Add phi-3-vision example (#11156)

* Add phi-3-vision example (HF-Automodels) * fix * fix * fix * Add phi-3-vision CPU example (HF-Automodels) * add in readme * fix * fix * fix * fix * use fp8 for gpu example * remove eval
intel-analytics · May 30, 2024 · dcbf4d3 · dcbf4d3
1 parent 93146b9
commit dcbf4d3
Show file tree

Hide file tree

Showing 6 changed files with 427 additions and 0 deletions.
diff --git a/README.md b/README.md
@@ -198,6 +198,7 @@ Over 50 models have been optimized/verified on `ipex-llm`, including *LLaMA/LLaM
 | Ziya-Coding-34B-v1.0 | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/ziya) | |
 | Phi-2 | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/phi-2) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/phi-2) |
 | Phi-3 | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/phi-3) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/phi-3) |
+| Phi-3-vision | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/phi-3-vision) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/phi-3-vision) |
 | Yuan2 | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/yuan2) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/yuan2) |
 | Gemma | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/gemma) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/gemma) |
 | DeciLM-7B | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/deciLM-7b) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/deciLM-7b) |

diff --git a/docs/readthedocs/source/index.rst b/docs/readthedocs/source/index.rst
@@ -555,6 +555,13 @@ Verified Models
          <td>
            <a href="https://github.com/intel-analytics/ipex-llm/tree/main/python/llm/example/GPU/HF-Transformers-AutoModels/Model/phi-3">link</a></td>
        </tr>
+       <tr>
+         <td>Phi-3-vision</td>
+         <td>
+           <a href="https://github.com/intel-analytics/ipex-llm/tree/main/python/llm/example/CPU/HF-Transformers-AutoModels/Model/phi-3-vision">link</a></td>
+         <td>
+           <a href="https://github.com/intel-analytics/ipex-llm/tree/main/python/llm/example/GPU/HF-Transformers-AutoModels/Model/phi-3-vision">link</a></td>
+       </tr>
        <tr>
          <td>Yuan2</td>
          <td>

