Skip to content

GitLab

Commit c36f0de4 authored by zhougaofeng's avatar zhougaofeng
Browse files

Update Dockerfile, eval_gsm8k.py, eval_math.py, LICENSE, read.md, README.MD,... 

Update Dockerfile, eval_gsm8k.py, eval_math.py, LICENSE, read.md, README.MD, requirements.txt, run_mistral.sh, run.sh, util.py, train_math.py, imgs/dcu.png, imgs/metamath.svg, data/README.md, data/test/MATH_test.jsonl, data/test/GSM8K_test.jsonl, data/test/GSM8K_Backward.jsonl, data/train/README.md, code_for_generating_data/ReadMe.md, code_for_generating_data/code/main_backward_reasoning.py, code_for_generating_data/code/main_forward_reasoning.py, code_for_generating_data/code/main_create_backward_questions.py, code_for_generating_data/code/main_rephrase_question.py, code_for_generating_data/code/path_init.py, code_for_generating_data/code/main_self_verification.py, code_for_generating_data/code/run_create_backward_questions.sh, code_for_generating_data/code/run_forward.sh, code_for_generating_data/code/run_backward.sh, code_for_generating_data/code/run_sv.sh, code_for_generating_data/code/run_rephrase.sh, code_for_generating_data/code/utils/__init__.py, code_for_generating_data/code/utils/answer_clean_utils.py, code_for_generating_data/code/utils/config_utils.py, code_for_generating_data/code/utils/log_utils.py, code_for_generating_data/code/utils/math_utils.py, code_for_generating_data/code/utils/openai_api_utils.py, code_for_generating_data/code/utils/parallel_utils.py, code_for_generating_data/code/utils/time_utils.py, code_for_generating_data/code/utils/path_utils.py, code_for_generating_data/configs/ansaug_cot_math.txt, code_for_generating_data/configs/ansaug_cot_gsm8k.txt, code_for_generating_data/configs/fobar_cot_gsm8k.txt, code_for_generating_data/configs/fobar_cot_math.txt, code_for_generating_data/configs/rephrase_cot_gsm8k.txt, code_for_generating_data/configs/rephrase_cot_math.txt, code_for_generating_data/configs/sv_cot_gsm8k.txt, code_for_generating_data/configs/sv_cot_math.txt, code_for_generating_data/configs/sv_rewrite_question_prompt_gsm8k.txt, code_for_generating_data/configs/sv_rewrite_question_prompt_math.txt, code_for_generating_data/data/gsm8k_train.json, code_for_generating_data/data/MATH_train.json files
parents
Pipeline #1598 canceled with stages
Hide whitespace changes
Inline Side-by-side
Showing with 1631 additions and 0 deletions
Dockerfile 0 → 100644
View file @ c36f0de4
FROM docker pull image.sourcefind.cn:5000/dcu/admin/base/pytorch:2.1.0-centos7.6-dtk24.04-py310
COPY requirement.txt requirement.txt
RUN source /opt/dtk-24.04/env.sh
ENV LANG C.UTF-8
RUN pip install -r requirement.txt -i http://mirrors.aliyun.com/pypi/simple/ --trusted-host mirrors.aliyun.com
LICENSE 0 → 100644
View file @ c36f0de4
Apache License
Version 2.0, January 2004
http://www.apache.org/licenses/
TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
1. Definitions.
"License" shall mean the terms and conditions for use, reproduction,
and distribution as defined by Sections 1 through 9 of this document.
"Licensor" shall mean the copyright owner or entity authorized by
the copyright owner that is granting the License.
"Legal Entity" shall mean the union of the acting entity and all
other entities that control, are controlled by, or are under common
control with that entity. For the purposes of this definition,
"control" means (i) the power, direct or indirect, to cause the
direction or management of such entity, whether by contract or
otherwise, or (ii) ownership of fifty percent (50%) or more of the
outstanding shares, or (iii) beneficial ownership of such entity.
"You" (or "Your") shall mean an individual or Legal Entity
exercising permissions granted by this License.
"Source" form shall mean the preferred form for making modifications,
including but not limited to software source code, documentation
source, and configuration files.
"Object" form shall mean any form resulting from mechanical
transformation or translation of a Source form, including but
not limited to compiled object code, generated documentation,
and conversions to other media types.
"Work" shall mean the work of authorship, whether in Source or
Object form, made available under the License, as indicated by a
copyright notice that is included in or attached to the work
(an example is provided in the Appendix below).
"Derivative Works" shall mean any work, whether in Source or Object
form, that is based on (or derived from) the Work and for which the
editorial revisions, annotations, elaborations, or other modifications
represent, as a whole, an original work of authorship. For the purposes
of this License, Derivative Works shall not include works that remain
separable from, or merely link (or bind by name) to the interfaces of,
the Work and Derivative Works thereof.
"Contribution" shall mean any work of authorship, including
the original version of the Work and any modifications or additions
to that Work or Derivative Works thereof, that is intentionally
submitted to Licensor for inclusion in the Work by the copyright owner
or by an individual or Legal Entity authorized to submit on behalf of
the copyright owner. For the purposes of this definition, "submitted"
means any form of electronic, verbal, or written communication sent
to the Licensor or its representatives, including but not limited to
communication on electronic mailing lists, source code control systems,
and issue tracking systems that are managed by, or on behalf of, the
Licensor for the purpose of discussing and improving the Work, but
excluding communication that is conspicuously marked or otherwise
designated in writing by the copyright owner as "Not a Contribution."
"Contributor" shall mean Licensor and any individual or Legal Entity
on behalf of whom a Contribution has been received by Licensor and
subsequently incorporated within the Work.
2. Grant of Copyright License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
copyright license to reproduce, prepare Derivative Works of,
publicly display, publicly perform, sublicense, and distribute the
Work and such Derivative Works in Source or Object form.
3. Grant of Patent License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
(except as stated in this section) patent license to make, have made,
use, offer to sell, sell, import, and otherwise transfer the Work,
where such license applies only to those patent claims licensable
by such Contributor that are necessarily infringed by their
Contribution(s) alone or by combination of their Contribution(s)
with the Work to which such Contribution(s) was submitted. If You
institute patent litigation against any entity (including a
cross-claim or counterclaim in a lawsuit) alleging that the Work
or a Contribution incorporated within the Work constitutes direct
or contributory patent infringement, then any patent licenses
granted to You under this License for that Work shall terminate
as of the date such litigation is filed.
4. Redistribution. You may reproduce and distribute copies of the
Work or Derivative Works thereof in any medium, with or without
modifications, and in Source or Object form, provided that You
meet the following conditions:
(a) You must give any other recipients of the Work or
Derivative Works a copy of this License; and
(b) You must cause any modified files to carry prominent notices
stating that You changed the files; and
(c) You must retain, in the Source form of any Derivative Works
that You distribute, all copyright, patent, trademark, and
attribution notices from the Source form of the Work,
excluding those notices that do not pertain to any part of
the Derivative Works; and
(d) If the Work includes a "NOTICE" text file as part of its
distribution, then any Derivative Works that You distribute must
include a readable copy of the attribution notices contained
within such NOTICE file, excluding those notices that do not
pertain to any part of the Derivative Works, in at least one
of the following places: within a NOTICE text file distributed
as part of the Derivative Works; within the Source form or
documentation, if provided along with the Derivative Works; or,
within a display generated by the Derivative Works, if and
wherever such third-party notices normally appear. The contents
of the NOTICE file are for informational purposes only and
do not modify the License. You may add Your own attribution
notices within Derivative Works that You distribute, alongside
or as an addendum to the NOTICE text from the Work, provided
that such additional attribution notices cannot be construed
as modifying the License.
