Add CPU examples of fuyu (#9393)

* add fuyu cpu examples

* add gpu example

* add comments

* add license

* remove gpu example

* fix inference time

README.md

@@ -163,6 +163,7 @@ Over 20 models have been optimized/verified on `bigdl-llm`, including *LLaMA/LLa
 | InternLM-XComposer  | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/internlm-xcomposer)   |    |
 | WizardCoder-Python | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/wizardcoder-python) | |
 | CodeShell | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/CodeShell) | |
+| Fuyu      | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/fuyu) | |
 
 
 ***For more details, please refer to the `bigdl-llm` [Document](https://test-bigdl-llm.readthedocs.io/en/main/doc/LLM/index.html), [Readme](python/llm), [Tutorial](https://github.com/intel-analytics/bigdl-llm-tutorial) and [API Doc](https://bigdl.readthedocs.io/en/latest/doc/PythonAPI/LLM/index.html).***

python/llm/README.md

@@ -70,6 +70,7 @@ Over 20 models have been optimized/verified on `bigdl-llm`, including *LLaMA/LLa
 | InternLM-XComposer    | [link](example/CPU/HF-Transformers-AutoModels/Model/internlm-xcomposer)   |   |
 | WizardCoder-Python | [link](example/CPU/HF-Transformers-AutoModels/Model/wizardcoder-python) | |
 | CodeShell | [link](example/CPU/HF-Transformers-AutoModels/Model/CodeShell) | |
+| Fuyu      | [link](example/CPU/HF-Transformers-AutoModels/Model/fuyu) | |
 
 ### Working with `bigdl-llm`
 

python/llm/example/CPU/HF-Transformers-AutoModels/Model/fuyu/README.md

@@ -0,0 +1,74 @@
+# Fuyu
+In this directory, you will find examples on how you could apply BigDL-LLM INT4 optimizations on Fuyu models. For illustration purposes, we utilize the [adept/fuyu-8b](https://huggingface.co/adept/fuyu-8b) as a reference Fuyu model.
+
+## 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: Predict Tokens using `generate()` API
+In the example [generate.py](./generate.py), we show a basic use case for an Fuyu model to predict the next N tokens using `generate()` API, 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
+
+pip install transformers==4.35 pillow # additional package required for Fuyu to conduct generation
+```
+
+### 2. Run
+After setting up the Python environment, you could run the example by following steps.
+
+> **Note**: When loading the model in 4-bit, BigDL-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 Fuyu 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:
+```powershell
+python ./generate.py --image-path demo.jpg
+```
+More information about arguments can be found in [Arguments Info](#23-arguments-info) section. The expected output can be found in [Sample Output](#24-sample-output) 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 ./generate.py --image-path demo.jpg
+```
+More information about arguments can be found in [Arguments Info](#23-arguments-info) section. The expected output can be found in [Sample Output](#24-sample-output) 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 Fuyu model (e.g. `adept/fuyu-8b`) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'adept/fuyu-8b'`.
+- `--prompt PROMPT`: argument defining the prompt to be inferred (with the image for chat). It is default to be `'Generate a coco-style caption.'`.
+- `--image-path IMAGE_PATH`: argument defining the input image that the chat will focus on. It is required and should be a local path (not url).
+- `--n-predict N_PREDICT`: argument defining the max number of tokens to predict. It is default to be `512`.
+
+
+#### 2.4 Sample Output
+#### [adept/fuyu-8b](https://huggingface.co/adept/fuyu-8b)
+
+```log
+Inference time: xxxx s
+-------------------- Prompt --------------------
+Generate a coco-style caption.
+-------------------- Output --------------------
+An orange bus parked on the side of a road.
+```
+
+The sample input image is (which is fetched from [COCO dataset](https://cocodataset.org/#explore?id=178242)):
+
+[demo.jpg](https://cocodataset.org/#explore?id=178242)
+
+<a href="http://farm6.staticflickr.com/5331/8954873157_539393fece_z.jpg"><img width=400px src="http://farm6.staticflickr.com/5331/8954873157_539393fece_z.jpg" ></a>

python/llm/example/CPU/HF-Transformers-AutoModels/Model/fuyu/generate.py

@@ -0,0 +1,66 @@
+#
+# 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.
+#
+
+from transformers import FuyuProcessor
+import torch
+import argparse
+import time
+from PIL import Image
+from bigdl.llm.transformers import AutoModelForCausalLM
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for Fuyu model')
+    parser.add_argument('--repo-id-or-model-path', type=str, default="adept/fuyu-8b",
+                        help='The huggingface repo id for the Fuyu model to be downloaded'
+                             ', or the path to the huggingface checkpoint folder')
+    parser.add_argument('--prompt', type=str, default="Generate a coco-style caption.",
+                        help='Prompt to infer')
+    parser.add_argument('--image-path', type=str, required=True,
+                        help='Image path for the input image that the chat will focus on')
+    parser.add_argument('--n-predict', type=int, default=512, help='Max tokens to predict')
+
+    args = parser.parse_args()
+    model_path = args.repo_id_or_model_path
+    prompt = args.prompt
+    image = Image.open(args.image_path)
+
+    # Load model
+    # For successful BigDL-LLM optimization on Fuyu, skip the 'vision_embed_tokens' module during optimization
+    model = AutoModelForCausalLM.from_pretrained(model_path, device_map='cpu',
+                                                 load_in_4bit = True,
+                                                 trust_remote_code=True,
+                                                 modules_to_not_convert=['vision_embed_tokens'])
+
+    # Load processor
+    processor = FuyuProcessor.from_pretrained(model_path)
+
+    # Generate predicted tokens
+    with torch.inference_mode():
+        inputs = processor(text=prompt, images=image, return_tensors="pt")
+        st = time.time()
+        generation_outputs = model.generate(**inputs,
+                                max_new_tokens=args.n_predict)
+        end = time.time()
+        outputs = processor.batch_decode(generation_outputs[:, -args.n_predict:], skip_special_tokens=True)
+        print(f'Inference time: {end-st} s')
+        print('-'*20, 'Prompt', '-'*20)
+        print(prompt)
+        print('-'*20, 'Output', '-'*20)
+        for output in outputs:
+            # '\x04' is the "beginning of answer" token
+            # See https://huggingface.co/adept/fuyu-8b#how-to-use
+            answer = output.split('\x04 ', 1)[1] if '\x04' in output else ''
+            print(answer)