diff --git a/python/llm/example/CPU/HF-Transformers-AutoModels/Model/phi-3-vision/README.md b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/phi-3-vision/README.md
@@ -0,0 +1,90 @@
+# phi-3-vision
+
+In this directory, you will find examples on how you could apply IPEX-LLM INT8 optimizations on phi-3-vision models. For illustration purposes, we utilize the [microsoft/Phi-3-vision-128k-instruct](https://huggingface.co/microsoft/Phi-3-vision-128k-instruct) as a reference phi-3-vision 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 phi-3-vision model to predict the next N tokens using `generate()` API, with IPEX-LLM INT8 optimizations.
+### 1. Install
+We suggest using conda to manage environment:
+
+On Linux:
+
+```bash
+conda create -n llm python=3.11 # recommend to use 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 pillow torchvision
+pip install transformers==4.37.0
+```
+
+On Windows:
+
+```cmd
+conda create -n llm python=3.11
+conda activate llm
+
+pip install --pre --upgrade ipex-llm[all]
+
+pip install pillow torchvision
+pip install transformers==4.37.0
+```
+
+### 2. Run
+```
+python ./generate.py --repo-id-or-model-path REPO_ID_OR_MODEL_PATH --image-url-or-path IMAGE_URL_OR_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 phi-3-vision model (e.g. `microsoft/Phi-3-vision-128k-instruct`) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'microsoft/Phi-3-vision-128k-instruct'`.
+- `--image-url-or-path IMAGE_URL_OR_PATH`: argument defining the image to be infered. It is default to be `'http://farm6.staticflickr.com/5268/5602445367_3504763978_z.jpg'`.
+- `--prompt PROMPT`: argument defining the prompt to be infered (with integrated prompt format for chat). It is default to be `'What is in the image?'`.
+- `--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 8-bit, IPEX-LLM converts linear layers in the model into INT8 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 phi-3-vision 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
+```
+
+#### 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
+#### [microsoft/Phi-3-vision-128k-instruct](https://huggingface.co/microsoft/Phi-3-vision-128k-instruct)
+
+```log
+Inference time: xxxx s
+-------------------- Prompt --------------------
+Message: [{'role': 'user', 'content': '<|image_1|>\nWhat is in the image?'}]
+Image link/path: http://farm6.staticflickr.com/5268/5602445367_3504763978_z.jpg
+-------------------- Output --------------------
+
+
+What is in the image?
+ The image shows a child holding a white teddy bear dressed in a pink dress.
+```
+
+The sample input image is (which is fetched from [COCO dataset](https://cocodataset.org/#explore?id=264959)):
+
+<a href="http://farm6.staticflickr.com/5268/5602445367_3504763978_z.jpg"><img width=400px src="http://farm6.staticflickr.com/5268/5602445367_3504763978_z.jpg" ></a>
diff --git a/python/llm/example/CPU/HF-Transformers-AutoModels/Model/phi-3-vision/generate.py b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/phi-3-vision/generate.py
@@ -0,0 +1,88 @@
+#
+# 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 os
+import time
+import torch
+import argparse
+import requests
+
+from PIL import Image
+from ipex_llm.transformers import AutoModelForCausalLM
+from transformers import AutoProcessor
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for phi-3 model')
+    parser.add_argument('--repo-id-or-model-path', type=str, default="microsoft/Phi-3-vision-128k-instruct",
+                        help='The huggingface repo id for the phi-3-vision model to be downloaded'
+                             ', or the path to the huggingface checkpoint folder')
+    parser.add_argument('--image-url-or-path', type=str,
+                        default="http://farm6.staticflickr.com/5268/5602445367_3504763978_z.jpg",
+                        help='The URL or path to the image to infer')
+    parser.add_argument('--prompt', type=str, default="What is in the image?",
+                        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
+    image_path = args.image_url_or_path
+
+    # Load model in INT8,
+    # which convert the relevant layers in the model into INT8 format
+    # We here use INT8 instead of INT4 for better output
+    # `_attn_implementation="eager"` is required for phi-3-vision
+    # `modules_to_not_convert=["vision_embed_tokens"]` is for acceleration and is optional
+    model = AutoModelForCausalLM.from_pretrained(model_path,
+                                                 trust_remote_code=True,
+                                                 load_in_low_bit="sym_int8",
+                                                 _attn_implementation="eager",
+                                                 modules_to_not_convert=["vision_embed_tokens"])
+
+    # Load processor
+    processor = AutoProcessor.from_pretrained(model_path, trust_remote_code=True)
+
+    # here the message formatting refers to https://huggingface.co/microsoft/Phi-3-vision-128k-instruct#sample-inference-code
+    messages = [
+        {"role": "user", "content": "<|image_1|>\n{prompt}".format(prompt=args.prompt)},
+    ]
+    prompt = processor.tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
+
+    if os.path.exists(image_path):
+       image = Image.open(image_path)
+    else:
+       image = Image.open(requests.get(image_path, stream=True).raw)
+
+    # Generate predicted tokens
+    with torch.inference_mode():
+        inputs = processor(prompt, [image], return_tensors="pt")
+        st = time.time()
+        output = model.generate(**inputs,
+                                eos_token_id=processor.tokenizer.eos_token_id,
