Add qwen2 example (#11252)

* Add GPU example for Qwen2 * Update comments in README * Update README for Qwen2 GPU example * Add CPU example for Qwen2 Sample Output under README pending * Update generate.py and README for CPU Qwen2 * Update GPU example for Qwen2 * Small update * Small fix * Add Qwen2 table * Update README for Qwen2 CPU and GPU Update sample output under README --------- Co-authored-by: Zijie Li <[email protected]>
intel-analytics · Jun 7, 2024 · 8c36b5b · 8c36b5b
1 parent 85df5e7
commit 8c36b5b
Show file tree

Hide file tree

Showing 10 changed files with 788 additions and 0 deletions.
diff --git a/README.md b/README.md
@@ -169,6 +169,7 @@ Over 50 models have been optimized/verified on `ipex-llm`, including *LLaMA/LLaM
 | InternLM   | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/internlm)  | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/internlm)   |
 | Qwen       | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/qwen)      | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/qwen)       |
 | Qwen1.5 | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/qwen1.5) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/qwen1.5) |
+| Qwen2 | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/qwen2) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/qwe2) |
 | Qwen-VL    | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/qwen-vl)   | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/qwen-vl)    |
 | Aquila     | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/aquila)    | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/aquila)     |
 | Aquila2     | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/aquila2)    | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/aquila2)     |

diff --git a/docs/readthedocs/source/index.rst b/docs/readthedocs/source/index.rst
@@ -363,6 +363,13 @@ Verified Models
          <td>
            <a href="https://github.com/intel-analytics/ipex-llm/tree/main/python/llm/example/GPU/HF-Transformers-AutoModels/Model/qwen1.5">link</a></td>
        </tr>
+       <tr>
+         <td>Qwen2</td>
+         <td>
+           <a href="https://github.com/intel-analytics/ipex-llm/tree/main/python/llm/example/CPU/HF-Transformers-AutoModels/Model/qwen2">link</a></td>
+         <td>
+           <a href="https://github.com/intel-analytics/ipex-llm/tree/main/python/llm/example/GPU/HF-Transformers-AutoModels/Model/qwen2">link</a></td>
+       </tr>
        <tr>
          <td>Qwen-VL</td>
          <td>

