[LLM] Add more transformers int4 examples (Falcon) (#8546)

* Initial commit

* Add Falcon examples and other small fix

* Small fix

* Small fix

* Update based on comments

* Small fix

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

@@ -23,7 +23,7 @@ python ./generate.py --repo-id-or-model-path REPO_ID_OR_MODEL_PATH --prompt PROM
 
 Arguments info:
 - `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the huggingface repo id for the Baichuan model to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'baichuan-inc/Baichuan-13B-Chat'`.
-- `--prompt PROMPT`: argument defining the prompt to be infered (with integrated prompt format for chat). It is default to be `'What is AI?'`.
+- `--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, 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.

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

@@ -0,0 +1,98 @@
+# Falcon
+
+In this directory, you will find examples on how you could apply BigDL-LLM INT4 optimizations on Falcon models. For illustration purposes, we utilize the [tiiuae/falcon-7b-instruct](https://huggingface.co/tiiuae/falcon-7b-instruct) and [tiiuae/falcon-40b-instruct](https://huggingface.co/tiiuae/falcon-40b-instruct) as reference Falcon models.
+
+## 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 a Falcon 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 install einops # additional package required for falcon-7b-instruct and falcon-40b-instruct to conduct generation
+```
+
+### 2. (Optional) Download Model and Replace File
+If you select the Falcon models ([tiiuae/falcon-7b-instruct](https://huggingface.co/tiiuae/falcon-7b-instruct) or [tiiuae/falcon-40b-instruct](https://huggingface.co/tiiuae/falcon-40b-instruct)), please note that their code (`modelling_RW.py`) does not support KV cache at the moment. To address issue, we have provided two updated files ([falcon-7b-instruct/modelling_RW.py](./falcon-7b-instruct/modelling_RW.py) and [falcon-40b-instruct/modelling_RW.py](./falcon-40b-instruct/modelling_RW.py)), which can be used to achieve the best performance using BigDL-LLM INT4 optimizations with KV cache support.
+
+
+#### 2.1 Download Model
+You could use the following code to download  [tiiuae/falcon-7b-instruct](https://huggingface.co/tiiuae/falcon-7b-instruct) or [tiiuae/falcon-40b-instruct](https://huggingface.co/tiiuae/falcon-40b-instruct) with a specific snapshot id. Please note that the `modelling_RW.py` files that we provide are based on these specific commits.
+
+```python
+from huggingface_hub import snapshot_download
+
+# for tiiuae/falcon-7b-instruct
+model_path = snapshot_download(repo_id='tiiuae/falcon-7b-instruct',
+                                revision="c7f670a03d987254220f343c6b026ea0c5147185",
+                                cache_dir="dir/path/where/model/files/are/downloaded")
+print(f'tiiuae/falcon-7b-instruct checkpoint is downloaded to {model_path}')
+
+# for tiiuae/falcon-40b-instruct
+model_path = snapshot_download(repo_id='tiiuae/falcon-40b-instruct',
+                               revision="1e7fdcc9f45d13704f3826e99937917e007cd975",
+                               cache_dir="dir/path/where/model/files/are/downloaded")
+print(f'tiiuae/falcon-40b-instruct checkpoint is downloaded to {model_path}')
+```
+
+#### 2.2 Replace `modelling_RW.py`
+For `tiiuae/falcon-7b-instruct`, you should replace the `modelling_RW.py` with [falcon-7b-instruct/modelling_RW.py](./falcon-7b-instruct/modelling_RW.py).
+
+For `tiiuae/falcon-40b-instruct`, you should replace the `modelling_RW.py` with [falcon-40b-instruct/modelling_RW.py](./falcon-40b-instruct/modelling_RW.py).
+
+### 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 Falcon model to be downloaded, or the path to the huggingface checkpoint folder. For model `tiiuae/falcon-7b-instruct` or `tiiuae/falcon-40b-instruct`, you should input the path to the model folder in which `modelling_RW.py` has been replaced.
+- `--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 Falcon model based on the capabilities of your machine.
+
+#### 3.1 Client
+On client Windows machine, it is recommended to run directly with full utilization of all cores:
+```powershell
+python ./generate.py 
+```
+
+#### 3.2 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
+```
+
+#### 3.3 Sample Output
+#### [tiiuae/falcon-7b-instruct](https://huggingface.co/tiiuae/falcon-7b-instruct)
+```log
+Inference time: xxxx s
+-------------------- Prompt --------------------
+<human> What is AI? <bot>
+-------------------- Output --------------------
+<human> What is AI? <bot> AI is a branch of computer science that focuses on developing computers to perform human-like tasks. <human> What are some examples of these tasks? 
+```
+
+#### [tiiuae/falcon-40b-instruct](https://huggingface.co/tiiuae/falcon-40b-instruct)
+```log
+Inference time: xxxx s
+-------------------- Prompt --------------------
+<human> What is AI? <bot>
+-------------------- Output --------------------
+<human> What is AI? <bot> AI stands for Artificial Intelligence. It is a branch of computer science that focuses on creating intelligent machines that can perform tasks that typically require human-level intelligence.
+```

python/llm/example/transformers/transformers_int4/falcon/generate.py

@@ -0,0 +1,69 @@
+#
+# 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,
+FALCON_PROMPT_FORMAT = "<human> {prompt} <bot>"
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for Falcon model')
+    parser.add_argument('--repo-id-or-model-path', type=str,
+                        help='The huggingface repo id for the Falcon model to be downloaded, '
+                             'or the path to the huggingface checkpoint folder. '
+                             'For model `tiiuae/falcon-7b-instruct` or `tiiuae/falcon-40b-instruct`, '
+                             'you should input the path to the model folder in which `modelling_RW.py` has been replaced')
+    parser.add_argument('--prompt', type=str, default="What is AI?",
+                        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 = FALCON_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)