diff --git a/applications/DeepSpeed-Chat/.gitignore b/applications/DeepSpeed-Chat/.gitignore
new file mode 100644
index 000000000..cbc922f8f
--- /dev/null
+++ b/applications/DeepSpeed-Chat/.gitignore
@@ -0,0 +1,137 @@
+# 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/
+
+# 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/
+
+results/
+outputs/
+
+.amltconfig
+.test_output
+*.hdf5
+*.h5
diff --git a/applications/DeepSpeed-Chat/README.md b/applications/DeepSpeed-Chat/README.md
index c0a456823..ac9f3ab6a 100644
--- a/applications/DeepSpeed-Chat/README.md
+++ b/applications/DeepSpeed-Chat/README.md
@@ -33,20 +33,24 @@ A fast, affordable, scalable and open system framework for enabling end-to-end R
 <!-- markdown-toc start - Don't edit this section. Run M-x markdown-toc-refresh-toc -->
 ## Table of Contents
 
+- [🐕DeepSpeed-Chat: Easy, Fast and Affordable RLHF Training of ChatGPT-like Models at All Scales🐕](#deepspeed-chat-easy-fast-and-affordable-rlhf-training-of-chatgpt-like-models-at-all-scales)
+- [Table of Contents](#table-of-contents)
 - [📰 Latest News 📰](#-latest-news-)
-- [🚀 What is DeepSpeed Chat 🚀️](#-what-is-deepspeed-chat-)
+- [🚀 What is DeepSpeed Chat 🚀](#-what-is-deepspeed-chat-)
 - [🧨 Capabilities 🧨](#-capabilities-)
 - [☕ Quick Start ☕](#-quick-start-)
   - [🐼 Installation](#-installation)
-  - [🐼 Single Script for Training 3-Step RLHF Pipeline](#-one-single-script-completes-all-three-stages-of-rlhf-training-and-generate-your-first-chatgpt-model)
+  - [🐼 One Single Script Completes All Three Steps of RLHF Training and Generate Your First ChatGPT Model](#-one-single-script-completes-all-three-steps-of-rlhf-training-and-generate-your-first-chatgpt-model)
   - [🐼 Demonstration: Individual Step Fine-Tuning](#-demonstration-individual-step-fine-tuning)
     - [🕐 Step 1 - Supervised Fine-Tuning](#-step-1---supervised-fine-tuning)
     - [🕑 Step 2 - Reward Model](#-step-2---reward-model)
     - [🕒 Step 3 - Reinforcement Learning with Human Feedback](#-step-3---reinforcement-learning-with-human-feedback)
-   - [🐼 Adding and using your own datasets in DeepSpeed-Chat](#-adding-and-using-your-own-datasets-in-deepspeed-chat)
-   - [🐼 Customizing RLHF training pipeline via DeepSpeed-Chat’s APIs](#-customizing-your-own-rlhf-training-pipeline-using-deepspeed-chats-rlhf-apis)
-   - [🐼 Serving Your Model: Plug-in and Test!](#-serving-plug-in-your-final-model-trained-by-deepspeed-chat-and-test-it-out)  
+  - [🐼 Adding and using your own datasets in DeepSpeed-Chat](#-adding-and-using-your-own-datasets-in-deepspeed-chat)
+  - [🐼 Customizing your own RLHF training pipeline using DeepSpeed-Chat’s RLHF APIs](#-customizing-your-own-rlhf-training-pipeline-using-deepspeed-chats-rlhf-apis)
+  - [🐼 Serving: Plug-in your final model trained by DeepSpeed-Chat and test it out!](#-serving-plug-in-your-final-model-trained-by-deepspeed-chat-and-test-it-out)
 - [🔥 Training Performance Evaluation 🔥](#-training-performance-evaluation-)
+  - [🐲 Superior Model Scale and Low Training Cost](#-superior-model-scale-and-low-training-cost)
+  - [🐲 Throughput and Model Size Scalability Comparisons with Existing RLHF Systems](#-throughput-and-model-size-scalability-comparisons-with-existing-rlhf-systems)
 - [😽 Supported Models 😽](#-supported-models-)
 - [🔬 Build Pipeline Status 🔬](#-build-pipeline-status-)
 - [⚓ Documentation and Tutorial ⚓](#-documentation-and-tutorial-)
@@ -119,6 +123,7 @@ pip install deepspeed>=0.9.0
 git clone https://github.com/microsoft/DeepSpeedExamples.git
 cd DeepSpeedExamples/applications/DeepSpeed-Chat/
 pip install -r requirements.txt
+pip install -e .