You may add Your own copyright statement to Your modifications and
may provide additional or different license terms and conditions
for use, reproduction, or distribution of Your modifications, or
for any such Derivative Works as a whole, provided Your use,
reproduction, and distribution of the Work otherwise complies with
the conditions stated in this License.
5. Submission of Contributions. Unless You explicitly state otherwise,
any Contribution intentionally submitted for inclusion in the Work
by You to the Licensor shall be under the terms and conditions of
this License, without any additional terms or conditions.
Notwithstanding the above, nothing herein shall supersede or modify
the terms of any separate license agreement you may have executed
with Licensor regarding such Contributions.
6. Trademarks. This License does not grant permission to use the trade
names, trademarks, service marks, or product names of the Licensor,
except as required for reasonable and customary use in describing the
origin of the Work and reproducing the content of the NOTICE file.
7. Disclaimer of Warranty. Unless required by applicable law or
agreed to in writing, Licensor provides the Work (and each
Contributor provides its Contributions) on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
implied, including, without limitation, any warranties or conditions
of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
PARTICULAR PURPOSE. You are solely responsible for determining the
appropriateness of using or redistributing the Work and assume any
risks associated with Your exercise of permissions under this License.
8. Limitation of Liability. In no event and under no legal theory,
whether in tort (including negligence), contract, or otherwise,
unless required by applicable law (such as deliberate and grossly
negligent acts) or agreed to in writing, shall any Contributor be
liable to You for damages, including any direct, indirect, special,
incidental, or consequential damages of any character arising as a
result of this License or out of the use or inability to use the
Work (including but not limited to damages for loss of goodwill,
work stoppage, computer failure or malfunction, or any and all
other commercial damages or losses), even if such Contributor
has been advised of the possibility of such damages.
9. Accepting Warranty or Additional Liability. While redistributing
the Work or Derivative Works thereof, You may choose to offer,
and charge a fee for, acceptance of support, warranty, indemnity,
or other liability obligations and/or rights consistent with this
License. However, in accepting such obligations, You may act only
on Your own behalf and on Your sole responsibility, not on behalf
of any other Contributor, and only if You agree to indemnify,
defend, and hold each Contributor harmless for any liability
incurred by, or claims asserted against, such Contributor by reason
of your accepting any such warranty or additional liability.
END OF TERMS AND CONDITIONS
APPENDIX: How to apply the Apache License to your work.
To apply the Apache License to your work, attach the following
boilerplate notice, with the fields enclosed by brackets "[]"
replaced with your own identifying information. (Don't include
the brackets!) The text should be enclosed in the appropriate
comment syntax for the file format. We also recommend that a
file or class name and description of purpose be included on the
same "printed page" as the copyright notice for easier
identification within third-party archives.
Copyright [yyyy] [name of copyright owner]
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
README.MD 0 → 100644
View file @ c36f0de4
# MetaMath: Bootstrap Your Own Mathematical Questions for Large Language Models
[![Code License](https://img.shields.io/badge/Code%20License-Apache_2.0-green.svg)](CODE_LICENSE)
[![Model Weight License](https://img.shields.io/badge/Model%20Weights%20License-LLaMA2-yellow)](MetaMath/LICENSE)
[![Python 3.9+](https://img.shields.io/badge/python-3.9+-blue.svg)](https://www.python.org/downloads/release/python-390/)
<p align="center">
🤗 <a href="https://huggingface.co/meta-math" target="_blank">HF Repo</a> • 📃 <a href="https://arxiv.org/abs/2309.12284" target="_blank">[MetaMath]</a><br>
</p>
<p align="center" width="100%">
<a ><img src="./imgs/metamath.svg" alt="MetaMath" style="width: 80%; min-width: 300px; display: block; margin: auto;"></a>
</p>
## News
- 🔥 Our **MetaMath-Llemma-7B** model achieves **30.0 pass@1** on the MATH Benchmarks, surpassing all the SOTA open-source LLM in 7B-13B scales! All the training scripts and the model are opened.
- 🔥 Our **MetaMath-Mistral-7B** model achieves **77.7 pass@1** on the [GSM8k Benchmarks](https://github.com/openai/grade-school-math), surpassing all the SOTA open-source LLM! All the training scripts and the model are opened.
- 🔥 The full **MetaMathQA** dataset is now released in the huggingface [MetaMathQA](https://huggingface.co/datasets/meta-math/MetaMathQA/tree/main)!
- 🔥 We released the GSM8K_Backward dataset is also released in the huggingface [GSM8K_Backward](https://huggingface.co/datasets/meta-math/GSM8K_Backward) to evaluate the reversal mathematical reasoning ability!
- 🔥 Although the data augmentation for **MetaMathQA** is sourced from **ChatGPT 3.5**, Our **MetaMath-70B** model outperforms the closed-source LLMs **ChatGPT 3.5** on the GSM8K!
- 🔥 Our **MetaMath-7B** model achieves **66.5 pass@1** on the [GSM8k Benchmarks](https://github.com/openai/grade-school-math), **11.6** points higher than the SOTA open-source LLM!
- 🔥 Our **MetaMath-7B** model achieves **19.8 pass@1** on the [MATH Benchmarks](https://github.com/hendrycks/math), **9.1** points higher than the SOTA open-source LLM!
| Model | Checkpoint | Paper | GSM8k | MATH | License|
| ----- |------| ---- |------|-------| ----- |
| MetaMath-70B-V1.0 | 🤗 <a href="https://huggingface.co/meta-math/MetaMath-70B-V1.0" target="_blank">HF Link</a> | 📃 <a href="https://arxiv.org/abs/2309.12284" target="_blank">[MetaMath]</a>| **82.3** | **26.6** | <a href="https://ai.meta.com/resources/models-and-libraries/llama-downloads/" target="_blank">Llama 2 </a> |
| MetaMath-13B-V1.0 | 🤗 <a href="https://huggingface.co/meta-math/MetaMath-13B-V1.0" target="_blank">HF Link</a> | 📃 <a href="https://arxiv.org/abs/2309.12284" target="_blank">[MetaMath]</a>| **72.3** | **22.4** | <a href="https://ai.meta.com/resources/models-and-libraries/llama-downloads/" target="_blank">Llama 2 </a> |
| MetaMath-7B-V1.0 | 🤗 <a href="https://huggingface.co/meta-math/MetaMath-7B-V1.0" target="_blank">HF Link</a> | 📃 <a href="https://arxiv.org/abs/2309.12284" target="_blank">[MetaMath]</a>| **66.5** | **19.8** | <a href="https://ai.meta.com/resources/models-and-libraries/llama-downloads/" target="_blank">Llama 2 </a>|
| MetaMath-Mistral-7B | 🤗 <a href="https://huggingface.co/meta-math/MetaMath-Mistral-7B" target="_blank">HF Link</a> | 📃 <a href="https://arxiv.org/abs/2309.12284" target="_blank">[MetaMath]</a>| **77.7** | **28.2** | <a href="http://www.apache.org/licenses/" target="_blank">Apache License 2.0 </a>|
| MetaMath-Llemma-7B | 🤗 <a href="https://huggingface.co/meta-math/MetaMath-Llemma-7B" target="_blank">HF Link</a> | 📃 <a href="https://arxiv.org/abs/2309.12284" target="_blank">[MetaMath]</a>| **69.2** | **30.0** | <a href="http://www.apache.org/licenses/" target="_blank">Apache License 2.0 </a>|
