Skip to content

GitLab

Unverified Commit b0ce5a10 authored by Fazzie-Maqianli's avatar Fazzie-Maqianli Committed by GitHub
Browse files

[Coati] first commit (#3283)

parent fd6add57
Hide whitespace changes
Inline Side-by-side
Showing with 1597 additions and 0 deletions
applications/Chat/.gitignore 0 → 100644
View file @ b0ce5a10
# Byte-compiled / optimized / DLL files
__pycache__/
*.py[cod]
*$py.class
# C extensions
*.so
# Distribution / packaging
.Python
build/
develop-eggs/
dist/
downloads/
eggs/
.eggs/
lib/
lib64/
parts/
sdist/
var/
wheels/
pip-wheel-metadata/
share/python-wheels/
*.egg-info/
.installed.cfg
*.egg
MANIFEST
# PyInstaller
# Usually these files are written by a python script from a template
# before PyInstaller builds the exe, so as to inject date/other infos into it.
*.manifest
*.spec
# Installer logs
pip-log.txt
pip-delete-this-directory.txt
# Unit test / coverage reports
htmlcov/
.tox/
.nox/
.coverage
.coverage.*
.cache
nosetests.xml
coverage.xml
*.cover
*.py,cover
.hypothesis/
.pytest_cache/
# Translations
*.mo
*.pot
# Django stuff:
*.log
local_settings.py
db.sqlite3
db.sqlite3-journal
# Flask stuff:
instance/
.webassets-cache
# Scrapy stuff:
.scrapy
# Sphinx documentation
docs/_build/
docs/.build/
# PyBuilder
target/
# Jupyter Notebook
.ipynb_checkpoints
# IPython
profile_default/
ipython_config.py
# pyenv
.python-version
# pipenv
# According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
# However, in case of collaboration, if having platform-specific dependencies or dependencies
# having no cross-platform support, pipenv may install dependencies that don't work, or not
# install all needed dependencies.
#Pipfile.lock
# PEP 582; used by e.g. github.com/David-OConnor/pyflow
__pypackages__/
# Celery stuff
celerybeat-schedule
celerybeat.pid
# SageMath parsed files
*.sage.py
# Environments
.env
.venv
env/
venv/
ENV/
env.bak/
venv.bak/
# Spyder project settings
.spyderproject
.spyproject
# Rope project settings
.ropeproject
# mkdocs documentation
/site
# mypy
.mypy_cache/
.dmypy.json
dmypy.json
# Pyre type checker
.pyre/
# IDE
.idea/
.vscode/
# macos
*.DS_Store
#data/
docs/.build
# pytorch checkpoint
*.pt
# wandb log
example/wandb/
applications/Chat/LICENSE 0 → 100644
View file @ b0ce5a10
Copyright 2021- HPC-AI Technology Inc. All rights reserved.
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 2021- HPC-AI Technology Inc.
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.
applications/Chat/README.md 0 → 100644
View file @ b0ce5a10
<h1 align="center">
<span>Coati - ColossalAI Talking Intelligence</span>
<img width="auto" height="50px", src="assets/logo_coati.png"/>
</h1>
## Table of Contents
- [Table of Contents](#table-of-contents)
- [What is Coati ?](#what-is-coati-)
- [Online demo](#online-demo)
- [Install](#install)
- [Install the environment](#install-the-environment)
- [Install the Transformers](#install-the-transformers)
- [How to use?](#how-to-use)
- [Supervised datasets collection](#supervised-datasets-collection)
- [Stage1 - Supervised instructs tuning](#stage1---supervised-instructs-tuning)
- [Stage2 - Training reward model](#stage2---training-reward-model)
- [Stage3 - Training model with reinforcement learning by human feedback](#stage3---training-model-with-reinforcement-learning-by-human-feedback)
- [Coati7B examples](#coati7b-examples)
- [FAQ](#faq)
- [How to save/load checkpoint](#how-to-saveload-checkpoint)
- [The Plan](#the-plan)
- [Real-time progress](#real-time-progress)
- [Invitation to open-source contribution](#invitation-to-open-source-contribution)
- [Quick Preview](#quick-preview)
- [Authors](#authors)
- [Citations](#citations)
- [Licenses](#licenses)
---
## What is Coati ?
Coati is a large language model developed by Colossal-AI, which is also a unified large language model framework that has implemented the following functions
- Supports comprehensive large-model training acceleration capabilities for ColossalAI, without requiring knowledge of complex distributed training algorithms
- Supervised datasets collection
- Supervised insturcts fine-tuning
- Training reward model
- Reinforcement learning with human feedback
- Quantization inference
- Fast model deploying
- Perfectly integration with the Hugging Face ecosystem, high degree of model customization
More details can be found in the [blog](https://www.hpc-ai.tech/blog/colossal-ai-chatgpt).
<p align="center">
<img src="https://raw.githubusercontent.com/hpcaitech/public_assets/main/applications/chatgpt/chatgpt.png" width=700/>
</p>
## Online demo
You can experience the performance of Coati7B on this page.
[chat.colossalai.org](https://chat.colossalai.org/)
> Warning: Due to model and dataset size limitations, Coati is just a baby model, Coati7B may output incorrect information and lack the ability for multi-turn dialogue. There is still significant room for improvement.
## Install
### Install the environment
```shell
conda creat -n coati
conda activate coati
pip install .
```
### Install the Transformers
Given Hugging Face hasn't officially supported the LLaMA models, We fork a branch of Transformers that can be compatible with our code
```shell
git clone https://github.com/hpcaitech/transformers
cd transformers
pip install .