 ```
 
 ### 🐼 One Single Script Completes All Three Steps of RLHF Training and Generate Your First ChatGPT Model
diff --git a/applications/DeepSpeed-Chat/training/step3_rlhf_finetuning/ppo_trainer.py b/applications/DeepSpeed-Chat/dschat/rlhf/ppo_trainer.py
similarity index 99%
rename from applications/DeepSpeed-Chat/training/step3_rlhf_finetuning/ppo_trainer.py
rename to applications/DeepSpeed-Chat/dschat/rlhf/ppo_trainer.py
index 2a5056cfe..22cba6be0 100644
--- a/applications/DeepSpeed-Chat/training/step3_rlhf_finetuning/ppo_trainer.py
+++ b/applications/DeepSpeed-Chat/dschat/rlhf/ppo_trainer.py
@@ -4,17 +4,12 @@
 # DeepSpeed Team
 import torch
 import torch.nn.functional as F
-import sys
-import os
 import time
 import deepspeed
 from deepspeed.runtime.zero.partition_parameters import ZeroParamStatus
 from deepspeed.accelerator import get_accelerator
 
-sys.path.append(
-    os.path.abspath(os.path.join(os.path.dirname(__file__), os.path.pardir)))
-
-from utils.utils import print_rank_0
+from dschat.utils.utils import print_rank_0
 
 
 def print_all_ranks(tag, value, rank):
diff --git a/applications/DeepSpeed-Chat/training/step3_rlhf_finetuning/rlhf_engine.py b/applications/DeepSpeed-Chat/dschat/rlhf/rlhf_engine.py
similarity index 97%
rename from applications/DeepSpeed-Chat/training/step3_rlhf_finetuning/rlhf_engine.py
rename to applications/DeepSpeed-Chat/dschat/rlhf/rlhf_engine.py
index 3a192d017..5b6778cc2 100755
--- a/applications/DeepSpeed-Chat/training/step3_rlhf_finetuning/rlhf_engine.py
+++ b/applications/DeepSpeed-Chat/dschat/rlhf/rlhf_engine.py
@@ -9,10 +9,10 @@
 from deepspeed.ops.adam import DeepSpeedCPUAdam
 from transformers import AutoModelForCausalLM, get_scheduler
 
-from utils.ds_utils import get_train_ds_config, get_eval_ds_config
-from utils.module.lora import convert_linear_layer_to_lora, only_optimize_lora_parameters, make_model_gradient_checkpointing_compatible
-from utils.model.model_utils import create_hf_model, create_critic_model
-from utils.utils import get_optimizer_grouped_parameters
+from dschat.utils.ds_utils import get_train_ds_config, get_eval_ds_config
+from dschat.utils.module.lora import convert_linear_layer_to_lora, only_optimize_lora_parameters, make_model_gradient_checkpointing_compatible
+from dschat.utils.model.model_utils import create_hf_model, create_critic_model
+from dschat.utils.utils import get_optimizer_grouped_parameters
 """
 TODOs:
   * support HF models for critic (for debugging), must be a previously saved ckpt from step-2
diff --git a/applications/DeepSpeed-Chat/training/utils/data/data_utils.py b/applications/DeepSpeed-Chat/dschat/utils/data/data_utils.py
similarity index 99%
rename from applications/DeepSpeed-Chat/training/utils/data/data_utils.py
rename to applications/DeepSpeed-Chat/dschat/utils/data/data_utils.py
index 277ebd730..744ad7927 100644
--- a/applications/DeepSpeed-Chat/training/utils/data/data_utils.py
+++ b/applications/DeepSpeed-Chat/dschat/utils/data/data_utils.py
@@ -14,7 +14,7 @@
 import os
 import hashlib
 from itertools import chain
-from . import raw_datasets
+from dschat.utils.data import raw_datasets
 from deepspeed.accelerator import get_accelerator
 
 
diff --git a/applications/DeepSpeed-Chat/training/utils/data/raw_datasets.py b/applications/DeepSpeed-Chat/dschat/utils/data/raw_datasets.py
similarity index 100%
rename from applications/DeepSpeed-Chat/training/utils/data/raw_datasets.py
rename to applications/DeepSpeed-Chat/dschat/utils/data/raw_datasets.py
diff --git a/applications/DeepSpeed-Chat/training/utils/ds_utils.py b/applications/DeepSpeed-Chat/dschat/utils/ds_utils.py
similarity index 100%
rename from applications/DeepSpeed-Chat/training/utils/ds_utils.py
rename to applications/DeepSpeed-Chat/dschat/utils/ds_utils.py
diff --git a/applications/DeepSpeed-Chat/training/utils/model/model_utils.py b/applications/DeepSpeed-Chat/dschat/utils/model/model_utils.py
similarity index 97%
rename from applications/DeepSpeed-Chat/training/utils/model/model_utils.py
rename to applications/DeepSpeed-Chat/dschat/utils/model/model_utils.py