## Comparing MetaMath with the LLM models.
🔥 Comprehensive Results
| Model | GSM8k Pass@1 | MATH Pass@1 |
|---------------------|--------------|-------------|
| MPT-7B | 6.8 | 3.0 |
| Falcon-7B | 6.8 | 2.3 |
| LLaMA-1-7B | 11.0 | 2.9 |
| LLaMA-2-7B | 14.6 | 2.5 |
| MPT-30B | 15.2 | 3.1 |
| LLaMA-1-13B | 17.8 | 3.9 |
| GPT-Neo-2.7B | 19.5 | -- |
| Falcon-40B | 19.6 | 2.5 |
| Baichuan-chat-13B | 23.9 | -- |
| Vicuna-v1.3-13B | 27.6 | -- |
| LLaMA-2-13B | 28.7 | 3.9 |
| InternLM-7B | 31.2 | -- |
| ChatGLM-2-6B | 32.4 | -- |
| GPT-J-6B | 34.9 | -- |
| LLaMA-1-33B | 35.6 | 3.9 |
| LLaMA-2-34B | 42.2 | 6.24 |
| RFT-7B | 50.3 | -- |
| LLaMA-1-65B | 50.9 | 10.6 |
| Qwen-7B | 51.6 | -- |
| WizardMath-7B | 54.9 | 10.7 |
| LLaMA-2-70B | 56.8 | 13.5 |
| WizardMath-13B | 63.9 | 14.0 |
| 🔥 MetaMath-7B | **66.5** | **19.8** |
| 🔥 MetaMath-13B | **72.3** | **22.4** |
| 🔥 MetaMath-Mistral-7B | **77.7** | **28.2** |
| 🔥 MetaMath-Llemma-7B | **69.2** | **30.0** |
| WizardMath-70B | 81.6 | 22.7 |
| 🔥 MetaMath-70B | **82.3** | **26.6** |
<h2 id="env">Quick Start</h2>
Clone Metamath and install the required packages:
```bash
git clone https://github.com/meta-math/MetaMath.git
cd MetaMath
pip install -r requirements.txt
```
If you encounter a Ray installation problem, please run:
```bash
pip install --upgrade ray
pip install --upgrade pyarrow
pip install pandas
```
<h2 id="Inference">Dataset Usage</h2>
Run the following command to load the data:
```python
from datasets import load_dataset
dataset = load_dataset("meta-math/MetaMathQA")
```
<h2 id="train">Training</h2>
you need to prepare the llama-2 base model and our **MetaMathQA** dataset huggingface [MetaMathQA](https://huggingface.co/datasets/meta-math/MetaMathQA/tree/main)
```
bash run.sh
```
or
```
CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 python3 -m torch.distributed.launch --master_addr ${MASTER_ADDR} --master_port ${MASTER_PORT} --nproc_per_node=8 --use_env train_math.py \
--model_name_or_path "meta-llama/Llama-2-7b-hf" \
--data_path "path/to/metamathqa" \
--data_length 10000000 \
--bf16 True \
--output_dir "path/to/save" \
--num_train_epochs 3 \
--per_device_train_batch_size 4 \
--per_device_eval_batch_size 4 \
--gradient_accumulation_steps 4 \
--evaluation_strategy "no" \
--save_strategy "steps" \
--save_steps 1000 \
--save_total_limit 2 \
--learning_rate 2e-5 \
--weight_decay 0. \
--warmup_ratio 0.03 \
--lr_scheduler_type "cosine" \
--logging_steps 1 \
--fsdp "full_shard auto_wrap" \
--fsdp_transformer_layer_cls_to_wrap 'LlamaDecoderLayer' \
--tf32 True
```
### Supervised fine-tuning
We supervised fine-tune MetaMath-7B with the following hyperparameters:
| Hyperparameter | LLaMA 2 7B |
|----------------|-------------|
| Batch size | 128 |
| Learning rate | 2e-5 |
| Epochs | 3 |
| Max length | 512 |
| LR scheduler | cosine |
<h2 id="evaluation">Evaluation</h2>
we use the vllm to help the fast generation:
```
python eval_gsm8k.py --model "path/to/save" --data_file ./data/test/GSM8K_test.jsonl
python eval_math.py --model "path/to/save" --data_file ./data/test/MATH_test.jsonl
```
where the "path/to/save" should be replaced by the finetuned model, you can also download our series of MetaMath models in huggingface:
🤗 <a href="https://huggingface.co/meta-math/MetaMath-7B-V1.0" target="_blank">MetaMath 7B</a> 🤗 <a href="https://huggingface.co/meta-math/MetaMath-13B-V1.0" target="_blank">MetaMath 13B</a> 🤗 <a href="https://huggingface.co/meta-math/MetaMath-70B-V1.0" target="_blank">MetaMath 70B</a>
The inference prompt for our MetaMath is:
```
"Below is an instruction that describes a task. Write a response that appropriately completes the request.\n\n### Instruction:\n{instruction}\n\n### Response: Let's think step by step."
```
Thanks for the open source code of [WizardMath](https://github.com/nlpxucan/WizardLM/tree/main/WizardMath) and [RFT](https://github.com/OFA-Sys/gsm8k-ScRel/tree/main). Some of our codes are based on them.
<h2 id="citation">Citation</h2>
Please cite the paper if you refer to our model, code, data or paper from MetaMath.
```
@article{yu2023metamath,
title={MetaMath: Bootstrap Your Own Mathematical Questions for Large Language Models},
author={Yu, Longhui and Jiang, Weisen and Shi, Han and Yu, Jincheng and Liu, Zhengying and Zhang, Yu and Kwok, James T and Li, Zhenguo and Weller, Adrian and Liu, Weiyang},
journal={arXiv preprint arXiv:2309.12284},
year={2023}
}
```
code_for_generating_data/ReadMe.md 0 → 100644
View file @ c36f0de4
# Code for generating data in MetaMath
> Tips
- set `export APIKEY="YOUR_API_KEY"` in `~/.bash_profile`
- argument `--num_repeat`: #outputs from ChatGPT for each input by tempearture sampling
- use argument `--part` to void overwriting previous generated data
- use argument `--cont` to continue a previous generating procedure, otherwise, will re-fetch data
### 0. Create backward questions
```bash
cd code
bash -x run_create_backward_questions.sh
```
### 1. AnsAug
```bash
cd code
bash -x run_forward.sh
```
### 2. Rephrasing
```bash
cd code
bash -x run_rephrase.sh
```
### 3. Self-Verification
```bash
cd code
bash -x run_sv.sh
```
### 4. FOBAR
```bash
cd code
bash -x run_backward.sh
```
\ No newline at end of file
code_for_generating_data/code/main_backward_reasoning.py 0 → 100644
View file @ c36f0de4
import path_init
from tqdm import tqdm
import argparse
import json
import os
import copy
from utils.log_utils import LogUtils
from utils.math_utils import MATH_DS_LIST
from utils.parallel_utils import batch_get_api_merge
from utils.path_utils import PathUtils
import numpy as np
from utils.answer_clean_utils import answer_cleansing
ds_path_dict = {
"GSM8K": "GSM8K/gsm8k_train-cleaned",
"MATH": "MATH/MATH_train-cleaned",
"GSM8K_SV": "GSM8K/gsm8k_train-cleaned_SV",
"MATH_SV": "MATH/MATH_train-cleaned_SV",
}
string_number_dict = {"one": 1, "two": 2, "three": 3, "four": 4, "five": 5, "six": 6,
"seven": 7, "eight": 8, "nine": 9, "ten": 10, "eleven": 11, "twelve": 12, "fifth": 5, "sixteen": 16, "half": "50"}
class BackwardReasoning():
def __init__(self, args):
self.args = args
self.ds_name = args.ds
self.temperature = args.temp
self.method = args.method_name
self.eng = args.eng
self.num_repeat = args.num_repeat
self.part = f"_{args.part}" if len(args.part) > 0 else ""
self.logger = LogUtils.get_or_init_logger(f"backward_cot_{self.args.method_name}_{self.ds_name}_{self.get_eng()}{self.part}", "backward")
self.inv_q_dict = {}
