Add ziya CPU example (#10114)

* ziya on CPU

* add README for ziya

* specify use_cache

* add arc CPU

* update prompt format

* update link

* add comments to emphasize use_cache

* update pip cmd

README.md

@@ -185,6 +185,7 @@ Over 40 models have been optimized/verified on `bigdl-llm`, including *LLaMA/LLa
 | RWKV5 |  | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/rwkv5) |
 | Bark | [link](python/llm/example/CPU/PyTorch-Models/Model/bark) | [link](python/llm/example/GPU/PyTorch-Models/Model/bark) |
 | SpeechT5 |  | [link](python/llm/example/GPU/PyTorch-Models/Model/speech-t5) |
+| Ziya-Coding-34B-v1.0 | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/ziya) | |
 
 ***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

@@ -81,7 +81,7 @@ Over 20 models have been optimized/verified on `bigdl-llm`, including *LLaMA/LLa
 | RWKV5 |  | [link](example/GPU/HF-Transformers-AutoModels/Model/rwkv5) |
 | Bark | [link](example/CPU/PyTorch-Models/Model/bark) | [link](example/GPU/PyTorch-Models/Model/bark) |
 | SpeechT5 |  | [link](example/GPU/PyTorch-Models/Model/speech-t5) |
-
+| Ziya-Coding-34B-v1.0 | [link](example/CPU/HF-Transformers-AutoModels/Model/ziya) | |
 
 ### Working with `bigdl-llm`
 

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

@@ -0,0 +1,88 @@
+# Ziya
+
+In this directory, you will find examples on how you could apply BigDL-LLM INT4 optimizations on Ziya models. For illustration purposes, we utilize the [IDEA-CCNL/Ziya-Coding-34B-v1.0](https://huggingface.co/IDEA-CCNL/Ziya-Coding-34B-v1.0) as a reference Ziya model.
+
+> **Note**: If you want to download the Hugging Face *Transformers* model, please refer to [here](https://huggingface.co/docs/hub/models-downloading#using-git).
+>
+> BigDL-LLM optimizes the *Transformers* model in INT4 precision at runtime, and thus no explicit conversion is needed.
+
+## 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 Ziya 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 einops  # additional package required for Ziya 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 Ziya 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 --prompt 'def quick_sort(arr):\n'
+```
+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-LLM env variables
+source bigdl-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 --prompt 'def quick_sort(arr):\n'
+```
+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`: str, argument defining the huggingface repo id for the Ziya model to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'IDEA-CCNL/Ziya-Coding-34B-v1.0'`.
+- `--prompt`: str, argument defining the prompt to be inferred (with integrated prompt format for chat). It is default to be `def quick_sort(arr):\n`.
+- `--n-predict`: int, argument defining the max number of tokens to predict. It is default to be `128`.
+
+#### 2.4 Sample Output
+#### [IDEA-CCNL/Ziya-Coding-34B-v1.0](https://huggingface.co/IDEA-CCNL/Ziya-Coding-34B-v1.0)
+```log
+Inference time: xxxx s
+-------------------- Prompt --------------------
+<human>: 
+def quick_sort(arr):\n
+<bot>: 
+
+-------------------- Output --------------------
+<s> <human>: 
+def quick_sort(arr):\n
+<bot>: 
+def partition(arr, low, high):
+
+    i = (low-1)
+    pivot = arr[high]
+    for j in range(low, high):
+        if arr[j] <= pivot:
+            arr[i], arr[j] = arr[j], arr[i]
+            i = i+1
+    arr[i], arr[high] = arr[high], arr[i]
+    return i
+
+def quick_sort(arr, low, high):
+    if low < high:
+        pi = partition(arr, low,
+```

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

@@ -0,0 +1,77 @@
+#
+# 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
+import numpy as np
+
+from transformers import AutoTokenizer
+
+# you could tune the prompt based on your own model,
+# here the prompt tuning refers to https://huggingface.co/IDEA-CCNL/Ziya-Coding-34B-v1.0
+ZIYA_PROMPT_FORMAT = "<human>: \n{prompt}\n<bot>: \n"
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for Ziya model')
+    parser.add_argument('--repo-id-or-model-path', type=str, default="IDEA-CCNL/Ziya-Coding-34B-v1.0",
+                        help='The huggingface repo id for the Ziya model to be downloaded'
+                             ', or the path to the huggingface checkpoint folder')
+    parser.add_argument('--prompt', type=str, default="def quick_sort(arr):\n",
+                        help='Prompt to infer') 
+    parser.add_argument('--n-predict', type=int, default=128,
+                        help='Max tokens to predict')
+
+    args = parser.parse_args()
+    model_path = args.repo_id_or_model_path
+
+    
+    from bigdl.llm.transformers import AutoModelForCausalLM
+    # 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 Ziya models
+    model = AutoModelForCausalLM.from_pretrained(model_path,
+                                                 load_in_4bit=True,
+                                                 trust_remote_code=True,
+                                                 use_cache=True)
+
+    # Load tokenizer
+    tokenizer = AutoTokenizer.from_pretrained(model_path,
+                                              trust_remote_code=True)
+    
+    # Generate predicted tokens
+    with torch.inference_mode():
+        prompt = ZIYA_PROMPT_FORMAT.format(prompt=args.prompt)
+        input_ids = tokenizer.encode(prompt, return_tensors="pt")
+        st = time.time()
+        
+        output = model.generate(input_ids,
+                                max_new_tokens=args.n_predict,
+                                do_sample = True,
+                                top_p = 0.85,
+                                temperature = 0.8,
+                                repetition_penalty = 0.95,
+                                eos_token_id = tokenizer.eos_token_id,
+                                pad_token_id = tokenizer.pad_token_id,
+                                )
+        end = time.time()
+        output_str = tokenizer.batch_decode(output)[0]
+        print(f'Inference time: {end-st} s')
+        print('-'*20, 'Prompt', '-'*20)
+        print(prompt)
+        print('-'*20, 'Output', '-'*20)
+        print(output_str)

