LLM: Add Replit CPU and GPU example (#9028)

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

@@ -22,6 +22,7 @@ You can use BigDL-LLM to run any Huggingface Transformer models with INT4 optimi
 | Whisper   | [link](whisper)   |
 | Qwen      | [link](qwen)      |
 | Aquila    | [link](aquila)    |
+| Replit    | [link](replit)    |
 | Mistral   | [link](mistral)   |
 
 ## Recommended Requirements

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

@@ -0,0 +1,66 @@
+# Replit
+In this directory, you will find examples on how you could apply BigDL-LLM INT4 optimizations on Replit models. For illustration purposes, we utilize the [replit/replit-code-v1-3b](https://huggingface.co/replit/replit-code-v1-3b) as a reference Replit model.
+
+## 0. 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 Replit 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
+```
+
+### 2. Run
+After setting up the Python environment, you could run the example by following steps.
+
+#### 2.1 Client
+On client Windows machine, it is recommended to run directly with full utilization of all cores:
+```powershell
+python ./generate.py --prompt 'def print_hello_world():'
+```
+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
+```
+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 Replit model to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'replit/replit-code-v1-3b'`.
+- `--prompt PROMPT`: argument defining the prompt to be infered (with integrated prompt format for chat). It is default to be `'def print_hello_world():'`.
+- `--n-predict N_PREDICT`: argument defining the max number of tokens to predict. It is default to be `32`.
+
+#### 2.4 Sample Output
+#### [replit/replit-code-v1-3b](https://huggingface.co/bigcode/replit/replit-code-v1-3b)
+```log
+-------------------- Prompt --------------------
+def print_hello_world():
+-------------------- Output --------------------
+def print_hello_world():
+    print("Hello")
+    print("World")
+
+print_hello_world()
+
+
+def print_hello_world():
+    print
+```

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

@@ -0,0 +1,67 @@
+#
+# 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 bigdl.llm.transformers import AutoModelForCausalLM
+from transformers import AutoTokenizer
+
+# you could tune the prompt based on your own model,
+REPLIT_PROMPT_FORMAT = "{prompt}"
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for Replit model')
+    parser.add_argument('--repo-id-or-model-path', type=str, default="replit/replit-code-v1-3b",
+                        help='The huggingface repo id for the Replit to be downloaded'
+                             ', or the path to the huggingface checkpoint folder')
+    parser.add_argument('--prompt', type=str, default="def print_hello_world():",
+                        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
+
+    # Load model in 4 bit,
+    # which convert the relevant layers in the model into INT4 format
+    model = AutoModelForCausalLM.from_pretrained(model_path,
+                                                 load_in_4bit=True,
+                                                 trust_remote_code=True)
+
+    # Load tokenizer
+    tokenizer = AutoTokenizer.from_pretrained(model_path,
+                                              trust_remote_code=True)
+    
+    # Generate predicted tokens
+    with torch.inference_mode():
+        prompt = REPLIT_PROMPT_FORMAT.format(prompt=args.prompt)
+        input_ids = tokenizer.encode(prompt, return_tensors="pt")
+        st = time.time()
+        # if your selected model is capable of utilizing previous key/value attentions
+        # to enhance decoding speed, but has `"use_cache": false` in its model config,
+        # it is important to set `use_cache=True` explicitly in the `generate` function
+        # to obtain optimal performance with BigDL-LLM INT4 optimizations
+        output = model.generate(input_ids,
+                                max_new_tokens=args.n_predict)
+        end = time.time()
+        output_str = tokenizer.decode(output[0], skip_special_tokens=True)
+        print(f'Inference time: {end-st} s')
+        print('-'*20, 'Prompt', '-'*20)
+        print(prompt)
+        print('-'*20, 'Output', '-'*20)
+        print(output_str)

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

@@ -2,6 +2,7 @@
 You can use BigDL-LLM to run almost every Huggingface Transformer models with INT4 optimizations on your laptops with Intel GPUs. This directory contains example scripts to help you quickly get started using BigDL-LLM to run some popular open-source models in the community. Each model has its own dedicated folder, where you can find detailed instructions on how to install and run it.
 