self.inv_question_path = os.path.join(PathUtils.DATA_HOME_PATH, f"{ds_path_dict[self.ds_name]}-backward-questions.json""")
with open(self.inv_question_path) as f:
inv_qs = json.load(f)
self.logger.info(f"number of backward question: {len(inv_qs)}")
for e in inv_qs:
if "inverse_question" in e:
if e["question"] not in self.inv_q_dict:
self.inv_q_dict[e["question"]] = []
self.inv_q_dict[e["question"]].append((e["inverse_question"], e['inverse_question_answer']))
self.save_file = os.path.join(PathUtils.DATA_HOME_PATH,
f"{ds_path_dict[self.ds_name]}_{self.args.method_name}_{self.get_eng()}-backward-answers{self.part}.json")
if not args.cont:
new_examples = []
self.todo_path = os.path.join(PathUtils.DATA_HOME_PATH, f"{ds_path_dict['GSM8K']}.json") if 'GSM8K' in self.ds_name else os.path.join(PathUtils.DATA_HOME_PATH, f"{ds_path_dict['MATH']}.json")
with open(self.todo_path) as f:
self.examples = json.load(f)
for e in self.examples:
candidate_answer = e['answer']
if e["question"] in self.inv_q_dict:
for temp_inv_e in self.inv_q_dict[e["question"]]:
new_e = copy.deepcopy(e)
new_e["candidate_answer"] = candidate_answer
new_e["inv_question"] = temp_inv_e[0]
if temp_inv_e[1] in string_number_dict:
new_e["inv_question_ans"] = str(string_number_dict[temp_inv_e[1]])
else:
new_e['inv_question_ans'] = temp_inv_e[1]
new_e['inv_question_ans'] = answer_cleansing(new_e['inv_question_ans'], ds_name=self.ds_name)
new_examples.append(new_e)
self.examples = np.repeat(new_examples, args.num_repeat).tolist()
self.save_data()
with open(self.save_file) as f:
self.examples = json.load(f)
self.unknown_var = "x"
if "MATH" in self.ds_name:
self.unknown_var = "X"
if self.method == "SV":
self.prompt = self.get_prompt("sv_cot_math.txt")
elif self.method == "fobar":
self.prompt = self.get_prompt("fobar_cot_math.txt")
else:
raise ValueError(f"unknown dataset: {self.method}")
def get_eng(self):
if "gpt-4" in self.eng:
return "gpt-4"
elif "gpt-3.5-turbo" in self.eng:
return "gpt-3.5-turbo"
else:
return self.eng
def save_data(self):
with open(self.save_file, 'w', encoding='utf-8') as f:
json.dump(self.examples, f, ensure_ascii=False, indent=4)
def get_prompt(self, prompt_file_name):
prompt_file = os.path.join(PathUtils.CONFIG_HOME_PATH, prompt_file_name)
with open(prompt_file, "r", encoding='utf-8') as f:
prompt = f.read().strip()
return prompt
def evaluate(self, end_idx):
num_correct = 0
for e in self.examples[0:end_idx]:
pred_ans = e['pred_inv_answer_cleaned']
gt_ans = answer_cleansing(e['inv_question_ans'], ds_name=self.ds_name)
if pred_ans == gt_ans:
num_correct += 1
return num_correct, end_idx, num_correct / end_idx
def get_inv_split_str(self):
if self.ds_name in MATH_DS_LIST:
return f"The value of ${self.unknown_var}$ is"
else:
return f"The value of {self.unknown_var} is"
def fetch_data_from_openai(self):
def wrap(e):
variable, special_token = (f"{self.unknown_var}", "") if "GSM8K" in self.ds_name else (f"${self.unknown_var}$", "### ")
if self.method == "fobar":
wrap_q = f"""{e['inv_question']}\n{special_token}If we know the answer to the above question is {e['candidate_answer']}, what is the value of unknown variable {variable}?"""
elif self.method == "SV":
wrap_q = f"""{e['inv_question']} What is the value of unknown variable {variable}?"""
else:
raise ValueError(f"unknown method: {self.method}")
return f"""{self.prompt}\n\nQuestion: {wrap_q}\nA: Let's think step by step.\n"""
def extract(e, reply):
e['inv_question_pred_answer'] = reply
e['pred_inv_answer_cleaned'] = answer_cleansing(pred=reply, ds_name=self.ds_name, split_str=self.get_inv_split_str())
todo_list = []
for i, example in tqdm(enumerate(self.examples), total=len(self.examples)):
if i % 10 == 0:
self.logger.info(f"processing: {i}/{len(self.examples)}")
if "inv_question_pred_answer" in example or 'inv_question' not in example:
self.logger.info(f"skip {i}th question, has no inv question.")
continue
todo_list.append(example)
if (len(todo_list) >= args.batch_size) or i >= (len(self.examples) - 1):
batch_get_api_merge(examples=todo_list, eng=self.args.eng, pre_fun=wrap, post_fun=extract,
logger=self.logger, n_processes=self.args.num_proc,
temperature=self.temperature, timeout=self.args.time_out, max_try=8)
self.save_data()
todo_list = []
num_correct, num_examples, acc = self.evaluate(i + 1)
self.logger.info(
"=" * 20 + f"processed: {i}/{len(self.examples)}, acc: {num_correct}/{num_examples}={100 * acc:.2f}")
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument('--eng', default="gpt-3.5-turbo", type=str)
parser.add_argument('--ds', default="GSM8K", type=str)
parser.add_argument('--temp', default=0.7, type=float)
parser.add_argument('--part', type=str)
parser.add_argument('--cont', action='store_true', help="true=continue previous fetching, default=false")
parser.add_argument('--method_name', default="fobar", type=str)
parser.add_argument('--num_repeat', default=20, type=int)
parser.add_argument('--batch_size', default=20, type=int)
parser.add_argument('--time_out', default=30, type=int)
parser.add_argument('--num_proc', default=16, type=int)
args = parser.parse_args()
rephrase_cot = BackwardReasoning(args)
rephrase_cot.fetch_data_from_openai()
\ No newline at end of file
code_for_generating_data/code/main_create_backward_questions.py 0 → 100644
View file @ c36f0de4
import path_init
import re
import json
import os
import copy
import argparse
from utils.answer_clean_utils import delete_extra_zero, string_number_dict, _strip_string
from utils.path_utils import PathUtils
from abc import ABC, abstractmethod
class InverseQuestions(ABC):
def __init__(self, args):
self.output_clean_path = f"{self.input_path}-cleaned.json"
self.output_path = f"{self.input_path}-cleaned-backward-questions.json"
self.load_dataset()
self.parse_examples()
print(f"#samples: {len(self.input_examples)}")
self.save_cleaned_data()
self.unknown_var = "x"
def load_dataset(self):
with open(os.path.join(PathUtils.DATA_HOME_PATH, f"{self.input_path}.json"), 'r') as f:
self.input_examples = json.load(f)
@abstractmethod
def parse_examples(self): pass
def replace_number_x(self, s):
if s in string_number_dict:
s = str(string_number_dict[s])
if s[-1] in (",", ".", "?", ";", "”", "'", "!", "\"", "%"):
try:
mo = re.match('.*([0-9])[^0-9]*$', s)
return self.unknown_var + s[mo.end(1):]
except:
print(f"the string is {s}")
elif s[0] in ("$"):
return "$" + self.unknown_var
else:
return self.unknown_var
@staticmethod
def search_number(s):
if s in string_number_dict:
return True
if re.search('[\d]', s) is not None:
if re.search('[a-zA-Z]', s) or re.search('[\\n:\(\)-*\"+–-]', s):
return None
else:
return True
def save_cleaned_data(self):
with open(os.path.join(PathUtils.DATA_HOME_PATH, self.output_clean_path), 'w',
encoding='utf-8') as f:
json.dump(self.input_examples, f, ensure_ascii=False, indent=4)
def save_data(self):
print(f"#samples for backward reasoning: {len(self.output_examples)}")
with open(os.path.join(PathUtils.DATA_HOME_PATH, self.output_path), 'w',
encoding='utf-8') as f:
json.dump(self.output_examples, f, ensure_ascii=False, indent=4)
def make_inv_question(self):
self.output_examples = []
num_example_has_backward_question = 0
for e in self.input_examples:
token_list = e['question'].split(' ')
numbers_idx = [idx for idx, _ in enumerate(token_list) if self.search_number(_) is not None]
if len(numbers_idx) > 0:
num_example_has_backward_question += 1
for x_idx in numbers_idx:
_e = copy.deepcopy(e)
_token_list = copy.deepcopy(token_list)
inverse_question_answer = _token_list[x_idx]
_token_list[x_idx] = self.replace_number_x(_token_list[x_idx])
_e['inverse_question'] = " ".join(_token_list)
_e['inverse_question_answer'] = inverse_question_answer
self.output_examples.append(_e)
print(f"has_inv_q: {num_example_has_backward_question}/{len(self.input_examples)}")
self.save_data()
class GSM8K(InverseQuestions):
def __init__(self, args):
self.ds_name = "GSM8K"
self.input_path = f"GSM8K/gsm8k_train"
super(GSM8K, self).__init__(args=args)
def parse_examples(self):
temp_examples = []
for e in self.input_examples:
q = e['question']
a = e['answer']
if a[-2:] == ".0":
a = a[:-2]
a = delete_extra_zero(a)
ans_detail = e['answer_detail']
temp_examples.append(dict(question=q, answer=a, answer_detail=ans_detail))
self.input_examples = temp_examples
class _MATH(InverseQuestions):
def __init__(self, args):
super(_MATH, self).__init__(args=args)
self.unknown_var = "X"
@staticmethod
def search_number(s):
if re.search('[\d]', s) is not None:
if re.search('[a-zA-Z]', s) or re.search('[\\n:\(\)-*\"+–-]', s):
return None
else:
return True
def find_math_answer(self, s):
assert ('boxed' in s)
# s = s.replace(",", "")
ans = s.split('boxed')[-1]
if (ans[0] == '{'):
stack = 1
a = ''
for c in ans[1:]:
if (c == '{'):
stack += 1
a += c
elif (c == '}'):
stack -= 1
if (stack == 0): break
a += c
else:
a += c
else:
a = ans.split('$')[0].strip()
a = _strip_string(a)
return a
def parse_examples(self):
temp_examples = []
for e in self.input_examples:
q = e['problem']
ans_detail = e['solution']
level = e['level']
type = e['type']
question_id = e['question_id']
a = self.find_math_answer(ans_detail)
temp_examples.append(dict(question=q, answer=a, answer_detail=ans_detail,
level=level, type=type, question_id=question_id))
self.input_examples = temp_examples
def replace_number_x(self, s):
if s[-1] in (",", ".", "?", ";", "”", "'", "!", "\"", "%"):
try:
mo = re.match('.*([0-9])[^0-9]*$', s)
return self.unknown_var + s[mo.end(1):]
except:
print(f"the string is {s}")
else:
return self.unknown_var
def make_inv_question(self):
self.output_examples = []
num_example_has_backward_question = 0
for e in self.input_examples:
token_list = e['question'].split(' ')
numbers_idx = [idx for idx, _ in enumerate(token_list) if self.search_number(_) is not None]
if len(numbers_idx) > 0:
num_example_has_backward_question += 1
for x_idx in numbers_idx:
_e = copy.deepcopy(e)
_token_list = copy.deepcopy(token_list)
inverse_question_answer = _token_list[x_idx]
_token_list[x_idx] = self.replace_number_x(_token_list[x_idx])
_e['inverse_question'] = " ".join(_token_list)
_e['inverse_question_answer'] = inverse_question_answer
self.output_examples.append(_e)
print(f"has_inv_q: {num_example_has_backward_question}/{len(self.input_examples)}")
self.save_data()
class MATH(_MATH):
def __init__(self, args):
self.ds_name = "MATH"
self.input_path = f"MATH/MATH_train"
super(MATH, self).__init__(args=args)
if __name__ == "__main__":
parser = argparse.ArgumentParser()
args = parser.parse_args()
for method in [GSM8K, MATH]:
method(args).make_inv_question()
code_for_generating_data/code/main_forward_reasoning.py 0 → 100644
View file @ c36f0de4
import path_init
from tqdm import tqdm
from statistics import mode
import argparse
from utils.parallel_utils import batch_get_api_merge
from utils.answer_clean_utils import answer_cleansing
from collections import Counter
from utils.log_utils import LogUtils
from utils.path_utils import PathUtils
import os
import json
import numpy as np
ds_path_dict = {
"GSM8K": "GSM8K/gsm8k_train-cleaned",
"MATH": "MATH/MATH_train-cleaned",
"GSM8K_rephrased": "GSM8K/gsm8k_train-cleaned_rephrased_questions",
"MATH_rephrased": "MATH/MATH_train-cleaned_rephrased_questions",
}
class ForwardReasoning():
def __init__(self, args):
self.args = args
self.ds_name = args.ds
self.temperature = args.temp
self.part = f"_{args.part}" if len(args.part) > 0 else ""
self.eng = args.eng
self.num_repeat = args.num_repeat
self.method_name = self.get_method_name()
self.logger = LogUtils.get_or_init_logger(f"{self.method_name}_{self.ds_name}_{self.get_eng()}{self.part}",
"forward")
self.json_file = os.path.join(PathUtils.DATA_HOME_PATH, f"{ds_path_dict[self.ds_name]}.json")
self.save_file = os.path.join(PathUtils.DATA_HOME_PATH,
f"{ds_path_dict[self.ds_name]}_{self.method_name}_answer_{self.get_eng()}_{self.part}.json")
self.save_stat_file = os.path.join(PathUtils.DATA_HOME_PATH,
f"{ds_path_dict[self.ds_name]}_{self.method_name}_answer_{self.get_eng()}_{self.part}_stat.json")
if not args.cont:
with open(self.json_file) as f:
self.examples = json.load(f)
self.examples = np.repeat(self.examples, self.num_repeat).tolist()
self.save_data()
with open(self.save_file) as f:
self.examples = json.load(f)
if "GSM8K" in self.ds_name:
self.prompt = self.get_prompt("ansaug_cot_gsm8k.txt")
elif "MATH" in self.ds_name:
self.prompt = self.get_prompt("ansaug_cot_math.txt")
else:
raise ValueError(f"unknown dataset={self.ds_name}")
def save_data(self):
with open(self.save_file, 'w', encoding='utf-8') as f:
json.dump(self.examples, f, ensure_ascii=False, indent=4)
def get_method_name(self):
return "SCComplexCoT"
def get_eng(self):
if "gpt-4" in self.eng:
return "gpt-4"
elif "gpt-3.5-turbo" in self.eng:
return "gpt-3.5-turbo"
else:
return self.eng
def get_prompt(self, prompt_file_name):
prompt_file = os.path.join(PathUtils.CONFIG_HOME_PATH, prompt_file_name)
with open(prompt_file, "r", encoding='utf-8') as f:
prompt = f.read().strip()