```
## How to use?
### Supervised datasets collection
we colllected 104K bilingual dataset of Chinese and English, and you can find the datasets in this repo
Here is how we collected the data
<p align="center">
<img src="assets/data-collect.png" width=500/>
</p>
### Stage1 - Supervised instructs tuning
Stage1 is supervised instructs fine-tuning, which uses the datasets mentioned earlier to fine-tune the model
you can run the `examples/train_sft.sh` to start a supervised instructs fine-tuning
```
torchrun --standalone --nproc_per_node=4 train_sft.py \
--pretrain "/path/to/LLaMa-7B/" \
--model 'llama' \
--strategy colossalai_zero2 \
--log_interval 10 \
--save_path /path/to/Coati-7B \
--dataset /path/to/data.json \
--batch_size 4 \
--accimulation_steps 8 \
--lr 2e-5 \
--max_datasets_size 512 \
--max_epochs 1 \
```
### Stage2 - Training reward model
Stage2 trains a reward model, which obtains corresponding scores by manually ranking different outputs for the same prompt and supervises the training of the reward model
you can run the `examples/train_rm.sh` to start a reward model training
```
torchrun --standalone --nproc_per_node=4 train_reward_model.py
--pretrain "/path/to/LLaMa-7B/" \
--model 'llama' \
--strategy colossalai_zero2 \
--loss_fn 'log_exp'\
--save_path 'rmstatic.pt' \
```
### Stage3 - Training model with reinforcement learning by human feedback
Stage3 uses reinforcement learning algorithm, which is the most complex part of the training process:
<p align="center">
<img src="assets/stage-3.jpeg" width=500/>
</p>
you can run the `examples/train_prompts.sh` to start training PPO with human feedback
```
torchrun --standalone --nproc_per_node=4 train_prompts.py prompts.csv \
--pretrain "/path/to/LLaMa-7B/" \
--model 'llama' \
--strategy colossalai_zero2
```
For more details, see `examples/`.
We also support training reward model with true-world data. See `examples/train_reward_model.py`.
## Coati7B examples
## FAQ
### How to save/load checkpoint
We have integrated the Transformers save and load pipeline, allowing users to freely call Hugging Face's language models and save them in the HF format.
```
from coati.models.llama import LlamaLM
from coati.trainer import SFTTrainer
model = LlamaLM(pretrained=args.pretrain)
tokenizer = AutoTokenizer.from_pretrained(args.pretrain)
trainer = SFTTrainer(model=model,
strategy=strategy,
optim=optim,
train_dataloader=train_dataloader,
eval_dataloader=eval_dataloader,
batch_size=args.batch_size,
max_epochs=args.max_epochs,
accimulation_steps = args.accimulation_steps
)
trainer.fit()
trainer.save_model(path=args.save_path, only_rank0=True, tokenizer=tokenizer)
```
## The Plan
- [x] implement PPO fine-tuning
- [x] implement training reward model
- [x] support LoRA
- [x] support inference
- [x] open source the reward model weight
- [x] support llama from [facebook](https://github.com/facebookresearch/llama)
- [x] implement PPO-ptx fine-tuning
- [ ] integrate with Ray
- [ ] support more RL paradigms, like Implicit Language Q-Learning (ILQL),
- [ ] support chain of throught by [langchain](https://github.com/hwchase17/langchain)
### Real-time progress
You will find our progress in github project broad
[Coati](https://github.com/orgs/hpcaitech/projects/17/views/1)
## Invitation to open-source contribution
Referring to the successful attempts of [BLOOM](https://bigscience.huggingface.co/) and [Stable Diffusion](https://en.wikipedia.org/wiki/Stable_Diffusion), any and all developers and partners with computing powers, datasets, models are welcome to join and build the Colossal-AI community, making efforts towards the era of big AI models from the starting point of replicating ChatGPT!
You may contact us or participate in the following ways:
1. [Leaving a Star ⭐](https://github.com/hpcaitech/ColossalAI/stargazers) to show your like and support. Thanks!
2. Posting an [issue](https://github.com/hpcaitech/ColossalAI/issues/new/choose), or submitting a PR on GitHub follow the guideline in [Contributing](https://github.com/hpcaitech/ColossalAI/blob/main/CONTRIBUTING.md).
3. Join the Colossal-AI community on
[Slack](https://join.slack.com/t/colossalaiworkspace/shared_invite/zt-z7b26eeb-CBp7jouvu~r0~lcFzX832w),
and [WeChat(微信)](https://raw.githubusercontent.com/hpcaitech/public_assets/main/colossalai/img/WeChat.png "qrcode") to share your ideas.
4. Send your official proposal to email contact@hpcaitech.com
Thanks so much to all of our amazing contributors!
## Quick Preview
<p id="ChatGPT_scaling" align="center">
<img src="https://raw.githubusercontent.com/hpcaitech/public_assets/main/applications/chatgpt/ChatGPT%20scaling.png" width=800/>
</p>
- Up to 7.73 times faster for single server training and 1.42 times faster for single-GPU inference
<p id="ChatGPT-1GPU" align="center">
<img src="https://raw.githubusercontent.com/hpcaitech/public_assets/main/applications/chatgpt/ChatGPT-1GPU.jpg" width=450/>
</p>
- Up to 10.3x growth in model capacity on one GPU
- A mini demo training process requires only 1.62GB of GPU memory (any consumer-grade GPU)
<p id="inference" align="center">
<img src="https://raw.githubusercontent.com/hpcaitech/public_assets/main/applications/chatgpt/LoRA%20data.jpg" width=600/>
</p>
- Increase the capacity of the fine-tuning model by up to 3.7 times on a single GPU
- Keep in a sufficiently high running speed
## Authors
Coati is developed by ColossalAI Team: [Fazzie](https://fazzie-key.cool/about/index.html), [FrankLeeeee](https://github.com/FrankLeeeee), [BlueRum](https://github.com/ht-zhou), [ver217](https://github.com/ver217)
The Phd student [Zangwei Zheng](https://github.com/zhengzangw) and [Xue Fuzhao](https://github.com/XueFuzhao) also contributed a lot to this project.
## Citations
```bibtex
@article{Hu2021LoRALA,
title = {LoRA: Low-Rank Adaptation of Large Language Models},
author = {Edward J. Hu and Yelong Shen and Phillip Wallis and Zeyuan Allen-Zhu and Yuanzhi Li and Shean Wang and Weizhu Chen},
journal = {ArXiv},
year = {2021},
volume = {abs/2106.09685}
}
@article{ouyang2022training,
title={Training language models to follow instructions with human feedback},
author={Ouyang, Long and Wu, Jeff and Jiang, Xu and Almeida, Diogo and Wainwright, Carroll L and Mishkin, Pamela and Zhang, Chong and Agarwal, Sandhini and Slama, Katarina and Ray, Alex and others},
journal={arXiv preprint arXiv:2203.02155},
year={2022}
}
@article{touvron2023llama,
title={LLaMA: Open and Efficient Foundation Language Models},
author={Touvron, Hugo and Lavril, Thibaut and Izacard, Gautier and Martinet, Xavier and Lachaux, Marie-Anne and Lacroix, Timoth{\'e}e and Rozi{\`e}re, Baptiste and Goyal, Naman and Hambro, Eric and Azhar, Faisal and Rodriguez, Aurelien and Joulin, Armand and Grave, Edouard and Lample, Guillaume},
journal={arXiv preprint arXiv:2302.13971},
year={2023}
}
@misc{alpaca,
author = {Rohan Taori and Ishaan Gulrajani and Tianyi Zhang and Yann Dubois and Xuechen Li and Carlos Guestrin and Percy Liang and Tatsunori B. Hashimoto },
title = {Stanford Alpaca: An Instruction-following LLaMA model},
year = {2023},
publisher = {GitHub},
journal = {GitHub repository},
howpublished = {\url{https://github.com/tatsu-lab/stanford_alpaca}},
}
```
## Licenses
Coati is licensed under the [Apache 2.0 License](LICENSE).