+                                num_beams=1,
+                                do_sample=False,
+                                max_new_tokens=args.n_predict,
+                                temperature=0.0)
+        end = time.time()
+        print(f'Inference time: {end-st} s')
+        output_str = processor.decode(output[0],
+                                      skip_special_tokens=True,
+                                      clean_up_tokenization_spaces=False)
+        print('-'*20, 'Prompt', '-'*20)
+        print(f'Message: {messages}')
+        print(f'Image link/path: {image_path}')
+        print('-'*20, 'Output', '-'*20)
+        print(output_str)
diff --git a/python/llm/example/GPU/HF-Transformers-AutoModels/Model/phi-3-vision/README.md b/python/llm/example/GPU/HF-Transformers-AutoModels/Model/phi-3-vision/README.md
@@ -0,0 +1,136 @@
+# phi-3-vision
+In this directory, you will find examples on how you could apply IPEX-LLM FP8 optimizations on phi-3-vision models on [Intel GPUs](../../../README.md). For illustration purposes, we utilize the [microsoft/Phi-3-vision-128k-instruct](https://huggingface.co/microsoft/Phi-3-vision-128k-instruct) as a reference phi-3-vision model.
+
+## 0. Requirements
+To run these examples with IPEX-LLM on Intel GPUs, we have some recommended requirements for your machine, please refer to [here](../../../README.md#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 phi-3-vision model to predict the next N tokens using `generate()` API, with IPEX-LLM FP8 optimizations on Intel GPUs.
+### 1. Install
+#### 1.1 Installation on Linux
+We suggest using conda to manage environment:
+```bash
+conda create -n llm python=3.11
+conda activate llm
+# below command will install intel_extension_for_pytorch==2.1.10+xpu as default
+pip install --pre --upgrade ipex-llm[xpu] --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
+
+pip install transformers==4.37.0
+```
+
+#### 1.2 Installation on Windows
+We suggest using conda to manage environment:
+```bash
+conda create -n llm python=3.11 libuv
+conda activate llm
+
+# below command will install intel_extension_for_pytorch==2.1.10+xpu as default
+pip install --pre --upgrade ipex-llm[xpu] --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
+
+pip install transformers==4.37.0
+```
+
+### 2. Configures OneAPI environment variables for Linux
+
+> [!NOTE]
+> Skip this step if you are running on Windows.
+
+This is a required step on Linux for APT or offline installed oneAPI. Skip this step for PIP-installed oneAPI.
+
+```bash
+source /opt/intel/oneapi/setvars.sh
+```
+
+### 3. Runtime Configurations
+For optimal performance, it is recommended to set several environment variables. Please check out the suggestions based on your device.
+#### 3.1 Configurations for Linux
+<details>
+
+<summary>For Intel Arc™ A-Series Graphics and Intel Data Center GPU Flex Series</summary>
+
+```bash
+export USE_XETLA=OFF
+export SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1
+export SYCL_CACHE_PERSISTENT=1
+```
+
+</details>
+
+<details>
+
+<summary>For Intel Data Center GPU Max Series</summary>
+
+```bash
+export LD_PRELOAD=${LD_PRELOAD}:${CONDA_PREFIX}/lib/libtcmalloc.so
+export SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1
+export SYCL_CACHE_PERSISTENT=1
+export ENABLE_SDP_FUSION=1
+```
+> Note: Please note that `libtcmalloc.so` can be installed by `conda install -c conda-forge -y gperftools=2.10`.
+</details>
+
+<details>
+
+<summary>For Intel iGPU</summary>
+
+```bash
+export SYCL_CACHE_PERSISTENT=1
+export BIGDL_LLM_XMX_DISABLED=1
+```
+
+</details>
+
+#### 3.2 Configurations for Windows
+<details>
+
+<summary>For Intel iGPU</summary>
+
+```cmd
+set SYCL_CACHE_PERSISTENT=1
+set BIGDL_LLM_XMX_DISABLED=1
+```
+
+</details>
+
+<details>
+
+<summary>For Intel Arc™ A-Series Graphics</summary>
+
+```cmd
+set SYCL_CACHE_PERSISTENT=1
+```
+
+</details>
+
+> [!NOTE]
+> For the first time that each model runs on Intel iGPU/Intel Arc™ A300-Series or Pro A60, it may take several minutes to compile.
+### 4. Running examples
+
+```
+python ./generate.py --prompt 'What is in the image?'
+```
+
+Arguments info:
+- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the huggingface repo id for the phi-3-vision model (e.g. `microsoft/Phi-3-vision-128k-instruct`) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'microsoft/Phi-3-vision-128k-instruct'`.
+- `--image-url-or-path IMAGE_URL_OR_PATH`: argument defining the image to be infered. It is default to be `'http://farm6.staticflickr.com/5268/5602445367_3504763978_z.jpg'`.
+- `--prompt PROMPT`: argument defining the prompt to be infered (with integrated prompt format for chat). It is default to be `'What is in the image?'`.
+- `--n-predict N_PREDICT`: argument defining the max number of tokens to predict. It is default to be `32`.
+
+#### Sample Output
+#### [microsoft/Phi-3-vision-128k-instruct](https://huggingface.co/microsoft/Phi-3-vision-128k-instruct)
+
+```log
+Inference time: xxxx s
+-------------------- Prompt --------------------
+Message: [{'role': 'user', 'content': '<|image_1|>\nWhat is in the image?'}]
+Image link/path: http://farm6.staticflickr.com/5268/5602445367_3504763978_z.jpg
+-------------------- Output --------------------
+
+
+What is in the image?
+ The image shows a young girl holding a white teddy bear. She is wearing a pink dress with a heart on it. The background includes a stone
+```
+
+The sample input image is (which is fetched from [COCO dataset](https://cocodataset.org/#explore?id=264959)):
+
+<a href="http://farm6.staticflickr.com/5268/5602445367_3504763978_z.jpg"><img width=400px src="http://farm6.staticflickr.com/5268/5602445367_3504763978_z.jpg" ></a>