return prompt
def save_ans_stat(self):
examples_collect = {}
for e in self.examples[0:len(self.examples)]:
question = e['question']
if question not in examples_collect:
examples_collect[question] = {}
for k in ["answer", "question", "answer_detail"]:
examples_collect[question][k] = e[k]
examples_collect[question]['pred_answer_cleaned_list'] = []
examples_collect[question]['pred_answer_cleaned_list'].append(e['pred_answer_cleaned'])
stat_list = []
for e in examples_collect.values():
counter = Counter(e['pred_answer_cleaned_list'])
e["ans_stat"] = dict(counter)
del e['pred_answer_cleaned_list']
stat_list.append(e)
with open(self.save_stat_file, 'w', encoding='utf-8') as f:
json.dump(stat_list, f, ensure_ascii=False, indent=4)
def evaluate(self, end_idx):
result_stat_dict = {}
for e in self.examples[0:end_idx]:
question = e['question']
if question not in result_stat_dict:
result_stat_dict[question] = []
result_stat_dict[question].append(e)
num_correct = 0
for q in result_stat_dict:
e_list = result_stat_dict[q]
answer = e_list[0]['answer']
pred_answers = [_['pred_answer_cleaned'] for _ in e_list]
freq_answer = mode(pred_answers)
if freq_answer == answer:
num_correct += 1
msg = f"acc: {100 * num_correct / len(result_stat_dict.keys()):.4f}"
self.logger.info(msg)
return num_correct, len(result_stat_dict.keys()), num_correct / len(result_stat_dict.keys())
def fetch_data_from_openai(self):
def wrap(e):
return "{}\n\nQuestion: {}\nA: Let's think step by step.\n".format(self.prompt, e['question'])
def extract(e, reply):
e['pred_answer'] = reply
e['pred_answer_cleaned'] = answer_cleansing(pred=reply, ds_name=self.ds_name)
todo_list = []
for i, example in tqdm(enumerate(self.examples), total=len(self.examples)):
if i % 10 == 0:
self.logger.info(f"processing: {i}/{len(self.examples)}")
if "pred_answer" in example and len(example['pred_answer']) > 10: # contain answer
continue
todo_list.append(example)
if len(todo_list) >= self.args.batch_size or i >= (len(self.examples) - 1):
if len(todo_list) > 0:
batch_get_api_merge(examples=todo_list, eng=self.args.eng, pre_fun=wrap, post_fun=extract,
logger=self.logger, n_processes=self.args.num_proc,
temperature=self.temperature, timeout=self.args.time_out, max_try=0)
todo_list = []
self.save_data()
num_correct, num_examples, acc = self.evaluate(i + 1)
self.logger.info(
"=" * 20 + f"processed: {i}/{len(self.examples)}, acc: {num_correct}/{num_examples}={100 * acc:.2f}")
self.save_ans_stat()
class SCComplexCoT(ForwardReasoning):
def __init__(self, args):
super(SCComplexCoT, self).__init__(args=args)
def get_method_name(self):
return "SCComplexCoT"
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument('--eng', default="gpt-3.5-turbo", type=str)
parser.add_argument('--ds', default="GSM8K", type=str)
parser.add_argument('--part', default="", type=str)
parser.add_argument('--temp', default=0.7, help="temperature", type=float)
parser.add_argument('--method_name', default="SCComplexCoT", type=str)
parser.add_argument('--cont', action='store_true', help="true=continue previous fetching, default=false")
parser.add_argument('--num_repeat', default=10, type=int, help="for self-consistency")
parser.add_argument('--batch_size', default=20, type=int)
parser.add_argument('--time_out', default=30, type=int)
parser.add_argument('--num_proc', default=16, type=int)
args = parser.parse_args()
method = SCComplexCoT(args)
method.fetch_data_from_openai()
method.logger.info("final evaluation")
num_correct, num_question, acc = method.evaluate(len(method.examples))
msg = f"finished acc: {100 * num_correct / num_question:.4f}"
code_for_generating_data/code/main_rephrase_question.py 0 → 100644
View file @ c36f0de4
from tqdm import tqdm
import path_init
import argparse
import json
import os
from utils.log_utils import LogUtils
from utils.parallel_utils import batch_get_api_merge
from utils.path_utils import PathUtils
import numpy as np
ds_path_dict = {
"GSM8K": "GSM8K/gsm8k_train-cleaned",
"MATH": "MATH/MATH_train-cleaned",
}
class RephraseQuestion():
def __init__(self, args):
self.args = args
self.ds_name = args.ds
self.temperature = args.temp
self.eng = args.eng
self.method_name = self.get_method_name()
self.num_repeat = args.num_repeat
self.logger = LogUtils.get_or_init_logger(f"rephrase_question_{self.ds_name}_{self.get_eng()}_rephrased_questions", "rephrase")
self.json_file = os.path.join(PathUtils.DATA_HOME_PATH, f"{ds_path_dict[self.ds_name]}.json")
self.save_file = os.path.join(PathUtils.DATA_HOME_PATH,
f"{ds_path_dict[self.ds_name]}_rephrased_questions.json")
if not args.cont:
with open(self.json_file) as f:
self.examples = json.load(f)
self.examples = np.repeat(self.examples, self.num_repeat).tolist()
self.save_data()
with open(self.save_file) as f:
self.examples = json.load(f)
if "GSM8K" in self.ds_name:
self.prompt = self.get_prompt(f"rephrase_cot_gsm8k.txt")
elif "MATH" in self.ds_name:
self.prompt = self.get_prompt(f"rephrase_cot_math.txt")
else:
raise ValueError(f"unknown dataset={self.ds_name}")
def get_method_name(self):
return "SCComplexCoT"
def get_prompt(self, prompt_file_name):
prompt_file = os.path.join(PathUtils.CONFIG_HOME_PATH, prompt_file_name)
with open(prompt_file, "r", encoding='utf-8') as f:
prompt = f.read().strip()
return prompt
def get_eng(self):
if "gpt-4" in self.eng:
return "gpt-4"
elif "gpt-3.5-turbo" in self.eng:
return "gpt-3.5-turbo"
else:
return self.eng
def save_data(self):
with open(self.save_file, 'w', encoding='utf-8') as f:
json.dump(self.examples, f, ensure_ascii=False, indent=4)
def fetch_data_from_openai(self):
def wrap(e):
return f"""{self.prompt}\n\nQuestion: {e['question']}\nRephrase the above question: """
def extract(e, reply):
original_question = e['question']
e['question'] = reply
e['original_question'] = original_question
todo_list = []
for i, example in tqdm(enumerate(self.examples), total=len(self.examples)):
if i % 10 == 0:
self.logger.info(f"processing: {i}/{len(self.examples)}")
if "original_question" in example:
continue
todo_list.append(example)
if len(todo_list) >= args.batch_size or i >= (len(self.examples) - 1):
if len(todo_list) > 0:
batch_get_api_merge(examples=todo_list, eng=self.args.eng, pre_fun=wrap, post_fun=extract,
logger=self.logger, n_processes=self.args.num_proc,
temperature=self.temperature, timeout=self.args.time_out, max_try=8)
todo_list = []
self.save_data()
self.logger.info("=" * 40 + f"processed: {i}/{len(self.examples)}")
self.logger.info("Finished.")