applications/Chat/benchmarks/README.md 0 → 100644
View file @ b0ce5a10
# Benchmarks
## Benchmark GPT on dummy prompt data
We provide various GPT models (string in parentheses is the corresponding model name used in this script):
- GPT2-S (s)
- GPT2-M (m)
- GPT2-L (l)
- GPT2-XL (xl)
- GPT2-4B (4b)
- GPT2-6B (6b)
- GPT2-8B (8b)
- GPT2-10B (10b)
- GPT2-12B (12b)
- GPT2-15B (15b)
- GPT2-18B (18b)
- GPT2-20B (20b)
- GPT2-24B (24b)
- GPT2-28B (28b)
- GPT2-32B (32b)
- GPT2-36B (36b)
- GPT2-40B (40b)
- GPT3 (175b)
We also provide various training strategies:
- ddp: torch DDP
- colossalai_gemini: ColossalAI GeminiDDP with `placement_policy="cuda"`, like zero3
- colossalai_gemini_cpu: ColossalAI GeminiDDP with `placement_policy="cpu"`, like zero3-offload
- colossalai_zero2: ColossalAI zero2
- colossalai_zero2_cpu: ColossalAI zero2-offload
- colossalai_zero1: ColossalAI zero1
- colossalai_zero1_cpu: ColossalAI zero1-offload
We only support `torchrun` to launch now. E.g.
```shell
# run GPT2-S on single-node single-GPU with min batch size
torchrun --standalone --nproc_per_node 1 benchmark_gpt_dummy.py --model s --strategy ddp --experience_batch_size 1 --train_batch_size 1
# run GPT2-XL on single-node 4-GPU
torchrun --standalone --nproc_per_node 4 benchmark_gpt_dummy.py --model xl --strategy colossalai_zero2
# run GPT3 on 8-node 8-GPU
torchrun --nnodes 8 --nproc_per_node 8 \
--rdzv_id=$JOB_ID --rdzv_backend=c10d --rdzv_endpoint=$HOST_NODE_ADDR \
benchmark_gpt_dummy.py --model 175b --strategy colossalai_gemini
```
> ⚠ Batch sizes in CLI args and outputed throughput/TFLOPS are all values of per GPU.
In this benchmark, we assume the model architectures/sizes of actor and critic are the same for simplicity. But in practice, to reduce training cost, we may use a smaller critic.
We also provide a simple shell script to run a set of benchmarks. But it only supports benchmark on single node. However, it's easy to run on multi-nodes by modifying launch command in this script.
Usage:
```shell
# run for GPUS=(1 2 4 8) x strategy=("ddp" "colossalai_zero2" "colossalai_gemini" "colossalai_zero2_cpu" "colossalai_gemini_cpu") x model=("s" "m" "l" "xl" "2b" "4b" "6b" "8b" "10b") x batch_size=(1 2 4 8 16 32 64 128 256)
./benchmark_gpt_dummy.sh
# run for GPUS=2 x strategy=("ddp" "colossalai_zero2" "colossalai_gemini" "colossalai_zero2_cpu" "colossalai_gemini_cpu") x model=("s" "m" "l" "xl" "2b" "4b" "6b" "8b" "10b") x batch_size=(1 2 4 8 16 32 64 128 256)
./benchmark_gpt_dummy.sh 2
# run for GPUS=2 x strategy=ddp x model=("s" "m" "l" "xl" "2b" "4b" "6b" "8b" "10b") x batch_size=(1 2 4 8 16 32 64 128 256)
./benchmark_gpt_dummy.sh 2 ddp
# run for GPUS=2 x strategy=ddp x model=l x batch_size=(1 2 4 8 16 32 64 128 256)
./benchmark_gpt_dummy.sh 2 ddp l
```
## Benchmark OPT with LoRA on dummy prompt data
We provide various OPT models (string in parentheses is the corresponding model name used in this script):
- OPT-125M (125m)
- OPT-350M (350m)
- OPT-700M (700m)
- OPT-1.3B (1.3b)
- OPT-2.7B (2.7b)
- OPT-3.5B (3.5b)
- OPT-5.5B (5.5b)
- OPT-6.7B (6.7b)
- OPT-10B (10b)
- OPT-13B (13b)
We only support `torchrun` to launch now. E.g.
```shell
# run OPT-125M with no lora (lora_rank=0) on single-node single-GPU with min batch size
torchrun --standalone --nproc_per_node 1 benchmark_opt_lora_dummy.py --model 125m --strategy ddp --experience_batch_size 1 --train_batch_size 1 --lora_rank 0
# run OPT-350M with lora_rank=4 on single-node 4-GPU
torchrun --standalone --nproc_per_node 4 benchmark_opt_lora_dummy.py --model 350m --strategy colossalai_zero2 --lora_rank 4
```
> ⚠ Batch sizes in CLI args and outputed throughput/TFLOPS are all values of per GPU.
In this benchmark, we assume the model architectures/sizes of actor and critic are the same for simplicity. But in practice, to reduce training cost, we may use a smaller critic.
applications/Chat/benchmarks/benchmark_gpt_dummy.py 0 → 100644
View file @ b0ce5a10
import argparse
from copy import deepcopy
import torch
import torch.distributed as dist
import torch.nn as nn
from coati.models.base import RewardModel
from coati.models.gpt import GPTActor, GPTCritic
from coati.trainer import PPOTrainer
from coati.trainer.callbacks import PerformanceEvaluator
from coati.trainer.strategies import ColossalAIStrategy, DDPStrategy, Strategy
from torch.optim import Adam
from transformers.models.gpt2.configuration_gpt2 import GPT2Config
from transformers.models.gpt2.tokenization_gpt2 import GPT2Tokenizer
from colossalai.nn.optimizer import HybridAdam
def get_model_numel(model: nn.Module, strategy: Strategy) -> int:
numel = sum(p.numel() for p in model.parameters())
if isinstance(strategy, ColossalAIStrategy) and strategy.stage == 3 and strategy.shard_init:
numel *= dist.get_world_size()
return numel
def preprocess_batch(samples) -> dict:
input_ids = torch.stack(samples)
attention_mask = torch.ones_like(input_ids, dtype=torch.long)
return {'input_ids': input_ids, 'attention_mask': attention_mask}
def print_rank_0(*args, **kwargs) -> None:
if dist.get_rank() == 0:
print(*args, **kwargs)
def print_model_numel(model_dict: dict) -> None:
B = 1024**3
M = 1024**2
K = 1024
outputs = ''
for name, numel in model_dict.items():
outputs += f'{name}: '
if numel >= B:
outputs += f'{numel / B:.2f} B\n'
elif numel >= M:
outputs += f'{numel / M:.2f} M\n'
elif numel >= K:
outputs += f'{numel / K:.2f} K\n'
else:
outputs += f'{numel}\n'
print_rank_0(outputs)
def get_gpt_config(model_name: str) -> GPT2Config:
model_map = {
's': GPT2Config(),
'm': GPT2Config(n_embd=1024, n_layer=24, n_head=16),
'l': GPT2Config(n_embd=1280, n_layer=36, n_head=20),
'xl': GPT2Config(n_embd=1600, n_layer=48, n_head=25),
'2b': GPT2Config(n_embd=2048, n_layer=40, n_head=16),
'4b': GPT2Config(n_embd=2304, n_layer=64, n_head=16),
'6b': GPT2Config(n_embd=4096, n_layer=30, n_head=16),
'8b': GPT2Config(n_embd=4096, n_layer=40, n_head=16),
'10b': GPT2Config(n_embd=4096, n_layer=50, n_head=16),
'12b': GPT2Config(n_embd=4096, n_layer=60, n_head=16),
'15b': GPT2Config(n_embd=4096, n_layer=78, n_head=16),
'18b': GPT2Config(n_embd=4096, n_layer=90, n_head=16),
'20b': GPT2Config(n_embd=8192, n_layer=25, n_head=16),
'24b': GPT2Config(n_embd=8192, n_layer=30, n_head=16),
'28b': GPT2Config(n_embd=8192, n_layer=35, n_head=16),
'32b': GPT2Config(n_embd=8192, n_layer=40, n_head=16),
'36b': GPT2Config(n_embd=8192, n_layer=45, n_head=16),
'40b': GPT2Config(n_embd=8192, n_layer=50, n_head=16),
'175b': GPT2Config(n_positions=2048, n_embd=12288, n_layer=96, n_head=96),
}
try:
return model_map[model_name]
except KeyError:
raise ValueError(f'Unknown model "{model_name}"')
def main(args):
if args.strategy == 'ddp':
strategy = DDPStrategy()
elif args.strategy == 'colossalai_gemini':
strategy = ColossalAIStrategy(stage=3, placement_policy='cuda', initial_scale=2**5)
elif args.strategy == 'colossalai_gemini_cpu':
strategy = ColossalAIStrategy(stage=3, placement_policy='cpu', initial_scale=2**5)
elif args.strategy == 'colossalai_zero2':
strategy = ColossalAIStrategy(stage=2, placement_policy='cuda')
elif args.strategy == 'colossalai_zero2_cpu':
strategy = ColossalAIStrategy(stage=2, placement_policy='cpu')
elif args.strategy == 'colossalai_zero1':
strategy = ColossalAIStrategy(stage=1, placement_policy='cuda')
elif args.strategy == 'colossalai_zero1_cpu':
strategy = ColossalAIStrategy(stage=1, placement_policy='cpu')
else:
raise ValueError(f'Unsupported strategy "{args.strategy}"')
model_config = get_gpt_config(args.model)
with strategy.model_init_context():
actor = GPTActor(config=model_config).cuda()
critic = GPTCritic(config=model_config).cuda()
initial_model = deepcopy(actor).cuda()
reward_model = RewardModel(deepcopy(critic.model), deepcopy(critic.value_head)).cuda()
actor_numel = get_model_numel(actor, strategy)
critic_numel = get_model_numel(critic, strategy)
initial_model_numel = get_model_numel(initial_model, strategy)
reward_model_numel = get_model_numel(reward_model, strategy)
print_model_numel({
'Actor': actor_numel,
'Critic': critic_numel,
'Initial model': initial_model_numel,
'Reward model': reward_model_numel
})
performance_evaluator = PerformanceEvaluator(actor_numel,
critic_numel,
initial_model_numel,
reward_model_numel,
enable_grad_checkpoint=False,
ignore_episodes=1)
if args.strategy.startswith('colossalai'):
actor_optim = HybridAdam(actor.parameters(), lr=5e-6)
critic_optim = HybridAdam(critic.parameters(), lr=5e-6)
else:
actor_optim = Adam(actor.parameters(), lr=5e-6)
critic_optim = Adam(critic.parameters(), lr=5e-6)
tokenizer = GPT2Tokenizer.from_pretrained('gpt2')
tokenizer.pad_token = tokenizer.eos_token
(actor, actor_optim), (critic, critic_optim), reward_model, initial_model = strategy.prepare(
(actor, actor_optim), (critic, critic_optim), reward_model, initial_model)
trainer = PPOTrainer(strategy,
actor,
critic,
reward_model,
initial_model,
actor_optim,
critic_optim,
max_epochs=args.max_epochs,
train_batch_size=args.train_batch_size,
experience_batch_size=args.experience_batch_size,
tokenizer=preprocess_batch,
max_length=512,
do_sample=True,
temperature=1.0,
top_k=50,
pad_token_id=tokenizer.pad_token_id,
eos_token_id=tokenizer.eos_token_id,
callbacks=[performance_evaluator])
random_prompts = torch.randint(tokenizer.vocab_size, (1000, 400), device=torch.cuda.current_device())
trainer.fit(random_prompts,
num_episodes=args.num_episodes,
max_timesteps=args.max_timesteps,
update_timesteps=args.update_timesteps)
print_rank_0(f'Peak CUDA mem: {torch.cuda.max_memory_allocated()/1024**3:.2f} GB')
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--model', default='s')
parser.add_argument('--strategy',
choices=[
'ddp', 'colossalai_gemini', 'colossalai_gemini_cpu', 'colossalai_zero2',
'colossalai_zero2_cpu', 'colossalai_zero1', 'colossalai_zero1_cpu'
],
default='ddp')
parser.add_argument('--num_episodes', type=int, default=3)
parser.add_argument('--max_timesteps', type=int, default=8)
parser.add_argument('--update_timesteps', type=int, default=8)
parser.add_argument('--max_epochs', type=int, default=3)
parser.add_argument('--train_batch_size', type=int, default=8)
parser.add_argument('--experience_batch_size', type=int, default=8)
args = parser.parse_args()
main(args)
applications/Chat/benchmarks/benchmark_gpt_dummy.sh 0 → 100755
View file @ b0ce5a10
#!/usr/bin/env bash
# Usage: $0 <?number-of-gpus> <?strategy> <?model>
set -xu
BASE=$(realpath $(dirname $0))
PY_SCRIPT=${BASE}/benchmark_gpt_dummy.py
export OMP_NUM_THREADS=8
function tune_batch_size() {
# we found when experience batch size is equal to train batch size
# peak CUDA memory usage of making experience phase is less than or equal to that of training phase
# thus, experience batch size can be larger than or equal to train batch size
for bs in 1 2 4 8 16 32 64 128 256; do
torchrun --standalone --nproc_per_node $1 $PY_SCRIPT --model $2 --strategy $3 --experience_batch_size $bs --train_batch_size $bs || return 1
done
}
if [ $# -eq 0 ]; then
num_gpus=(1 2 4 8)
else
num_gpus=($1)
fi
if [ $# -le 1 ]; then
strategies=("ddp" "colossalai_zero2" "colossalai_gemini" "colossalai_zero2_cpu" "colossalai_gemini_cpu")
else
strategies=($2)
fi
if [ $# -le 2 ]; then
models=("s" "m" "l" "xl" "2b" "4b" "6b" "8b" "10b")
else
models=($3)
fi
for num_gpu in ${num_gpus[@]}; do
for strategy in ${strategies[@]}; do
for model in ${models[@]}; do
tune_batch_size $num_gpu $model $strategy || break
done
done
done
applications/Chat/benchmarks/benchmark_opt_lora_dummy.py 0 → 100644
View file @ b0ce5a10
import argparse
from copy import deepcopy
import torch
import torch.distributed as dist
import torch.nn as nn
from coati.models.base import RewardModel
from coati.models.opt import OPTActor, OPTCritic
from coati.trainer import PPOTrainer
from coati.trainer.callbacks import PerformanceEvaluator
from coati.trainer.strategies import ColossalAIStrategy, DDPStrategy, Strategy
from torch.optim import Adam
from transformers import AutoTokenizer
from transformers.models.opt.configuration_opt import OPTConfig
from colossalai.nn.optimizer import HybridAdam
def get_model_numel(model: nn.Module, strategy: Strategy) -> int:
numel = sum(p.numel() for p in model.parameters())
if isinstance(strategy, ColossalAIStrategy) and strategy.stage == 3 and strategy.shard_init:
numel *= dist.get_world_size()
return numel
def preprocess_batch(samples) -> dict:
input_ids = torch.stack(samples)
attention_mask = torch.ones_like(input_ids, dtype=torch.long)
return {'input_ids': input_ids, 'attention_mask': attention_mask}
def print_rank_0(*args, **kwargs) -> None:
if dist.get_rank() == 0:
print(*args, **kwargs)
def print_model_numel(model_dict: dict) -> None:
B = 1024**3
M = 1024**2
K = 1024
outputs = ''
for name, numel in model_dict.items():
outputs += f'{name}: '
if numel >= B:
outputs += f'{numel / B:.2f} B\n'
elif numel >= M:
outputs += f'{numel / M:.2f} M\n'
elif numel >= K:
outputs += f'{numel / K:.2f} K\n'
else:
outputs += f'{numel}\n'
print_rank_0(outputs)
def get_gpt_config(model_name: str) -> OPTConfig:
model_map = {
'125m': OPTConfig.from_pretrained('facebook/opt-125m'),
'350m': OPTConfig(hidden_size=1024, ffn_dim=4096, num_hidden_layers=24, num_attention_heads=16),
'700m': OPTConfig(hidden_size=1280, ffn_dim=5120, num_hidden_layers=36, num_attention_heads=20),
'1.3b': OPTConfig.from_pretrained('facebook/opt-1.3b'),
'2.7b': OPTConfig.from_pretrained('facebook/opt-2.7b'),
'3.5b': OPTConfig(hidden_size=3072, ffn_dim=12288, num_hidden_layers=32, num_attention_heads=32),
'5.5b': OPTConfig(hidden_size=3840, ffn_dim=15360, num_hidden_layers=32, num_attention_heads=32),
'6.7b': OPTConfig.from_pretrained('facebook/opt-6.7b'),
'10b': OPTConfig(hidden_size=5120, ffn_dim=20480, num_hidden_layers=32, num_attention_heads=32),
'13b': OPTConfig.from_pretrained('facebook/opt-13b'),
}
try:
return model_map[model_name]
except KeyError:
raise ValueError(f'Unknown model "{model_name}"')
def main(args):
if args.strategy == 'ddp':
strategy = DDPStrategy()
elif args.strategy == 'colossalai_gemini':
strategy = ColossalAIStrategy(stage=3, placement_policy='cuda', initial_scale=2**5)
elif args.strategy == 'colossalai_gemini_cpu':
strategy = ColossalAIStrategy(stage=3, placement_policy='cpu', initial_scale=2**5)
elif args.strategy == 'colossalai_zero2':
strategy = ColossalAIStrategy(stage=2, placement_policy='cuda')
elif args.strategy == 'colossalai_zero2_cpu':
strategy = ColossalAIStrategy(stage=2, placement_policy='cpu')
elif args.strategy == 'colossalai_zero1':
strategy = ColossalAIStrategy(stage=1, placement_policy='cuda')
elif args.strategy == 'colossalai_zero1_cpu':
strategy = ColossalAIStrategy(stage=1, placement_policy='cpu')
else:
raise ValueError(f'Unsupported strategy "{args.strategy}"')
torch.cuda.set_per_process_memory_fraction(args.cuda_mem_frac)
model_config = get_gpt_config(args.model)
with strategy.model_init_context():
actor = OPTActor(config=model_config, lora_rank=args.lora_rank).cuda()
critic = OPTCritic(config=model_config, lora_rank=args.lora_rank).cuda()
initial_model = deepcopy(actor).cuda()
reward_model = RewardModel(deepcopy(critic.model), deepcopy(critic.value_head)).cuda()
actor_numel = get_model_numel(actor, strategy)
critic_numel = get_model_numel(critic, strategy)
initial_model_numel = get_model_numel(initial_model, strategy)
reward_model_numel = get_model_numel(reward_model, strategy)
print_model_numel({
'Actor': actor_numel,
'Critic': critic_numel,
'Initial model': initial_model_numel,
'Reward model': reward_model_numel
})
performance_evaluator = PerformanceEvaluator(actor_numel,
critic_numel,
initial_model_numel,
reward_model_numel,
enable_grad_checkpoint=False,
ignore_episodes=1)
if args.strategy.startswith('colossalai'):
actor_optim = HybridAdam(actor.parameters(), lr=5e-6)
critic_optim = HybridAdam(critic.parameters(), lr=5e-6)
else:
actor_optim = Adam(actor.parameters(), lr=5e-6)
critic_optim = Adam(critic.parameters(), lr=5e-6)
tokenizer = AutoTokenizer.from_pretrained('facebook/opt-350m')
tokenizer.pad_token = tokenizer.eos_token
(actor, actor_optim), (critic, critic_optim), reward_model, initial_model = strategy.prepare(
(actor, actor_optim), (critic, critic_optim), reward_model, initial_model)
trainer = PPOTrainer(strategy,
actor,
critic,
reward_model,
initial_model,
actor_optim,
critic_optim,
max_epochs=args.max_epochs,
train_batch_size=args.train_batch_size,
experience_batch_size=args.experience_batch_size,
tokenizer=preprocess_batch,
max_length=512,
do_sample=True,
temperature=1.0,
top_k=50,
pad_token_id=tokenizer.pad_token_id,
eos_token_id=tokenizer.eos_token_id,
callbacks=[performance_evaluator])
random_prompts = torch.randint(tokenizer.vocab_size, (1000, 400), device=torch.cuda.current_device())
trainer.fit(random_prompts,
num_episodes=args.num_episodes,
max_timesteps=args.max_timesteps,
update_timesteps=args.update_timesteps)
print_rank_0(f'Peak CUDA mem: {torch.cuda.max_memory_allocated()/1024**3:.2f} GB')
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--model', default='125m')
parser.add_argument('--strategy',
choices=[
'ddp', 'colossalai_gemini', 'colossalai_gemini_cpu', 'colossalai_zero2',
'colossalai_zero2_cpu', 'colossalai_zero1', 'colossalai_zero1_cpu'
],
default='ddp')
parser.add_argument('--num_episodes', type=int, default=3)
parser.add_argument('--max_timesteps', type=int, default=8)
parser.add_argument('--update_timesteps', type=int, default=8)
parser.add_argument('--max_epochs', type=int, default=3)
parser.add_argument('--train_batch_size', type=int, default=8)
parser.add_argument('--experience_batch_size', type=int, default=8)
parser.add_argument('--lora_rank', type=int, default=4)
parser.add_argument('--cuda_mem_frac', type=float, default=1.0)
args = parser.parse_args()
main(args)
applications/Chat/coati/dataset/__init__.py 0 → 100644
View file @ b0ce5a10
from .prompt_dataset import PromptDataset
from .reward_dataset import HhRlhfDataset, RmStaticDataset
from .sft_dataset import DataCollatorForSupervisedDataset, SFTDataset, SupervisedDataset
from .utils import is_rank_0
__all__ = [
'RmStaticDataset', 'HhRlhfDataset', 'is_rank_0', 'SFTDataset', 'SupervisedDataset',
'DataCollatorForSupervisedDataset', 'PromptDataset'
]
applications/Chat/coati/dataset/prompt_dataset.py 0 → 100644
View file @ b0ce5a10
import copy
import random
from dataclasses import dataclass, field
from typing import Callable, Dict, Sequence
import torch
import torch.distributed as dist
import transformers
from torch.utils.data import Dataset
from tqdm import tqdm
from colossalai.logging import get_dist_logger
from .utils import is_rank_0, jload
logger = get_dist_logger()
class PromptDataset(Dataset):
"""Dataset for supervised fine-tuning."""
def __init__(self, data_path: str, tokenizer: transformers.PreTrainedTokenizer, max_datasets_size: int = None):
super(PromptDataset, self).__init__()
self.prompt = []
logger.info("Loading data...")
list_data_dict = jload(data_path)
logger.info(f"Loaded {len(list_data_dict)} examples.")
if max_datasets_size is not None:
logger.info(f"Limiting dataset to {max_datasets_size} examples.")
list_data_dict = list_data_dict[:max_datasets_size]
for data_dict in list_data_dict:
token = tokenizer(data_dict["instruction"],
return_tensors='pt',
max_length=96,
padding='max_length',
truncation=True)
for idx in token['input_ids']:
self.prompt.append(idx.to(torch.cuda.current_device()))
def __len__(self):
return len(self.prompt)
def __getitem__(self, i) -> Dict[str, torch.Tensor]:
return self.prompt[i]
applications/Chat/coati/dataset/reward_dataset.py 0 → 100644
View file @ b0ce5a10
from typing import Callable
from torch.utils.data import Dataset
from tqdm import tqdm
from .utils import is_rank_0
# Dahaos/rm-static
class RmStaticDataset(Dataset):
"""
Dataset for reward model
Args:
dataset: dataset for reward model
tokenizer: tokenizer for reward model
max_length: max length of input
special_token: special token at the end of sentence
"""
def __init__(self, dataset, tokenizer: Callable, max_length: int, special_token=None) -> None:
super().__init__()
self.chosen = []
self.reject = []
if special_token is None:
self.end_token = tokenizer.eos_token
else:
self.end_token = special_token
for data in tqdm(dataset, disable=not is_rank_0()):
prompt = data['prompt']
chosen = prompt + data['chosen'] + self.end_token
chosen_token = tokenizer(chosen,
max_length=max_length,
padding="max_length",
truncation=True,
return_tensors="pt")
self.chosen.append({
"input_ids": chosen_token['input_ids'],
"attention_mask": chosen_token['attention_mask']
})
reject = prompt + data['rejected'] + self.end_token
reject_token = tokenizer(reject,
max_length=max_length,
padding="max_length",
truncation=True,
return_tensors="pt")
self.reject.append({
"input_ids": reject_token['input_ids'],
"attention_mask": reject_token['attention_mask']
})
def __len__(self):
length = len(self.chosen)
return length
def __getitem__(self, idx):
return self.chosen[idx]["input_ids"], self.chosen[idx]["attention_mask"], self.reject[idx][
"input_ids"], self.reject[idx]["attention_mask"]
# Anthropic/hh-rlhf
class HhRlhfDataset(Dataset):
"""
Dataset for reward model
Args:
dataset: dataset for reward model
tokenizer: tokenizer for reward model
max_length: max length of input
special_token: special token at the end of sentence
"""
def __init__(self, dataset, tokenizer: Callable, max_length: int, special_token=None) -> None:
super().__init__()
self.chosen = []
self.reject = []
if special_token is None:
self.end_token = tokenizer.eos_token
else:
self.end_token = special_token
for data in tqdm(dataset, disable=not is_rank_0()):
chosen = data['chosen'] + self.end_token
chosen_token = tokenizer(chosen,
max_length=max_length,
padding="max_length",
truncation=True,
return_tensors="pt")
self.chosen.append({
"input_ids": chosen_token['input_ids'],
"attention_mask": chosen_token['attention_mask']
})
reject = data['rejected'] + self.end_token
reject_token = tokenizer(reject,
max_length=max_length,
padding="max_length",
truncation=True,
return_tensors="pt")
self.reject.append({
"input_ids": reject_token['input_ids'],
"attention_mask": reject_token['attention_mask']
})
def __len__(self):
length = len(self.chosen)
return length
def __getitem__(self, idx):
return self.chosen[idx]["input_ids"], self.chosen[idx]["attention_mask"], self.reject[idx][
"input_ids"], self.reject[idx]["attention_mask"]
applications/Chat/coati/dataset/sft_dataset.py 0 → 100644
View file @ b0ce5a10
# Copyright 2023 Rohan Taori, Ishaan Gulrajani, Tianyi Zhang, Yann Dubois, Xuechen Li
#
# 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.
import copy
import random
from dataclasses import dataclass, field
from typing import Callable, Dict, Sequence
import torch
import torch.distributed as dist
import transformers
from torch.utils.data import Dataset
from tqdm import tqdm
from colossalai.logging import get_dist_logger
from .utils import is_rank_0, jload
logger = get_dist_logger()
IGNORE_INDEX = -100
PROMPT_DICT = {
"prompt_input":
("Below is an instruction that describes a task, paired with an input that provides further context. "
"Write a response that appropriately completes the request.\n\n"
"### Instruction:\n{instruction}\n\n### Input:\n{input}\n\n### Response:"),
"prompt_no_input": ("Below is an instruction that describes a task. "
"Write a response that appropriately completes the request.\n\n"
"### Instruction:\n{instruction}\n\n### Response:"),
}
class SFTDataset(Dataset):
"""
Dataset for sft model
Args:
dataset: dataset for supervised model
tokenizer: tokenizer for supervised model
max_length: max length of input
"""
def __init__(self, dataset, tokenizer: Callable, max_length: int = 512) -> None:
super().__init__()
# self.prompts = []
self.input_ids = []
for data in tqdm(dataset, disable=not is_rank_0()):
prompt = data['prompt'] + data['completion'] + "<|endoftext|>"
prompt_token = tokenizer(prompt,
max_length=max_length,
padding="max_length",
truncation=True,
return_tensors="pt")
# self.prompts.append(prompt_token)s
self.input_ids.append(prompt_token)
self.labels = copy.deepcopy(self.input_ids)
def __len__(self):
length = len(self.prompts)
return length
def __getitem__(self, idx):
# dict(input_ids=self.input_ids[i], labels=self.labels[i])
return dict(input_ids=self.input_ids[idx], labels=self.labels[idx])
# return dict(self.prompts[idx], self.prompts[idx])
def _tokenize_fn(strings: Sequence[str], tokenizer: transformers.PreTrainedTokenizer) -> Dict:
"""Tokenize a list of strings."""
tokenized_list = [
tokenizer(
text,
return_tensors="pt",
padding="longest",
max_length=tokenizer.model_max_length,
truncation=True,
) for text in strings
]
input_ids = labels = [tokenized.input_ids[0] for tokenized in tokenized_list]
input_ids_lens = labels_lens = [
tokenized.input_ids.ne(tokenizer.pad_token_id).sum().item() for tokenized in tokenized_list
]
return dict(
input_ids=input_ids,
labels=labels,
input_ids_lens=input_ids_lens,
labels_lens=labels_lens,
)
def preprocess(
sources: Sequence[str],
targets: Sequence[str],
tokenizer: transformers.PreTrainedTokenizer,
) -> Dict:
"""Preprocess the data by tokenizing."""
examples = [s + t for s, t in zip(sources, targets)]
examples_tokenized, sources_tokenized = [_tokenize_fn(strings, tokenizer) for strings in (examples, sources)]
input_ids = examples_tokenized["input_ids"]
labels = copy.deepcopy(input_ids)
for label, source_len in zip(labels, sources_tokenized["input_ids_lens"]):
label[:source_len] = IGNORE_INDEX
return dict(input_ids=input_ids, labels=labels)
class SupervisedDataset(Dataset):
"""Dataset for supervised fine-tuning."""
def __init__(self, data_path: str, tokenizer: transformers.PreTrainedTokenizer, max_datasets_size: int = None):
super(SupervisedDataset, self).__init__()
logger.info("Loading data...")
list_data_dict = jload(data_path)
logger.info(f"Loaded {len(list_data_dict)} examples.")
if max_datasets_size is not None:
logger.info(f"Limiting dataset to {max_datasets_size} examples.")
list_data_dict = list_data_dict[:max_datasets_size]
logger.info("Formatting inputs...")