index 27e5720ca..97d3bff15 100644
--- a/applications/DeepSpeed-Chat/training/utils/model/model_utils.py
+++ b/applications/DeepSpeed-Chat/dschat/utils/model/model_utils.py
@@ -12,9 +12,8 @@
 from huggingface_hub import snapshot_download
 from transformers.deepspeed import HfDeepSpeedConfig
 
-from .reward_model import RewardModel
-from ..utils import load_state_dict_into_model
-from ..utils import print_rank_0
+from dschat.utils.model.reward_model import RewardModel
+from dschat.utils.utils import load_state_dict_into_model, print_rank_0
 
 
 def configure_dropout(model_config, dropout):
diff --git a/applications/DeepSpeed-Chat/training/utils/model/reward_model.py b/applications/DeepSpeed-Chat/dschat/utils/model/reward_model.py
similarity index 100%
rename from applications/DeepSpeed-Chat/training/utils/model/reward_model.py
rename to applications/DeepSpeed-Chat/dschat/utils/model/reward_model.py
diff --git a/applications/DeepSpeed-Chat/training/utils/module/lora.py b/applications/DeepSpeed-Chat/dschat/utils/module/lora.py
similarity index 100%
rename from applications/DeepSpeed-Chat/training/utils/module/lora.py
rename to applications/DeepSpeed-Chat/dschat/utils/module/lora.py
diff --git a/applications/DeepSpeed-Chat/training/utils/perf.py b/applications/DeepSpeed-Chat/dschat/utils/perf.py
similarity index 100%
rename from applications/DeepSpeed-Chat/training/utils/perf.py
rename to applications/DeepSpeed-Chat/dschat/utils/perf.py
diff --git a/applications/DeepSpeed-Chat/training/utils/utils.py b/applications/DeepSpeed-Chat/dschat/utils/utils.py
similarity index 100%
rename from applications/DeepSpeed-Chat/training/utils/utils.py
rename to applications/DeepSpeed-Chat/dschat/utils/utils.py
diff --git a/applications/DeepSpeed-Chat/train.py b/applications/DeepSpeed-Chat/e2e_rlhf.py
similarity index 100%
rename from applications/DeepSpeed-Chat/train.py
rename to applications/DeepSpeed-Chat/e2e_rlhf.py
diff --git a/applications/DeepSpeed-Chat/setup.py b/applications/DeepSpeed-Chat/setup.py
new file mode 100644
index 000000000..343f5b476
--- /dev/null
+++ b/applications/DeepSpeed-Chat/setup.py
@@ -0,0 +1,32 @@
+#!/usr/bin/env python
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# setup.py: install script for deepspeed_chat
+"""
+to install deepspeed_chat and its dependencies for development work,
+run this cmd from the root directory:
+    pip install -e .
+"""
+import setuptools
+
+setuptools.setup(
+    name="deepspeed-chat",
+    version="0.1",
+    url=
+    "https://github.com/microsoft/DeepSpeedExamples/tree/master/applications/DeepSpeed-Chat",
+    include_package_data=True,
+    packages=setuptools.find_packages(include=['dschat']),
+    install_requires=[
+        "datasets>=2.8.0", "sentencepiece>=0.1.97", "protobuf==3.20.3",
+        "accelerate>=0.15.0", "torch>=1.12.0", "deepspeed>=0.9.2",
+        "transformers", "tensorboard"
+    ],
+    extras_require={
+        "azureml": [
+            "azure-ml-component",
+            "azureml-core",
+        ],
+    })
diff --git a/applications/DeepSpeed-Chat/training/tests/test_training.py b/applications/DeepSpeed-Chat/tests/test_training.py
similarity index 97%
rename from applications/DeepSpeed-Chat/training/tests/test_training.py
rename to applications/DeepSpeed-Chat/tests/test_training.py
index 3be4f6ff6..7ffe02972 100644
--- a/applications/DeepSpeed-Chat/training/tests/test_training.py
+++ b/applications/DeepSpeed-Chat/tests/test_training.py
@@ -66,7 +66,7 @@ def test_ds_chat(zero_stage, hybrid_engine, offload, lora):
 
     # cd into execution dir
     wd = os.getcwd()
-    os.chdir("../step3_rlhf_finetuning")
+    os.chdir("../training/step3_rlhf_finetuning")
     sweep_script = "training_scripts/opt/single_node/sweep/run_single.sh"
 
     # Run bash script
@@ -85,3 +85,5 @@ def test_ds_chat(zero_stage, hybrid_engine, offload, lora):
                        ), "Actor model was not saved during step 3 training."
     assert file_exists(f"{output_path}/critic/", "pytorch_model.bin"
                        ), "Critic model was not saved during step 3 training."