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument('--eng', default="gpt-3.5-turbo", type=str)
parser.add_argument('--ds', default="GSM8K", type=str)
parser.add_argument('--temp', default=0.7, type=float)
parser.add_argument('--cont', action='store_true', help="true=continue previous fetching, default=false")
parser.add_argument('--num_repeat', default=40, type=int)
parser.add_argument('--batch_size', default=20, type=int)
parser.add_argument('--time_out', default=30, type=int)
parser.add_argument('--num_proc', default=20, type=int)
args = parser.parse_args()
rephrase_cot = RephraseQuestion(args)
rephrase_cot.fetch_data_from_openai()
\ No newline at end of file
code_for_generating_data/code/main_self_verification.py 0 → 100644
View file @ c36f0de4
import path_init
import argparse
import json
import os
from utils.log_utils import LogUtils
from utils.math_utils import MATH_DS_LIST
from utils.parallel_utils import batch_get_api_merge
from utils.path_utils import PathUtils
import numpy as np
from utils.answer_clean_utils import answer_cleansing
ds_path_dict = {
"GSM8K": "GSM8K/gsm8k_train-cleaned",
"MATH": "MATH/MATH_train-cleaned",
}
class SelfVerification():
def __init__(self, args):
self.args = args
self.ds_name = args.ds
self.temperature = args.temp
self.eng = args.eng
self.method_name = self.get_method_name()
self.num_repeat = args.num_repeat
self.logger = LogUtils.get_or_init_logger(f"SV_{self.ds_name}_{self.get_eng()}_rewritten_questions", "for_tuning")
self.json_file = os.path.join(PathUtils.DATA_HOME_PATH, f"{ds_path_dict[self.ds_name]}-backward-questions.json")
self.save_file = os.path.join(PathUtils.DATA_HOME_PATH,
f"{ds_path_dict[self.ds_name]}_SV-backward-questions.json")
if not args.cont:
with open(self.json_file) as f:
self.examples = json.load(f)
self.examples = np.repeat(self.examples, self.num_repeat).tolist()
self.save_data()
with open(self.save_file) as f:
self.examples = json.load(f)
if "GSM8K" in self.ds_name:
self.prompt = self.get_prompt("sv_rewrite_question_prompt_gsm8k.txt")
else:
self.prompt = self.get_prompt("sv_rewrite_question_prompt_math.txt")
def get_method_name(self):
return "SCComplexCoT"
def get_eng(self):
if "gpt-4" in self.eng:
return "gpt-4"
elif "gpt-3.5-turbo" in self.eng:
return "gpt-3.5-turbo"
else:
return self.eng
def get_prompt(self, prompt_file_name):
prompt_file = os.path.join(PathUtils.CONFIG_HOME_PATH, prompt_file_name)
with open(prompt_file, "r", encoding='utf-8') as f:
prompt = f.read().strip()
return prompt
def save_data(self):
with open(self.save_file, 'w', encoding='utf-8') as f:
json.dump(self.examples, f, ensure_ascii=False, indent=4)
def fetch_data_from_openai(self):
def wrap(e):
text = e['inverse_question'].replace(',', '.')
position_fullstop = text[::-1].find('.')
answer = answer_cleansing(e['answer'], ds_name=self.ds_name)
question = text[len(text) - position_fullstop:].strip()
e['base_text'] = e['inverse_question'][:len(text) - position_fullstop].strip()
return f"{self.prompt}\n\nQuestion: {question} The answer is {answer}.\n Result: "
def extract(e, reply):
e['inverse_question'] = f"{e['base_text']} {reply}"
todo_list = []
for i, example in enumerate(self.examples):
if i % 10 == 0:
self.logger.info(f"processing: {i}/{len(self.examples)}")
if "rephrased_question" in example:
continue
todo_list.append(example)
if len(todo_list) >= args.batch_size or i >= (len(self.examples) - 1):
if len(todo_list) > 0:
batch_get_api_merge(examples=todo_list, eng=self.args.eng, pre_fun=wrap, post_fun=extract,
logger=self.logger, n_processes=self.args.num_proc,
temperature=self.temperature, timeout=self.args.time_out, max_try=8)
todo_list = []
self.save_data()
self.logger.info("=" * 40 + f"processed: {i}/{len(self.examples)}")
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument('--eng', default="gpt-3.5-turbo", type=str)
parser.add_argument('--ds', default="GSM8K", type=str)
parser.add_argument('--temp', default=0.7, type=float)
parser.add_argument('--cont', action='store_true', help="true=continue previous fetching, default=false")
parser.add_argument('--num_repeat', default=40, type=int)
parser.add_argument('--batch_size', default=20, type=int)
parser.add_argument('--time_out', default=10, type=int)
parser.add_argument('--num_proc', default=16, type=int)
args = parser.parse_args()
rephrase_cot = SelfVerification(args)
rephrase_cot.fetch_data_from_openai()
\ No newline at end of file
code_for_generating_data/code/path_init.py 0 → 100644
View file @ c36f0de4
from pathlib import Path
from utils.path_utils import PathUtils
"""
set the current working path
"""
work_path = Path().resolve().parent
PathUtils.HOME_PATH = work_path
PathUtils.set_path()
print(work_path)
def null_fun():
pass
\ No newline at end of file
code_for_generating_data/code/run_backward.sh 0 → 100644
View file @ c36f0de4
source ~/.bash_profile
model_name=gpt-3.5-turbo
ds_name="GSM8K"
#ds_name="MATH"
python main_backward_reasoning.py --part part1 --method_name fobar --eng $model_name --ds $ds_name --temp 0.7 --num_repeat 2 --batch_size 500 --time_out 30 --num_proc 20
code_for_generating_data/code/run_create_backward_questions.sh 0 → 100644
View file @ c36f0de4
python main_create_backward_questions.py
\ No newline at end of file
code_for_generating_data/code/run_forward.sh 0 → 100644
View file @ c36f0de4
source ~/.bash_profile
model_name=gpt-3.5-turbo
method_name="SCComplexCoT"
ds_name="GSM8K"
#ds_name="MATH"
python main_forward_reasoning.py --part part1 --eng $model_name --ds $ds_name --method $method_name --temp 0.7 --num_repeat 2 --batch_size 500 --time_out 30 --num_proc 20
\ No newline at end of file
code_for_generating_data/code/run_rephrase.sh 0 → 100644
View file @ c36f0de4
source ~/.bash_profile
model_name=gpt-3.5-turbo
ds_name="GSM8K"
#ds_name="MATH"
python main_rephrase_question.py --eng $model_name --ds $ds_name --temp 0.7 --num_repeat 2 --batch_size 500 --time_out 30 --num_proc 20
python main_forward_reasoning.py --part part1 --eng $model_name --ds "${ds_name}_rephrased" --method $method_name --temp 0.7 --num_repeat 2 --batch_size 500 --time_out 30 --num_proc 20
code_for_generating_data/code/run_sv.sh 0 → 100644
View file @ c36f0de4
source ~/.bash_profile
model_name=gpt-3.5-turbo
ds_name="GSM8K"
#ds_name="MATH"
python main_self_verification.py --eng $model_name --ds $ds_name --temp 0.7 --num_repeat 2 --batch_size 500 --time_out 30 --num_proc 20
python main_backward_reasoning.py --part part1 --method_name SV --eng $model_name --ds "${ds_name}_SV" --temp 0.7 --num_repeat 2 --batch_size 500 --time_out 30 --num_proc 20
code_for_generating_data/code/utils/answer_clean_utils.py 0 → 100644
View file @ c36f0de4
import re
from utils.math_utils import MATH_DS_LIST
string_number_dict = {"one": 1, "two": 2, "three": 3, "four": 4, "five": 5,
"six": 6, "seven": 7, "eight": 8, "nine": 9, "ten": 10,
"eleven": 11, "twelve": 12, "fifth": 5,
"sixteen": 16, "half": "50%"}
def delete_extra_zero(n):
try:
n=float(n)
except:
# print("None {}".format(n))
return n
if isinstance(n, int):
return str(n)
if isinstance(n, float):
n = str(n).rstrip('0') # 删除小数点后多余的0
n = int(n.rstrip('.')) if n.endswith('.') else float(n) # 只剩小数点直接转int，否则转回float
n=str(n)
return n
def extract_math_answer(pred_str, split_str='The answer is '):
if(split_str in pred_str):
pred = pred_str.split(split_str)[-1].strip()
elif('the answer is ' in pred_str):
pred = pred_str.split('the answer is ')[-1].strip()
elif 'boxed' in pred_str:
ans = pred_str.split('boxed')[-1]
if (ans[0] == '{'):
stack = 1
a = ''
for c in ans[1:]:
if (c == '{'):
stack += 1
a += c
elif (c == '}'):
stack -= 1
if (stack == 0): break
a += c
else:
a += c
else:
a = ans.split('$')[0].strip()
a = _strip_string(a)
pred=a
else:
pattern = '-?\d*\.?\d+'
pred = re.findall(pattern, pred_str)
if(len(pred) >= 1):
# print(pred_str)
pred = pred[-1]
else:
pred = ''
if pred != "" and len(pred) >= 1:
if pred[-1] == ".":
pred = pred[:-1]
if len(pred) >= 1 and pred[-1] == "/":
pred = pred[:-1]
pred = _strip_string(pred)
if 'boxed' in pred:
ans = pred.split('boxed')[-1]
if (ans[0] == '{'):
stack = 1
a = ''
for c in ans[1:]:
if (c == '{'):
stack += 1
a += c
elif (c == '}'):
stack -= 1
if (stack == 0): break
a += c
else:
a += c
else:
a = ans.split('$')[0].strip()
a = _strip_string(a)
pred=a
return pred
def answer_cleansing(pred, ds_name, split_str="The answer is"):
if ds_name in MATH_DS_LIST:
return extract_math_answer(pred, split_str)
preds = pred.split(split_str)
pred = preds[-1]
pred = pred.replace(",", "")
pred = [delete_extra_zero(s.replace(",", "")) for s in re.findall(r'-?\d+/?\.?\d*', pred)]