diff --git a/python/llm/example/CPU/HF-Transformers-AutoModels/Model/qwen2/README.md b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/qwen2/README.md
@@ -0,0 +1,83 @@
+# Qwen2
+
+In this directory, you will find examples on how you could apply IPEX-LLM INT4 optimizations on Qwen2 models. For illustration purposes, we utilize the [Qwen/Qwen2-7B-Instruct](https://huggingface.co/Qwen/Qwen2-7B-Instruct) as a reference Qwen2 model.
+
+## 0. Requirements
+To run these examples with IPEX-LLM, we have some recommended requirements for your machine, please refer to [here](../README.md#recommended-requirements) for more information.
+
+## Example: Predict Tokens using `generate()` API
+In the example [generate.py](./generate.py), we show a basic use case for a Qwen model to predict the next N tokens using `generate()` API, with IPEX-LLM INT4 optimizations.
+### 1. Install
+We suggest using conda to manage environment:
+
+On Linux:
+
+```bash
+conda create -n llm python=3.11
+conda activate llm
+
+# install ipex-llm with 'all' option
+pip install --pre --upgrade ipex-llm[all] --extra-index-url https://download.pytorch.org/whl/cpu
+pip install transformers==4.37.0 # install the transformers which support Qwen2
+```
+
+On Windows:
+
+```cmd
+conda create -n llm python=3.11
+conda activate llm
+
+pip install --pre --upgrade ipex-llm[all]
+pip install transformers==4.37.0
+```
+
+### 2. Run
+```
+python ./generate.py --repo-id-or-model-path REPO_ID_OR_MODEL_PATH --prompt PROMPT --n-predict N_PREDICT
+```
+
+Arguments info:
+- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the huggingface repo id for the Qwen model to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'Qwen/Qwen2-7B-Instruct'`.
+- `--prompt PROMPT`: argument defining the prompt to be infered (with integrated prompt format for chat). It is default to be `'AI是什么?'`.
+- `--n-predict N_PREDICT`: argument defining the max number of tokens to predict. It is default to be `32`.
+
+> **Note**: When loading the model in 4-bit, IPEX-LLM converts linear layers in the model into INT4 format. In theory, a *X*B model saved in 16-bit will requires approximately 2*X* GB of memory for loading, and ~0.5*X* GB memory for further inference.
+>
+> Please select the appropriate size of the Qwen model based on the capabilities of your machine.
+
+#### 2.1 Client
+On client Windows machine, it is recommended to run directly with full utilization of all cores:
+```cmd
+python ./generate.py 
+```
+
+#### 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 IPEX-LLM env variables
+source ipex-llm-init
+
+# e.g. for a server with 48 cores per socket
+export OMP_NUM_THREADS=48
+numactl -C 0-47 -m 0 python ./generate.py
+```
+
+#### 2.3 Sample Output
+##### [Qwen/Qwen2-7B-Instruct](https://huggingface.co/Qwen/Qwen2-7B-Instruct)
+```log
+Inference time: xxxx s
+-------------------- Prompt --------------------
+AI是什么？
+-------------------- Output --------------------
+AI，即人工智能（Artificial Intelligence），是一种计算机科学领域，旨在开发能够模拟、延伸和增强人类智能的算法和系统。人工 智能涉及许多
+```
+
+```log
+Inference time: xxxx s
+-------------------- Prompt --------------------
+What is AI?
+-------------------- Output --------------------
+AI stands for Artificial Intelligence, which is the simulation of human intelligence in machines that are programmed to think and learn like humans and mimic their actions. The term may
+```
diff --git a/python/llm/example/CPU/HF-Transformers-AutoModels/Model/qwen2/generate.py b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/qwen2/generate.py
@@ -0,0 +1,80 @@
+#
+# 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
+import numpy as np
+
+from transformers import AutoTokenizer
+
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser(description='Qwen2-7B-Instruct')
+    parser.add_argument('--repo-id-or-model-path', type=str, default="Qwen/Qwen2-7B-Instruct",
+                        help='The huggingface repo id for the Qwen2 model to be downloaded'
+                             ', or the path to the huggingface checkpoint folder')
+    parser.add_argument('--prompt', type=str, default="AI是什么？",
+                        help='Prompt to infer') 
+    parser.add_argument('--n-predict', type=int, default=32,
+                        help='Max tokens to predict')
+
+    args = parser.parse_args()
+    model_path = args.repo_id_or_model_path
+
+
+    from ipex_llm.transformers import AutoModelForCausalLM
+    model = AutoModelForCausalLM.from_pretrained(model_path,
+                                                 load_in_4bit=True,
+                                                 optimize_model=True,
+                                                 trust_remote_code=True,
+                                                 use_cache=True)
+
+    # Load tokenizer
+    tokenizer = AutoTokenizer.from_pretrained(model_path,
+                                              trust_remote_code=True)
+
+    prompt = args.prompt
+
+    # Generate predicted tokens
+    with torch.inference_mode():
+        # The following code for generation is adapted from https://huggingface.co/Qwen/Qwen2-7B-Instruct#quickstart
+        messages = [
+            {"role": "system", "content": "You are a helpful assistant."},
+            {"role": "user", "content": prompt}
+            ]
+        text = tokenizer.apply_chat_template(
+            messages,
+            tokenize=False,
+            add_generation_prompt=True
+            )
+        model_inputs = tokenizer([text], return_tensors="pt")
+        st = time.time()
+        generated_ids = model.generate(
+            model_inputs.input_ids,
+            max_new_tokens=args.n_predict
+            )
+        end = time.time()
+        generated_ids = [
+            output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
+            ]
+
+        response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
+        print(f'Inference time: {end-st} s')
+        print('-'*20, 'Prompt', '-'*20)
+        print(prompt)
+        print('-'*20, 'Output', '-'*20)
+        print(response)
diff --git a/python/llm/example/CPU/PyTorch-Models/Model/qwen2/README.md b/python/llm/example/CPU/PyTorch-Models/Model/qwen2/README.md
