Update README.md
Browse files
README.md
CHANGED
@@ -198,7 +198,7 @@ We fine-tune LLaMA-7B with the following hyperparameters:
|
|
198 |
|
199 |
| Hyperparameter | Global Batch Size | Learning rate | Epochs | Max length | Weight decay |
|
200 |
|:----------------|-------------------:|---------------:|--------:|------------:|--------------:|
|
201 |
-
| LLaMA 7B | 128 | 2e-5 | 3 |
|
202 |
|
203 |
To reproduce the training procedure, please use the following command:
|
204 |
|
@@ -259,7 +259,7 @@ python3 ./eval/generate.py \
|
|
259 |
--prompt_type alpaca \
|
260 |
--do_sample
|
261 |
|
262 |
-
2. Evaluation
|
263 |
python3 ./eval/chatgpt_score.py \
|
264 |
--baseline_file ./eval/answer/<MODEL_1>.jsonl \ # answer of baseline model to compare with
|
265 |
--answer_file ./eval/answer/<MODEL_2>.jsonl \ # answer of evaluation model
|
@@ -285,10 +285,10 @@ python3 ./eval/generate.py \
|
|
285 |
--max_new_tokens 512 \
|
286 |
--prompt_type alpaca
|
287 |
|
288 |
-
2. Evaluation
|
289 |
python3 ./eval/auto_eval.py \
|
290 |
--question_file ./eval/question/MATH_eval_set_sample.jsonl \
|
291 |
-
--answer_file ./eval/answer/<MODEL_ID>.jsonl
|
292 |
```
|
293 |
|
294 |
## Limitations
|
|
|
198 |
|
199 |
| Hyperparameter | Global Batch Size | Learning rate | Epochs | Max length | Weight decay |
|
200 |
|:----------------|-------------------:|---------------:|--------:|------------:|--------------:|
|
201 |
+
| LLaMA 7B | 128 | 2e-5 | 3 | 512| 0 |
|
202 |
|
203 |
To reproduce the training procedure, please use the following command:
|
204 |
|
|
|
259 |
--prompt_type alpaca \
|
260 |
--do_sample
|
261 |
|
262 |
+
# 2. Evaluation
|
263 |
python3 ./eval/chatgpt_score.py \
|
264 |
--baseline_file ./eval/answer/<MODEL_1>.jsonl \ # answer of baseline model to compare with
|
265 |
--answer_file ./eval/answer/<MODEL_2>.jsonl \ # answer of evaluation model
|
|
|
285 |
--max_new_tokens 512 \
|
286 |
--prompt_type alpaca
|
287 |
|
288 |
+
# 2. Evaluation
|
289 |
python3 ./eval/auto_eval.py \
|
290 |
--question_file ./eval/question/MATH_eval_set_sample.jsonl \
|
291 |
+
--answer_file ./eval/answer/<MODEL_ID>.jsonl # answer of evaluation model
|
292 |
```
|
293 |
|
294 |
## Limitations
|