 |
|
39 |
### Test result
|
40 |
|
41 |
| |INT8|FP32|
|
42 |
|---|:---:|:---:|
|
43 |
+
| **Accuracy (eval-f1)** |0.9177|0.9138|
|
44 |
+
| **Model size (MB)** |127|499|
|
45 |
|
46 |
### Load with Intel® Neural Compressor:
|
47 |
|