Commit
•
1b31e6d
1
Parent(s):
8f1f2c6
Update README.md (#3)
Browse files- Update README.md (7c4eb899a6b9a03b48793d82b551de8eb2c64116)
- Update README.md (b1a44217b69169dc854d10f08e7321abbd7d2ede)
Co-authored-by: Adam Stachowicz <[email protected]>
README.md
CHANGED
|
@@ -24,7 +24,7 @@ The only difference is that there are a few new training arguments specific to H
|
|
| 24 |
|
| 25 |
[Here](https://github.com/huggingface/optimum-habana/blob/main/examples/question-answering/run_qa.py) is a question-answering example script to fine-tune a model on SQuAD. You can run it with RoBERTa Large with the following command:
|
| 26 |
```bash
|
| 27 |
-
python run_qa.py \
|
| 28 |
--model_name_or_path roberta-large \
|
| 29 |
--gaudi_config_name Habana/roberta-large \
|
| 30 |
--dataset_name squad \
|
|
@@ -37,7 +37,9 @@ python run_qa.py \
|
|
| 37 |
--max_seq_length 384 \
|
| 38 |
--output_dir /tmp/squad/ \
|
| 39 |
--use_habana \
|
| 40 |
-
--
|
|
|
|
|
|
|
| 41 |
--throughput_warmup_steps 3 \
|
| 42 |
--bf16
|
| 43 |
```
|
|
|
|
| 24 |
|
| 25 |
[Here](https://github.com/huggingface/optimum-habana/blob/main/examples/question-answering/run_qa.py) is a question-answering example script to fine-tune a model on SQuAD. You can run it with RoBERTa Large with the following command:
|
| 26 |
```bash
|
| 27 |
+
PT_HPU_LAZY_MODE=0 python run_qa.py \
|
| 28 |
--model_name_or_path roberta-large \
|
| 29 |
--gaudi_config_name Habana/roberta-large \
|
| 30 |
--dataset_name squad \
|
|
|
|
| 37 |
--max_seq_length 384 \
|
| 38 |
--output_dir /tmp/squad/ \
|
| 39 |
--use_habana \
|
| 40 |
+
--torch_compile_backend hpu_backend \
|
| 41 |
+
--torch_compile \
|
| 42 |
+
--use_lazy_mode false \
|
| 43 |
--throughput_warmup_steps 3 \
|
| 44 |
--bf16
|
| 45 |
```
|