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

@@ -0,0 +1,78 @@
+# Ziya
+In this directory, you will find examples on how you could use BigDL-LLM `optimize_model` API to accelerate Ziya models. For illustration purposes, we utilize the [IDEA-CCNL/Ziya-Coding-34B-v1.0](https://huggingface.co/IDEA-CCNL/Ziya-Coding-34B-v1.0) as a reference Ziya 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 a Ziya model to predict the next N tokens using `generate()` API, with BigDL-LLM 'optimize_model' API.
+### 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 einops  # additional package required for Ziya to conduct generation
+```
+
+### 2. Run
+After setting up the Python environment, you could run the example by following steps.
+#### 2.1 Client
+On client Windows machines, it is recommended to run directly with full utilization of all cores:
+```powershell
+python ./generate.py --prompt 'def quick_sort(arr):\n'
+```
+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-LLM env variables
+source bigdl-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
+```
+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`: str, argument defining the huggingface repo id for the Ziya model to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'IDEA-CCNL/Ziya-Coding-34B-v1.0'`.
+- `--prompt`: str, argument defining the prompt to be inferred (with integrated prompt format for chat). It is default to be `def quick_sort(arr):\n`.
+- `--n-predict`: int, argument defining the max number of tokens to predict. It is default to be `128`.
+
+#### 2.4 Sample Output
+#### [IDEA-CCNL/Ziya-Coding-34B-v1.0](https://huggingface.co/IDEA-CCNL/Ziya-Coding-34B-v1.0)
+```log
+Inference time: xxxx s
+-------------------- Prompt --------------------
+<human>: 
+def quick_sort(arr):\n
+<bot>: 
+
+-------------------- Output --------------------
+<s> <human>: 
+def quick_sort(arr):\n
+<bot>: 
+def partition(arr, low, high):
+
+    i = (low-1)
+    pivot = arr[high]
+    for j in range(low, high):
+        if arr[j] <= pivot:
+            arr[i], arr[j] = arr[j], arr[i]
+            i = i+1
+    arr[i], arr[high] = arr[high], arr[i]
+    return i
+
+def quick_sort(arr, low, high):
+    if low < high:
+        pi = partition(arr, low,
+```

python/llm/example/CPU/PyTorch-Models/Model/ziya/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 time
+import argparse
+import numpy as np
+
+from transformers import AutoTokenizer
+
+
+ZIYA_PROMPT_FORMAT = "<human>: \n{prompt}\n<bot>: \n"
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for Ziya model')
+    parser.add_argument('--repo-id-or-model-path', type=str, default="IDEA-CCNL/Ziya-Coding-34B-v1.0",
+                        help='The huggingface repo id for the Ziya model to be downloaded'
+                             ', or the path to the huggingface checkpoint folder')
+    parser.add_argument('--prompt', type=str, default="def quick_sort(arr):\n",
+                        help='Prompt to infer') 
+    parser.add_argument('--n-predict', type=int, default=128,
+                        help='Max tokens to predict')
+
+    args = parser.parse_args()
+    model_path = args.repo_id_or_model_path
+
+    
+    from transformers import AutoModelForCausalLM
+    from bigdl.llm import optimize_model
+    # 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 `optimization_model` API optimizations,
+    # it is important to set use_cache=True for Ziya models
+    model = AutoModelForCausalLM.from_pretrained(model_path,
+                                                 trust_remote_code=True,
+                                                 use_cache=True)
+    model = optimize_model(model)
+
+    # Load tokenizer
+    tokenizer = AutoTokenizer.from_pretrained(model_path,
+                                              trust_remote_code=True)
+    
+    # Generate predicted tokens
+    with torch.inference_mode():
+        prompt = ZIYA_PROMPT_FORMAT.format(prompt=args.prompt)
+        input_ids = tokenizer.encode(prompt, return_tensors="pt")
+ 
+        st = time.time()
+        output = model.generate(input_ids,
+                                max_new_tokens=args.n_predict,
+                                do_sample = True,
+                                top_p = 0.85,
+                                temperature = 0.8,
+                                repetition_penalty = 0.95,
+                                eos_token_id = tokenizer.eos_token_id,
+                                pad_token_id = tokenizer.pad_token_id,
+                                )
+        end = time.time()
+        output_str = tokenizer.batch_decode(output)[0]
+        print(f'Inference time: {end-st} s')
+        print('-'*20, 'Prompt', '-'*20)
+        print(prompt)
+        print('-'*20, 'Output', '-'*20)
+        print(output_str)
+