prompt_input, prompt_no_input = PROMPT_DICT["prompt_input"], PROMPT_DICT["prompt_no_input"]
sources = [
prompt_input.format_map(example) if example.get("input", "") != "" else prompt_no_input.format_map(example)
for example in list_data_dict
]
targets = [f"{example['output']}{tokenizer.eos_token}" for example in list_data_dict]
logger.info("Tokenizing inputs... This may take some time...")
data_dict = preprocess(sources, targets, tokenizer)
self.input_ids = data_dict["input_ids"]
self.labels = data_dict["labels"]
def __len__(self):
return len(self.input_ids)
def __getitem__(self, i) -> Dict[str, torch.Tensor]:
return dict(input_ids=self.input_ids[i], labels=self.labels[i])
@dataclass
class DataCollatorForSupervisedDataset(object):
"""Collate examples for supervised fine-tuning."""
tokenizer: transformers.PreTrainedTokenizer
def __call__(self, instances: Sequence[Dict]) -> Dict[str, torch.Tensor]:
input_ids, labels = tuple([instance[key] for instance in instances] for key in ("input_ids", "labels"))
input_ids = torch.nn.utils.rnn.pad_sequence(input_ids,
batch_first=True,
padding_value=self.tokenizer.pad_token_id)
labels = torch.nn.utils.rnn.pad_sequence(labels, batch_first=True, padding_value=IGNORE_INDEX)
return dict(
input_ids=input_ids,
labels=labels,
attention_mask=input_ids.ne(self.tokenizer.pad_token_id),
)
applications/Chat/coati/dataset/utils.py 0 → 100644
View file @ b0ce5a10
import io
import json
import torch.distributed as dist
def is_rank_0() -> bool:
return not dist.is_initialized() or dist.get_rank() == 0
def _make_r_io_base(f, mode: str):
if not isinstance(f, io.IOBase):
f = open(f, mode=mode)
return f
def jload(f, mode="r"):
"""Load a .json file into a dictionary."""
f = _make_r_io_base(f, mode)
jdict = json.load(f)
f.close()
return jdict
applications/Chat/coati/experience_maker/__init__.py 0 → 100644
View file @ b0ce5a10
from .base import Experience, ExperienceMaker
from .naive import NaiveExperienceMaker
__all__ = ['Experience', 'ExperienceMaker', 'NaiveExperienceMaker']
applications/Chat/coati/experience_maker/base.py 0 → 100644
View file @ b0ce5a10
from abc import ABC, abstractmethod
from dataclasses import dataclass
from typing import Optional
import torch
import torch.nn as nn
from coati.models.base import Actor
@dataclass
class Experience:
"""Experience is a batch of data.
These data should have the the sequence length and number of actions.
Left padding for sequences is applied.
Shapes of each tensor:
sequences: (B, S)
action_log_probs: (B, A)
values: (B)
reward: (B)
advatanges: (B)
attention_mask: (B, S)
action_mask: (B, A)
"A" is the number of actions.
"""
sequences: torch.Tensor
action_log_probs: torch.Tensor
values: torch.Tensor
reward: torch.Tensor
advantages: torch.Tensor
attention_mask: Optional[torch.LongTensor]
action_mask: Optional[torch.BoolTensor]
@torch.no_grad()
def to_device(self, device: torch.device) -> None:
self.sequences = self.sequences.to(device)
self.action_log_probs = self.action_log_probs.to(device)
self.values = self.values.to(device)
self.reward = self.reward.to(device)
self.advantages = self.advantages.to(device)
if self.attention_mask is not None:
self.attention_mask = self.attention_mask.to(device)
if self.action_mask is not None:
self.action_mask = self.action_mask.to(device)
def pin_memory(self):
self.sequences = self.sequences.pin_memory()
self.action_log_probs = self.action_log_probs.pin_memory()
self.values = self.values.pin_memory()
self.reward = self.reward.pin_memory()
self.advantages = self.advantages.pin_memory()
if self.attention_mask is not None:
self.attention_mask = self.attention_mask.pin_memory()
if self.action_mask is not None:
self.action_mask = self.action_mask.pin_memory()
return self
class ExperienceMaker(ABC):
def __init__(self,
actor: Actor,
critic: nn.Module,
reward_model: nn.Module,
initial_model: Actor,
kl_coef: float = 0.1) -> None:
super().__init__()
self.actor = actor
self.critic = critic
self.reward_model = reward_model
self.initial_model = initial_model
self.kl_coef = kl_coef
@abstractmethod
def make_experience(self, input_ids: torch.Tensor, **generate_kwargs) -> Experience:
pass
applications/Chat/coati/experience_maker/naive.py 0 → 100644
View file @ b0ce5a10
import torch
from coati.models.utils import compute_reward, normalize
from .base import Experience, ExperienceMaker
class NaiveExperienceMaker(ExperienceMaker):
"""
Naive experience maker.
"""
@torch.no_grad()
def make_experience(self, input_ids: torch.Tensor, **generate_kwargs) -> Experience:
self.actor.eval()
self.critic.eval()
self.initial_model.eval()
self.reward_model.eval()
sequences, attention_mask, action_mask = self.actor.generate(input_ids,
return_action_mask=True,
**generate_kwargs)
num_actions = action_mask.size(1)
action_log_probs = self.actor(sequences, num_actions, attention_mask)
base_action_log_probs = self.initial_model(sequences, num_actions, attention_mask)
value = self.critic(sequences, action_mask, attention_mask)
r = self.reward_model(sequences, attention_mask)
reward = compute_reward(r, self.kl_coef, action_log_probs, base_action_log_probs, action_mask=action_mask)
advantage = reward - value
# TODO(ver217): maybe normalize adv
if advantage.ndim == 1:
advantage = advantage.unsqueeze(-1)
return Experience(sequences, action_log_probs, value, reward, advantage, attention_mask, action_mask)
applications/Chat/coati/models/__init__.py 0 → 100644
View file @ b0ce5a10
from .base import Actor, Critic, RewardModel
from .loss import LogExpLoss, LogSigLoss, PolicyLoss, PPOPtxActorLoss, ValueLoss
__all__ = ['Actor', 'Critic', 'RewardModel', 'PolicyLoss', 'ValueLoss', 'PPOPtxActorLoss', 'LogSigLoss', 'LogExpLoss']
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