python/llm/example/CPU/PyTorch-Models/Model/fuyu/README.md

@@ -0,0 +1,74 @@
+# Fuyu
+In this directory, you will find examples on how you could apply BigDL-LLM INT4 optimizations on Fuyu models. For illustration purposes, we utilize the [adept/fuyu-8b](https://huggingface.co/adept/fuyu-8b) as a reference Fuyu model.
+
+## 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: Predict Tokens using `generate()` API
+In the example [generate.py](./generate.py), we show a basic use case for an Fuyu model to predict the next N tokens using `generate()` API, 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
+
+pip install transformers==4.35 pillow # additional package required for Fuyu to conduct generation
+```
+
+### 2. Run
+After setting up the Python environment, you could run the example by following steps.
+
+> **Note**: When loading the model in 4-bit, BigDL-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 Fuyu 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:
+```powershell
+python ./generate.py --image-path demo.jpg
+```
+More information about arguments can be found in [Arguments Info](#23-arguments-info) section. The expected output can be found in [Sample Output](#24-sample-output) 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 ./generate.py --image-path demo.jpg
+```
+More information about arguments can be found in [Arguments Info](#23-arguments-info) section. The expected output can be found in [Sample Output](#24-sample-output) 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 Fuyu model (e.g. `adept/fuyu-8b`) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'adept/fuyu-8b'`.
+- `--prompt PROMPT`: argument defining the prompt to be inferred (with the image for chat). It is default to be `'Generate a coco-style caption.'`.
+- `--image-path IMAGE_PATH`: argument defining the input image that the chat will focus on. It is required and should be a local path (not url).
+- `--n-predict N_PREDICT`: argument defining the max number of tokens to predict. It is default to be `512`.
+
+
+#### 2.4 Sample Output
+#### [adept/fuyu-8b](https://huggingface.co/adept/fuyu-8b)
+
+```log
+Inference time: xxxx s
+-------------------- Prompt --------------------
+Generate a coco-style caption.
+-------------------- Output --------------------
+An orange bus parked on the side of a road.
+```
+
+The sample input image is (which is fetched from [COCO dataset](https://cocodataset.org/#explore?id=178242)):
+
+[demo.jpg](https://cocodataset.org/#explore?id=178242)
+
+<a href="http://farm6.staticflickr.com/5331/8954873157_539393fece_z.jpg"><img width=400px src="http://farm6.staticflickr.com/5331/8954873157_539393fece_z.jpg" ></a>

python/llm/example/CPU/PyTorch-Models/Model/fuyu/generate.py

@@ -0,0 +1,68 @@
+#
+# 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.
+#
+
+from transformers import AutoModelForCausalLM, FuyuProcessor
+import torch
+import argparse
+import time
+from PIL import Image
+from bigdl.llm import optimize_model
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for Fuyu model')
+    parser.add_argument('--repo-id-or-model-path', type=str, default="adept/fuyu-8b",
+                        help='The huggingface repo id for the Fuyu model to be downloaded'
+                             ', or the path to the huggingface checkpoint folder')
+    parser.add_argument('--prompt', type=str, default="Generate a coco-style caption.",
+                        help='Prompt to infer')
+    parser.add_argument('--image-path', type=str, required=True,
+                        help='Image path for the input image that the chat will focus on')
+    parser.add_argument('--n-predict', type=int, default=512, help='Max tokens to predict')
+
+    args = parser.parse_args()
+    model_path = args.repo_id_or_model_path
+    prompt = args.prompt
+    image = Image.open(args.image_path)
+
+    # Load model
+    model = AutoModelForCausalLM.from_pretrained(model_path, device_map='cpu', trust_remote_code=True)
+
+    # With only one line to enable BigDL-LLM optimization on model
+    # For successful BigDL-LLM optimization on Fuyu, skip the 'vision_embed_tokens' module during optimization
+    model = optimize_model(model,
+                           low_bit='sym_int4',
+                           modules_to_not_convert=['vision_embed_tokens'])
+
+    # Load processor
+    processor = FuyuProcessor.from_pretrained(model_path)
+
+    # Generate predicted tokens
+    with torch.inference_mode():
+        inputs = processor(text=prompt, images=image, return_tensors="pt")
+        st = time.time()
+        generation_outputs = model.generate(**inputs,
+                                max_new_tokens=args.n_predict)
+        end = time.time()
+        outputs = processor.batch_decode(generation_outputs[:, -args.n_predict:], skip_special_tokens=True)
+        print(f'Inference time: {end-st} s')
+        print('-'*20, 'Prompt', '-'*20)
+        print(prompt)
+        print('-'*20, 'Output', '-'*20)
+        for output in outputs:
+            # '\x04' is the "beginning of answer" token
+            # See https://huggingface.co/adept/fuyu-8b#how-to-use
+            answer = output.split('\x04 ', 1)[1] if '\x04' in output else ''
+            print(answer)