From c1d25a51a82f39af01422ac08ba9140f306b5d48 Mon Sep 17 00:00:00 2001
From: binbin Deng <108676127+plusbang@users.noreply.github.com>
Date: Mon, 18 Sep 2023 16:18:35 +0800
Subject: [PATCH] LLM: add `optimize_model` example for bert (#8975)

---
 python/llm/example/pytorch-models/README.md   |  1 +
 .../llm/example/pytorch-models/bert/README.md | 48 ++++++++++++++++
 .../pytorch-models/bert/extract_feature.py    | 55 +++++++++++++++++++
 3 files changed, 104 insertions(+)
 create mode 100644 python/llm/example/pytorch-models/bert/README.md
 create mode 100644 python/llm/example/pytorch-models/bert/extract_feature.py

diff --git a/python/llm/example/pytorch-models/README.md b/python/llm/example/pytorch-models/README.md
index ad60d414..ddbb5c21 100644
--- a/python/llm/example/pytorch-models/README.md
+++ b/python/llm/example/pytorch-models/README.md
@@ -7,6 +7,7 @@ You can use `optimize_model` API to accelerate general PyTorch models on Intel s
 | LLaMA 2   | [link](llama2)    |
 | ChatGLM   | [link](chatglm)   | 
 | Openai Whisper | [link](openai-whisper)   | 
+| BERT | [link](bert)   | 
 
 ## Recommended Requirements
 To run the examples, we recommend using Intel® Xeon® processors (server), or >= 12th Gen Intel® Core™ processor (client).
diff --git a/python/llm/example/pytorch-models/bert/README.md b/python/llm/example/pytorch-models/bert/README.md
new file mode 100644
index 00000000..522d1cb8
--- /dev/null
+++ b/python/llm/example/pytorch-models/bert/README.md
@@ -0,0 +1,48 @@
+# BERT
+In this directory, you will find examples on how you could use BigDL-LLM `optimize_model` API to accelerate BERT models. For illustration purposes, we utilize the [bert-large-uncased](https://huggingface.co/bert-large-uncased) as reference BERT models.
+
+## Requirements
+To run these examples with BigDL-LLM, we have some recommended requirements for your machine, please refer to [here](../README.md#recommended-requirements) for more information.
+
+## Example: Extract the feature of given text
+In the example [extract_feature.py](./extract_feature.py), we show a basic use case for a BERT model to extract the feature of given text, with BigDL-LLM INT4 optimizations.
+### 1. Install
+We suggest using conda to manage the Python environment. For more information about conda installation, please refer to [here](https://docs.conda.io/en/latest/miniconda.html#).
+
+After installing conda, create a Python environment for BigDL-LLM:
+```bash
+conda create -n llm python=3.9 # recommend to use Python 3.9
+conda activate llm
+
+pip install --pre --upgrade bigdl-llm[all] # install the latest bigdl-llm nightly build with 'all' option
+```
+
+### 2. Run
+After setting up the Python environment, you could run the example by following steps.
+
+#### 2.1 Client
+On client Windows machines, it is recommended to run directly with full utilization of all cores:
+```powershell
+python ./extract_feature.py --text 'This is an example text for feature extraction.'
+```
+More information about arguments can be found in [Arguments Info](#23-arguments-info) section.
+
+#### 2.2 Server
+For optimal performance on server, it is recommended to set several environment variables (refer to [here](../README.md#best-known-configuration-on-linux) for more information), and run the example with all the physical cores of a single socket.
+
+E.g. on Linux,
+```bash
+# set BigDL-Nano env variables
+source bigdl-nano-init
+
+# e.g. for a server with 48 cores per socket
+export OMP_NUM_THREADS=48
+numactl -C 0-47 -m 0 python ./extract_feature.py --text 'This is an example text for feature extraction.'
+```
+More information about arguments can be found in [Arguments Info](#23-arguments-info) section.
+
+#### 2.3 Arguments Info
+In the example, several arguments can be passed to satisfy your requirements:
+
+- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the huggingface repo id for the BERT model (e.g. `bert-large-uncased`) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'bert-large-uncased'`.
+- `--text TEXT`: argument defining the text to be extracted features. It is default to be `'This is an example text for feature extraction.'`.
diff --git a/python/llm/example/pytorch-models/bert/extract_feature.py b/python/llm/example/pytorch-models/bert/extract_feature.py
new file mode 100644
index 00000000..aab47ff7
--- /dev/null
+++ b/python/llm/example/pytorch-models/bert/extract_feature.py
@@ -0,0 +1,55 @@
+#
+# Copyright 2016 The BigDL Authors.
+#
+# 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 torch
+import time
+import argparse
+
+from transformers import BertTokenizer, BertModel
+from bigdl.llm import optimize_model
+
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser(description='Extract the feature of given text using BERT model')
+    parser.add_argument('--repo-id-or-model-path', type=str, default="bert-large-uncased",
+                        help='The huggingface repo id for the BERT (e.g. `bert-large-uncased`) to be downloaded'
+                             ', or the path to the huggingface checkpoint folder')
+    parser.add_argument('--text', type=str, default="This is an example text for feature extraction.",
+                        help='Text to extract features')
+
+    args = parser.parse_args()
+    model_path = args.repo_id_or_model_path
+
+    # Load model
+    model = BertModel.from_pretrained(model_path,
+                                      torch_dtype="auto",
+                                      low_cpu_mem_usage=True)
+    
+    # With only one line to enable BigDL-LLM optimization on model
+    model = optimize_model(model)
+
+    # Load tokenizer
+    tokenizer = BertTokenizer.from_pretrained(model_path)
+    
+    # Extract the feature of given text
+    text = args.text
+    encoded_input = tokenizer(text, return_tensors='pt')
+    st = time.time()
+    output = model(**encoded_input)
+    end = time.time()
+    print(f'Time cost: {end-st} s')
+    print('-'*20, 'Output', '-'*20)
+    print(output)