 ## Verified models
+
 | Model          | Example                                                  |
 |----------------|----------------------------------------------------------|
 | Aquila         | [link](aquila)                                           |
@@ -21,6 +22,8 @@ You can use BigDL-LLM to run almost every Huggingface Transformer models with IN
 | StarCoder      | [link](starcoder)                                        |
 | Vicuna         | [link](vicuna)                                           |
 | Whisper        | [link](whisper)                                          |
+| Replit         | [link](replit)                                           |
+
 
 ## Verified Hardware Platforms
 

python/llm/example/GPU/HF-Transformers-AutoModels/Model/replit/README.md

@@ -0,0 +1,64 @@
+# Replit
+In this directory, you will find examples on how you could apply BigDL-LLM INT4 optimizations on Replit models on [Intel GPUs](../README.md). For illustration purposes, we utilize the [replit/replit-code-v1-3b](https://huggingface.co/replit/replit-code-v1-3b) as a reference Replit model.
+
+## 0. Requirements
+To run these examples with BigDL-LLM on Intel GPUs, 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 Replit model to predict the next N tokens using `generate()` API, with BigDL-LLM INT4 optimizations on Intel GPUs.
+### 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
+conda activate llm
+# below command will install intel_extension_for_pytorch==2.0.110+xpu as default
+# you can install specific ipex/torch version for your need
+pip install --pre --upgrade bigdl-llm[xpu] -f https://developer.intel.com/ipex-whl-stable-xpu
+```
+
+### 2. Configures OneAPI environment variables
+```bash
+source /opt/intel/oneapi/setvars.sh
+```
+
+### 3. Run
+
+For optimal performance on Arc, it is recommended to set several environment variables.
+
+```bash
+export USE_XETLA=OFF
+export SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1
+```
+
+```
+python ./generate.py --prompt 'def print_hello_world():'
+```
+More information about arguments can be found in [Arguments Info](#31-arguments-info) section. The expected output can be found in [Sample Output](#32-sample-output) section.
+
+#### 3.1 Arguments Info
+In the example, several arguments can be passed to satisfy your requirements:
+
+Arguments info:
+- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the huggingface repo id for the Replit model to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'replit/replit-code-v1-3b'`.
+- `--prompt PROMPT`: argument defining the prompt to be infered (with integrated prompt format for chat). It is default to be `'def print_hello_world():'`.
+- `--n-predict N_PREDICT`: argument defining the max number of tokens to predict. It is default to be `32`.
+
+#### 3.2 Sample Output
+#### [replit/replit-code-v1-3b](https://huggingface.co/replit/replit-code-v1-3b)
+```log
+Inference time: xxxx s
+-------------------- Prompt --------------------
+def print_hello_world():
+-------------------- Output --------------------
+def print_hello_world():
+    print("Hello")
+    print("World")
+
+print_hello_world()
+
+
+def print_hello_world():
+    print
+```

python/llm/example/GPU/HF-Transformers-AutoModels/Model/replit/generate.py

@@ -0,0 +1,78 @@
+#
+# 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 intel_extension_for_pytorch as ipex
+import time
+import argparse
+
+from bigdl.llm.transformers import AutoModelForCausalLM
+from transformers import AutoTokenizer
+
+# you could tune the prompt based on your own model,
+REPLIT_PROMPT_FORMAT = "{prompt}"
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for Replit model')
+    parser.add_argument('--repo-id-or-model-path', type=str, default="replit/replit-code-v1-3b",
+                        help='The huggingface repo id for the Replit to be downloaded'
+                             ', or the path to the huggingface checkpoint folder')
+    parser.add_argument('--prompt', type=str, default="def print_hello_world():",
+                        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
+
+    # Load model in 4 bit,
+    # which convert the relevant layers in the model into INT4 format
+    model = AutoModelForCausalLM.from_pretrained(model_path,
+                                                 load_in_4bit=True,
+                                                 optimize_model=True,
+                                                 trust_remote_code=True,
+                                                 use_cache=True)
+    model = model.to('xpu')
+
+    # Load tokenizer
+    tokenizer = AutoTokenizer.from_pretrained(model_path,
+                                              trust_remote_code=True)
+    
+    # Generate predicted tokens
+    with torch.inference_mode():
+        prompt = REPLIT_PROMPT_FORMAT.format(prompt=args.prompt)
+        input_ids = tokenizer.encode(prompt, return_tensors="pt").to('xpu')
+
+        # ipex model needs a warmup, then inference time can be accurate
+        output = model.generate(input_ids,
+                                max_new_tokens=args.n_predict)
+
+        # start inference
+        st = time.time()
+        # if your selected model is capable of utilizing previous key/value attentions
+        # to enhance decoding speed, but has `"use_cache": false` in its model config,
+        # it is important to set `use_cache=True` explicitly in the `generate` function
+        # to obtain optimal performance with BigDL-LLM INT4 optimizations
+        output = model.generate(input_ids,
+                                max_new_tokens=args.n_predict)
+        torch.xpu.synchronize()
+        end = time.time()
+        output_str = tokenizer.decode(output[0], skip_special_tokens=True)
+        print(f'Inference time: {end-st} s')
+        print('-'*20, 'Prompt', '-'*20)
+        print(prompt)
+        print('-'*20, 'Output', '-'*20)
+        print(output_str)