# If there is no candidate in list, null is set.
if len(pred) == 0:
pred = ""
else:
pred = pred[-1]
# (For arithmetic tasks) if a word ends with period, it will be omitted ...
if pred != "":
if pred[-1] == ".":
pred = pred[:-1]
if pred[-1] == "/":
pred = pred[:-1]
return pred
def _fix_fracs(string):
substrs = string.split("\\frac")
new_str = substrs[0]
if len(substrs) > 1:
substrs = substrs[1:]
for substr in substrs:
new_str += "\\frac"
if substr[0] == "{":
new_str += substr
else:
try:
assert len(substr) >= 2
except:
return string
a = substr[0]
b = substr[1]
if b != "{":
if len(substr) > 2:
post_substr = substr[2:]
new_str += "{" + a + "}{" + b + "}" + post_substr
else:
new_str += "{" + a + "}{" + b + "}"
else:
if len(substr) > 2:
post_substr = substr[2:]
new_str += "{" + a + "}" + b + post_substr
else:
new_str += "{" + a + "}" + b
string = new_str
return string
def _fix_a_slash_b(string):
if len(string.split("/")) != 2:
return string
a = string.split("/")[0]
b = string.split("/")[1]
try:
a = int(a)
b = int(b)
assert string == "{}/{}".format(a, b)
new_string = "\\frac{" + str(a) + "}{" + str(b) + "}"
return new_string
except:
return string
def _remove_right_units(string):
# "\\text{ " only ever occurs (at least in the val set) when describing units
if "\\text{ " in string:
splits = string.split("\\text{ ")
# assert len(splits) == 2
return splits[0]
else:
return string
def _fix_sqrt(string):
if "\\sqrt" not in string:
return string
splits = string.split("\\sqrt")
new_string = splits[0]
for split in splits[1:]:
if split[0] != "{":
a = split[0]
new_substr = "\\sqrt{" + a + "}" + split[1:]
else:
new_substr = "\\sqrt" + split
new_string += new_substr
return new_string
def _strip_string(string):
# linebreaks
string = string.replace("\n", "")
# print(string)
# remove inverse spaces
string = string.replace("\\!", "")
# print(string)
# replace \\ with \
string = string.replace("\\\\", "\\")
# print(string)
# replace tfrac and dfrac with frac
string = string.replace("tfrac", "frac")
string = string.replace("dfrac", "frac")
# print(string)
# remove \left and \right
string = string.replace("\\left", "")
string = string.replace("\\right", "")
# print(string)
# Remove circ (degrees)
string = string.replace("^{\\circ}", "")
string = string.replace("^\\circ", "")
# remove dollar signs
string = string.replace("\\$", "")
# remove units (on the right)
string = _remove_right_units(string)
# remove percentage
string = string.replace("\\%", "")
string = string.replace("\%", "")
# " 0." equivalent to " ." and "{0." equivalent to "{." Alternatively, add "0" if "." is the start of the string
string = string.replace(" .", " 0.")
string = string.replace("{.", "{0.")
# if empty, return empty string
if len(string) == 0:
return string
if string[0] == ".":
string = "0" + string
# to consider: get rid of e.g. "k = " or "q = " at beginning
if len(string.split("=")) == 2:
if len(string.split("=")[0]) <= 2:
string = string.split("=")[1]
# fix sqrt3 --> sqrt{3}
string = _fix_sqrt(string)
# remove spaces
string = string.replace(" ", "")
# \frac1b or \frac12 --> \frac{1}{b} and \frac{1}{2}, etc. Even works with \frac1{72} (but not \frac{72}1). Also does a/b --> \\frac{a}{b}
string = _fix_fracs(string)
# manually change 0.5 --> \frac{1}{2}
if string == "0.5":
string = "\\frac{1}{2}"
# NOTE: X/Y changed to \frac{X}{Y} in dataset, but in simple cases fix in case the model output is X/Y
string = _fix_a_slash_b(string)
return string
\ No newline at end of file
code_for_generating_data/code/utils/config_utils.py 0 → 100644
View file @ c36f0de4
from utils.path_utils import PathUtils
import yaml
import torch
import os
class ConfigUtils(object):
def __init__(self):
pass
@staticmethod
def get_device(device_id=0):
device = torch.device("cpu")
if torch.cuda.is_available():
print("GPU is available, using GPU:{}".format(device_id))
device = torch.device('cuda:{}'.format(device_id))
else:
print("GPU is unavailable, using CPU")
return device
@staticmethod
def get_config_dict(config_file_name):
config_file_full_path = os.path.join(PathUtils.CONFIG_HOME_PATH, config_file_name)
return yaml.load(open(config_file_full_path, "r"), Loader=yaml.Loader)
code_for_generating_data/code/utils/log_utils.py 0 → 100644
View file @ c36f0de4
import logging
import os
from pathlib import Path
from utils.path_utils import PathUtils
class LogUtils:
LOGGER_FORMATTER = logging.Formatter('%(asctime)s %(levelname)s: %(message)s', datefmt='%Y-%m-%d %H:%M:%S')
LOGGER_DICT = {}
def __init__(self):
pass
@staticmethod
def get_or_init_logger(file_name, dir_name=None, level=logging.DEBUG) -> logging.Logger:
"""
log: time-level-prefix-msg
:param file_name:
:param prefix:
:param dir_name:
:param level:
:return:
"""
if dir_name is None:
raise ValueError("job id should not be None!")
log_dir_path = "{}/{}".format(PathUtils.Log_HOME_PATH, dir_name)
log_file = "{}/log_{}.log".format(log_dir_path, file_name)
# return the logger if exists
if log_file in LogUtils.LOGGER_DICT:
return LogUtils.LOGGER_DICT[log_file]
# create a logger if not exist
os.makedirs(os.path.dirname(log_file), exist_ok=True)
handler = logging.FileHandler(log_file, mode='a')
handler.setFormatter(LogUtils.LOGGER_FORMATTER)
logger = logging.getLogger(file_name)
logger.setLevel(level)
logger.addHandler(handler)
LogUtils.LOGGER_DICT[log_file] = logger
os.utime(Path(log_dir_path))
return logger
@staticmethod
def get_stat_from_dict(stat_dict, keys=None, is_simple=True):
if not keys:
keys = stat_dict.keys()
msg = []
for key in keys:
if is_simple:
msg.append(stat_dict[key].simple_repr())
else:
msg.append(str(stat_dict[key]))
return "; ".join(msg)
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment