jimboHsueh
commited on
Commit
·
a4ab28e
1
Parent(s):
40517c3
Upload README-4.md
Browse files- README-4.md +38 -0
README-4.md
ADDED
@@ -0,0 +1,38 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# **HW2**
|
2 |
+
|
3 |
+
## **Environment**
|
4 |
+
```
|
5 |
+
pip install -r requirements.txt
|
6 |
+
```
|
7 |
+
|
8 |
+
## **Download**
|
9 |
+
Download adapter checkpoints
|
10 |
+
```
|
11 |
+
bash ./download.sh
|
12 |
+
```
|
13 |
+
|
14 |
+
## **Fine tuning mT5 model**
|
15 |
+
```
|
16 |
+
python train.py --base_model_path jimboHsueh/save_hw3 \
|
17 |
+
--peft_path ./PETH \
|
18 |
+
--train_data_path train.json \
|
19 |
+
--per_device_train_batch_size 2 \
|
20 |
+
--gradient_accumulation_steps 8 \
|
21 |
+
--test_data_path public_test.json \
|
22 |
+
--output_dir HW3_4 \
|
23 |
+
--push_to_hub
|
24 |
+
```
|
25 |
+
-**base_model_path**: Path to Taiwan_llama.
|
26 |
+
|
27 |
+
-**peft_path**: Path to peft model.
|
28 |
+
|
29 |
+
-**train_data_path**: Path to train.json.
|
30 |
+
|
31 |
+
-**test_data_path**: Path to test.json.
|
32 |
+
|
33 |
+
## **Generate prediction.json**
|
34 |
+
```
|
35 |
+
bash ./download.sh
|
36 |
+
bash ./run.sh python /path/to/Taiwan-LLaMa-folder /path/to/adapter_checkpoint \
|
37 |
+
/path/to/input.json /path/to/output.json
|
38 |
+
```
|