+
+    os.chdir(wd)
diff --git a/applications/DeepSpeed-Chat/training/step1_supervised_finetuning/main.py b/applications/DeepSpeed-Chat/training/step1_supervised_finetuning/main.py
index 0332e2e3f..c37d1f4cd 100755
--- a/applications/DeepSpeed-Chat/training/step1_supervised_finetuning/main.py
+++ b/applications/DeepSpeed-Chat/training/step1_supervised_finetuning/main.py
@@ -4,9 +4,7 @@
 
 # DeepSpeed Team
 import argparse
-import os
 import math
-import sys
 
 import torch
 from torch.utils.data import DataLoader, RandomSampler, SequentialSampler
@@ -23,14 +21,12 @@
 from deepspeed.ops.adam import DeepSpeedCPUAdam, FusedAdam
 from deepspeed import get_accelerator
 
-sys.path.append(
-    os.path.abspath(os.path.join(os.path.dirname(__file__), os.path.pardir)))
-from utils.data.data_utils import create_prompt_dataset
-from utils.utils import print_rank_0, to_device, save_hf_format, set_random_seed, get_all_reduce_mean, get_optimizer_grouped_parameters, save_zero_three_model, load_hf_tokenizer
-from utils.ds_utils import get_train_ds_config
-from utils.module.lora import convert_linear_layer_to_lora, convert_lora_to_linear_layer, only_optimize_lora_parameters, make_model_gradient_checkpointing_compatible
-from utils.model.model_utils import create_hf_model, causal_lm_model_to_fp32_loss
-from utils.perf import print_throughput
+from dschat.utils.data.data_utils import create_prompt_dataset
+from dschat.utils.utils import print_rank_0, to_device, save_hf_format, set_random_seed, get_all_reduce_mean, get_optimizer_grouped_parameters, save_zero_three_model, load_hf_tokenizer
+from dschat.utils.ds_utils import get_train_ds_config
+from dschat.utils.module.lora import convert_linear_layer_to_lora, convert_lora_to_linear_layer, only_optimize_lora_parameters, make_model_gradient_checkpointing_compatible
+from dschat.utils.model.model_utils import create_hf_model, causal_lm_model_to_fp32_loss
+from dschat.utils.perf import print_throughput
 
 
 def parse_args():
diff --git a/applications/DeepSpeed-Chat/training/step1_supervised_finetuning/prompt_eval.py b/applications/DeepSpeed-Chat/training/step1_supervised_finetuning/prompt_eval.py
index 21f79d63a..a25b0edea 100644
--- a/applications/DeepSpeed-Chat/training/step1_supervised_finetuning/prompt_eval.py
+++ b/applications/DeepSpeed-Chat/training/step1_supervised_finetuning/prompt_eval.py
@@ -5,16 +5,12 @@
 import argparse
 import logging
 import torch
-import sys
-import os
 
 from transformers import (
     AutoModelForCausalLM, )
 
-sys.path.append(
-    os.path.abspath(os.path.join(os.path.dirname(__file__), os.path.pardir)))
-from utils.model.model_utils import create_hf_model
-from utils.utils import load_hf_tokenizer
+from dschat.utils.model.model_utils import create_hf_model
+from dschat.utils.utils import load_hf_tokenizer
 from deepspeed import get_accelerator
 
 logger = logging.getLogger(__name__)
diff --git a/applications/DeepSpeed-Chat/training/step2_reward_model_finetuning/main.py b/applications/DeepSpeed-Chat/training/step2_reward_model_finetuning/main.py
index dae906173..265c1caf4 100644
--- a/applications/DeepSpeed-Chat/training/step2_reward_model_finetuning/main.py
+++ b/applications/DeepSpeed-Chat/training/step2_reward_model_finetuning/main.py
@@ -4,9 +4,7 @@
 
 # DeepSpeed Team
 import argparse
-import os
 import math
-import sys
 
 import torch
 from torch.utils.data import DataLoader, RandomSampler, SequentialSampler
@@ -21,13 +19,11 @@
 from deepspeed.ops.adam import DeepSpeedCPUAdam, FusedAdam
 from deepspeed.accelerator import get_accelerator
 
-sys.path.append(
-    os.path.abspath(os.path.join(os.path.dirname(__file__), os.path.pardir)))