@@ -0,0 +1,84 @@
+# Qwen2
+In this directory, you will find examples on how you could use IPEX-LLM `optimize_model` API to accelerate Qwen2 models. For illustration purposes, we utilize the [Qwen/Qwen2-7B-Instruct](https://huggingface.co/Qwen/Qwen2-7B-Instruct) as reference Qwen2 model.
+
+## Requirements
+To run these examples with IPEX-LLM, we have some recommended requirements for your machine, please refer to [here](../README.md#recommended-requirements) for more information.
+
+## Example: Predict Tokens using `generate()` API
+In the example [generate.py](./generate.py), we show a basic use case for a Qwen2 model to predict the next N tokens using `generate()` API, with IPEX-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://conda-forge.org/download/).
+
+After installing conda, create a Python environment for IPEX-LLM:
+
+On Linux:
+
+```bash
+conda create -n llm python=3.11 # recommend to use Python 3.11
+conda activate llm
+
+# install the latest ipex-llm nightly build with 'all' option
+pip install --pre --upgrade ipex-llm[all] --extra-index-url https://download.pytorch.org/whl/cpu
+pip install transformers==4.37.0 # install transformers which supports Qwen2
+```
+
+On Windows:
+
+```cmd
+conda create -n llm python=3.11
+conda activate llm
+
+pip install --pre --upgrade ipex-llm[all]
+pip install transformers==4.37.0
+```
+
+### 2. Run
+```
+python ./generate.py --repo-id-or-model-path REPO_ID_OR_MODEL_PATH --prompt PROMPT --n-predict N_PREDICT
+```
+
+Arguments info:
+- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the huggingface repo id for the Qwen2 to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'Qwen/Qwen2-7B-Instruct'`.
+- `--prompt PROMPT`: argument defining the prompt to be infered (with integrated prompt format for chat). It is default to be `'AI是什么？'`.
+- `--n-predict N_PREDICT`: argument defining the max number of tokens to predict. It is default to be `32`.
+
+> **Note**: When loading the model in 4-bit, IPEX-LLM converts linear layers in the model into INT4 format. In theory, a *X*B model saved in 16-bit will requires approximately 2*X* GB of memory for loading, and ~0.5*X* GB memory for further inference.
+>
+> Please select the appropriate size of the Qwen model based on the capabilities of your machine.
+
+#### 2.1 Client
+On client Windows machines, it is recommended to run directly with full utilization of all cores:
+```cmd
+python ./generate.py --prompt 'What is AI?'
+```
+
+#### 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 IPEX-LLM env variables
+source ipex-llm-init
+
+# e.g. for a server with 48 cores per socket
+export OMP_NUM_THREADS=48
+numactl -C 0-47 -m 0 python ./generate.py --prompt 'What is AI?'
+```
+
+#### 2.3 Sample Output
+##### [Qwen/Qwen2-7B-Instruct](https://huggingface.co/Qwen/Qwen2-7B-Instruct)
+```log
+Inference time: xxxx s
+-------------------- Prompt --------------------
+AI是什么？
+-------------------- Output --------------------
+AI，即人工智能（Artificial Intelligence），是一门研究、开发用于模拟、延伸和扩展人的智能的理论、方法、技术及应用系统的学科
+```
+
+```log
+Inference time: xxxx s
+-------------------- Prompt --------------------
+What is AI?
+-------------------- Output --------------------
+AI stands for Artificial Intelligence, which refers to the development of computer systems that can perform tasks that typically require human intelligence. These tasks may include learning from experience,
+```
diff --git a/python/llm/example/CPU/PyTorch-Models/Model/qwen2/generate.py b/python/llm/example/CPU/PyTorch-Models/Model/qwen2/generate.py
@@ -0,0 +1,82 @@
+#
+# 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
+import numpy as np
+
+from transformers import AutoTokenizer
+
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser(description='Qwen2-7B-Instruct')
+    parser.add_argument('--repo-id-or-model-path', type=str, default="Qwen/Qwen2-7B-Instruct",
+                        help='The huggingface repo id for the Qwen2 model to be downloaded'
+                             ', or the path to the huggingface checkpoint folder')
+    parser.add_argument('--prompt', type=str, default="AI是什么？",
+                        help='Prompt to infer') 
+    parser.add_argument('--n-predict', type=int, default=32,
+                        help='Max tokens to predict')
+
+    args = parser.parse_args()
+    model_path = args.repo_id_or_model_path
+
+
+    from transformers import AutoModelForCausalLM
+    model = AutoModelForCausalLM.from_pretrained(model_path,
+                                                 trust_remote_code=True,
+                                                 torch_dtype='auto',
+                                                 low_cpu_mem_usage=True,
+                                                 use_cache=True)
+
+
+    # Load tokenizer
+    tokenizer = AutoTokenizer.from_pretrained(model_path,
+                                              trust_remote_code=True)
+    from ipex_llm import optimize_model
+    model = optimize_model(model)
+
+    prompt = args.prompt
+    # Generate predicted tokens
+    with torch.inference_mode():
+        # The following code for generation is adapted from https://huggingface.co/Qwen/Qwen2-7B-Instruct#quickstart
+        messages = [
+            {"role": "system", "content": "You are a helpful assistant."},
+            {"role": "user", "content": prompt}
+            ]
+        text = tokenizer.apply_chat_template(
+            messages,
+            tokenize=False,
+            add_generation_prompt=True
+            )
+        model_inputs = tokenizer([text], return_tensors="pt")
+        st = time.time()
+        generated_ids = model.generate(
+            model_inputs.input_ids,
+            max_new_tokens=args.n_predict
+            )
+        end = time.time()
+        generated_ids = [
+            output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
+            ]
+
+        response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
+        print(f'Inference time: {end-st} s')
+        print('-'*20, 'Prompt', '-'*20)
+        print(prompt)
+        print('-'*20, 'Output', '-'*20)
+        print(response)