Update README.md
Browse files
README.md
CHANGED
|
@@ -35,6 +35,8 @@ The original fp32 model comes from the fine-tuned model [roberta-base-mrpc](http
|
|
| 35 |
|
| 36 |
The calibration dataloader is the train dataloader. The default calibration sampling size 300 isn't divisible exactly by batch size 8, so the real sampling size is 304.
|
| 37 |
|
|
|
|
|
|
|
| 38 |
### Test result
|
| 39 |
|
| 40 |
- Batch size = 8
|
|
|
|
| 35 |
|
| 36 |
The calibration dataloader is the train dataloader. The default calibration sampling size 300 isn't divisible exactly by batch size 8, so the real sampling size is 304.
|
| 37 |
|
| 38 |
+
Embedding module **roberta.embeddings.token_type_embeddings** is fallbacked to fp32 due to *Unexpected exception RuntimeError('Expect weight, indices, and offsets to be contiguous.')*
|
| 39 |
+
|
| 40 |
### Test result
|
| 41 |
|
| 42 |
- Batch size = 8
|