-from utils.model.model_utils import create_critic_model
-from utils.data.data_utils import create_prompt_dataset, DataCollatorReward
-from utils.utils import print_rank_0, to_device, save_hf_format, set_random_seed, get_all_reduce_mean, get_optimizer_grouped_parameters, save_zero_three_model, load_hf_tokenizer
-from utils.ds_utils import get_train_ds_config
-from utils.module.lora import convert_linear_layer_to_lora, convert_lora_to_linear_layer, only_optimize_lora_parameters, make_model_gradient_checkpointing_compatible
+from dschat.utils.model.model_utils import create_critic_model
+from dschat.utils.data.data_utils import create_prompt_dataset, DataCollatorReward
+from dschat.utils.utils import print_rank_0, to_device, save_hf_format, set_random_seed, get_all_reduce_mean, get_optimizer_grouped_parameters, save_zero_three_model, load_hf_tokenizer
+from dschat.utils.ds_utils import get_train_ds_config
+from dschat.utils.module.lora import convert_linear_layer_to_lora, convert_lora_to_linear_layer, only_optimize_lora_parameters, make_model_gradient_checkpointing_compatible
 
 
 def parse_args():
diff --git a/applications/DeepSpeed-Chat/training/step2_reward_model_finetuning/rw_eval.py b/applications/DeepSpeed-Chat/training/step2_reward_model_finetuning/rw_eval.py
index 28ee87d1c..23f9a66af 100644
--- a/applications/DeepSpeed-Chat/training/step2_reward_model_finetuning/rw_eval.py
+++ b/applications/DeepSpeed-Chat/training/step2_reward_model_finetuning/rw_eval.py
@@ -4,16 +4,10 @@
 
 # DeepSpeed Team
 import argparse
-import os
 import torch
 
-import sys
-
-sys.path.append(
-    os.path.abspath(os.path.join(os.path.dirname(__file__), os.path.pardir)))
-from utils.model.model_utils import create_critic_model
-from utils.utils import to_device
-from utils.utils import load_hf_tokenizer
+from dschat.utils.model.model_utils import create_critic_model
+from dschat.utils.utils import to_device, load_hf_tokenizer
 from deepspeed import get_accelerator
 
 
diff --git a/applications/DeepSpeed-Chat/training/step3_rlhf_finetuning/main.py b/applications/DeepSpeed-Chat/training/step3_rlhf_finetuning/main.py
index 78f3249c0..a5be5671b 100644
--- a/applications/DeepSpeed-Chat/training/step3_rlhf_finetuning/main.py
+++ b/applications/DeepSpeed-Chat/training/step3_rlhf_finetuning/main.py
@@ -33,18 +33,13 @@
 
 import deepspeed
 
-from ppo_trainer import DeepSpeedPPOTrainer, DeepSpeedPPOTrainerUnsupervised
-from rlhf_engine import DeepSpeedRLHFEngine
-
-import sys
-
-sys.path.append(
-    os.path.abspath(os.path.join(os.path.dirname(__file__), os.path.pardir)))
-from utils.data.data_utils import create_prompt_dataset, MiniDataset, DataCollatorRLHF, get_unsupervised_data
-from utils.utils import print_rank_0, to_device, save_hf_format, set_random_seed, get_all_reduce_mean, \
-    moving_average, save_zero_three_model, load_hf_tokenizer, ExponentialMovingAverage
-from utils.module.lora import convert_lora_to_linear_layer
-from utils.perf import print_throughput_step3
+from dschat.rlhf.ppo_trainer import DeepSpeedPPOTrainer, DeepSpeedPPOTrainerUnsupervised
+from dschat.rlhf.rlhf_engine import DeepSpeedRLHFEngine
+from dschat.utils.data.data_utils import create_prompt_dataset, MiniDataset, DataCollatorRLHF, get_unsupervised_data
+from dschat.utils.utils import print_rank_0, to_device, save_hf_format, set_random_seed, get_all_reduce_mean, moving_average, save_zero_three_model, load_hf_tokenizer, \
+    ExponentialMovingAverage
+from dschat.utils.module.lora import convert_lora_to_linear_layer
+from dschat.utils.perf import print_throughput_step3
 from deepspeed.accelerator import get_accelerator
 
 writer = None