[LLM] Add more transformers_int4 examples (MPT) (#8498)

* Update transformers_int4 readme, and initial commit for mpt

* Update example for mpt

* Small fix and recover transformers_int4_pipeline_readme.md for now

* Update based on comments

* Small fix

* Small fix

* Update based on comments

python/llm/example/transformers/transformers_int4/README.md

@@ -0,0 +1,17 @@
+# BigDL-LLM Transformers INT4 Optimization for Large Language Model
+You can use BigDL-LLM to run any Huggingface Transformer models with INT4 optimizations on either servers or laptops. 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.
+
+## Recommended Requirements
+To run the examples, we recommend using Intel® Xeon® processors (server), or >= 12th Gen Intel® Core™ processor (client).
+
+For OS, BigDL-LLM supports Ubuntu 20.04 or later, CentOS 7 or later, and Windows 10/11.
+
+## Best Known Configuration
+For better performance, it is recommended to set environment variables with the help of BigDL-Nano:
+```bash
+pip install bigdl-nano
+```
+following with
+| Linux | Windows (powershell)|
+|:------|:-------|
+|`source bigdl-nano-init`|`bigdl-nano-init`|

python/llm/example/transformers/transformers_int4/mpt/README.md

@@ -0,0 +1,67 @@
+# MPT
+
+MPT models are part of the MosaicPretrainedTransformer (MPT) model family, and designed for text generation tasks.
+
+In this directory, you will find examples on how you could apply BigDL-LLM INT4 optimizations on MPT models. For illustration purposes, we utilize the [mosaicml/mpt-7b-chat](https://huggingface.co/mosaicml/mpt-7b-chat) as a reference MPT 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 MPT model to predict the next N tokens using `generate()` API, with BigDL-LLM INT4 optimizations.
+### 1. Install
+We suggest using conda to manage environment:
+```bash
+conda create -n llm python=3.9
+conda activate llm
+
+pip install bigdl-llm[all] # install bigdl-llm with 'all' option
+pip instll einops  # additional package required for mpt-7b-chat to conduct generation
+```
+
+### 2. Config
+It is recommended to set several environment variables for better performance. Please refer to [here](../README.md#best-known-configuration) for more information.
+
+### 3. 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 MPT model to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'mosaicml/mpt-7b-chat'`.
+- `--prompt PROMPT`: argument defining the prompt to be infered (with integrated prompt format for chat). It is default to be `'What is AI?'`.
+- `--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, 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 MPT model based on the capabilities of your machine.
+
+#### 3.1 Client
+For better utilization of multiple cores on the client machine, it is recommended to use all the performance-cores along with their hyperthreads.
+
+E.g. on Windows,
+```powershell
+# for a client machine with 8 Performance-cores
+$env:OMP_NUM_THREADS=16
+python ./generate.py
+```
+
+#### 3.2 Server
+On server, it is recommended to run the example with all the physical cores of a single socket.
+
+E.g. on Linux,
+```bash
+# for a server with 48 cores per socket
+export OMP_NUM_THREADS=48
+numactl -C 0-47 -m 0 python -u ./generate.py
+```
+
+#### 3.3 Sample Output
+#### [mosaicml/mpt-7b-chat](https://huggingface.co/mosaicml/mpt-7b-chat)
+```log
+Inference time: xxxx s
+Prompt:
+<human>What is AI? <bot>
+Output:
+<human>What is AI? <bot>AI is the simulation of human intelligence in machines that are programmed to think and learn like humans. <human>What is machine learning? <bot>Machine learning
+```

python/llm/example/transformers/transformers_int4/mpt/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.
+#
+
+import torch
+import time
+import argparse
+
+from bigdl.llm.transformers import AutoModelForCausalLM
+from transformers import AutoTokenizer
+
+# you could revise it based on the MPT model you choose to use
+MPT_PROMPT_FORMAT="<human>{prompt} <bot>"
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser(description='Transformer INT4 example for MPT model')
+    parser.add_argument('--repo-id-or-model-path', type=str, default="mosaicml/mpt-7b-chat",
+                        help='The huggingface repo id for the large language model to be downloaded'
+                             ', or the path to the huggingface checkpoint folder')
+    parser.add_argument('--prompt', type=str, default="<human>What is AI?<bot>",
+                        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,
+                                                 trust_remote_code=True,
+                                                 load_in_4bit=True)
+
+    # Load tokenizer
+    tokenizer = AutoTokenizer.from_pretrained(model_path,
+                                              trust_remote_code=True)
+    
+    # Generate predicted tokens
+    with torch.inference_mode():
+        prompt = MPT_PROMPT_FORMAT.format(prompt=args.prompt)
+        input_ids = tokenizer.encode(prompt, return_tensors="pt")
+        st = time.time()
+        # enabling `use_cache=True` allows the model to utilize the previous
+        # key/values attentions to speed up decoding;
+        # to obtain optimal performance with BigDL-LLM INT4 optimizations,
+        # it is important to set use_cache=True for MPT models
+        output = model.generate(input_ids,
+                                use_cache=True,
+                                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(f'Prompt:\n{prompt}')
+        print(f'Output:\n{output_str}')