diff --git a/python/llm/example/transformers/transformers_int4/GPU/llama2/README.md b/python/llm/example/transformers/transformers_int4/GPU/llama2/README.md index 23bf9257..547d5918 100644 --- a/python/llm/example/transformers/transformers_int4/GPU/llama2/README.md +++ b/python/llm/example/transformers/transformers_int4/GPU/llama2/README.md @@ -43,34 +43,32 @@ Arguments info: ```log Inference time: xxxx s -------------------- Prompt -------------------- -### HUMAN: -What is AI? +[INST] <> -### RESPONSE: +<> +What is AI? [/INST] -------------------- Output -------------------- -### HUMAN: -What is AI? +[INST] <> -### RESPONSE: +<> -AI is a term used to describe the development of computer systems that can perform tasks that typically require human intelligence, such as understanding natural language, recognizing images +What is AI? [/INST] Artificial intelligence (AI) is the broader field of research and development aimed at creating machines that can perform tasks that typically require human intelligence, ``` #### [meta-llama/Llama-2-13b-chat-hf](https://huggingface.co/meta-llama/Llama-2-13b-chat-hf) ```log Inference time: xxxx s -------------------- Prompt -------------------- -### HUMAN: -What is AI? +[INST] <> -### RESPONSE: +<> +What is AI? [/INST] -------------------- Output -------------------- -### HUMAN: -What is AI? +[INST] <> -### RESPONSE: +<> -AI, or artificial intelligence, refers to the ability of machines to perform tasks that would typically require human intelligence, such as learning, problem-solving, +What is AI? [/INST] AI stands for Artificial Intelligence, which refers to the ability of machines or computers to perform tasks that would typically require human intelligence, such as ``` diff --git a/python/llm/example/transformers/transformers_int4/GPU/llama2/generate.py b/python/llm/example/transformers/transformers_int4/GPU/llama2/generate.py index ac66c963..78d16246 100644 --- a/python/llm/example/transformers/transformers_int4/GPU/llama2/generate.py +++ b/python/llm/example/transformers/transformers_int4/GPU/llama2/generate.py @@ -24,12 +24,22 @@ from transformers import LlamaTokenizer # you could tune the prompt based on your own model, # here the prompt tuning refers to https://huggingface.co/georgesung/llama2_7b_chat_uncensored#prompt-style -LLAMA2_PROMPT_FORMAT = """### HUMAN: -{prompt} - -### RESPONSE: +DEFAULT_SYSTEM_PROMPT = """\ """ +def get_prompt(message: str, chat_history: list[tuple[str, str]], + system_prompt: str) -> str: + texts = [f'[INST] <>\n{system_prompt}\n<>\n\n'] + # The first user input is _not_ stripped + do_strip = False + for user_input, response in chat_history: + user_input = user_input.strip() if do_strip else user_input + do_strip = True + texts.append(f'{user_input} [/INST] {response.strip()} [INST] ') + message = message.strip() if do_strip else message + texts.append(f'{message} [/INST]') + return ''.join(texts) + if __name__ == '__main__': parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for Llama2 model') parser.add_argument('--repo-id-or-model-path', type=str, default="meta-llama/Llama-2-7b-chat-hf", @@ -56,7 +66,7 @@ if __name__ == '__main__': # Generate predicted tokens with torch.inference_mode(): - prompt = LLAMA2_PROMPT_FORMAT.format(prompt=args.prompt) + prompt = get_prompt(args.prompt, [], system_prompt=DEFAULT_SYSTEM_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, diff --git a/python/llm/example/transformers/transformers_int4/GPU/starcoder/generate.py b/python/llm/example/transformers/transformers_int4/GPU/starcoder/generate.py new file mode 100644 index 00000000..98e2c8fa --- /dev/null +++ b/python/llm/example/transformers/transformers_int4/GPU/starcoder/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 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, +StarCoder_PROMPT_FORMAT = "{prompt}" + +if __name__ == '__main__': + parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for StarCoder model') + parser.add_argument('--repo-id-or-model-path', type=str, default="bigcode/starcoder", + help='The huggingface repo id for the StarCoder 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=False, + trust_remote_code=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 = StarCoder_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) diff --git a/python/llm/example/transformers/transformers_int4/GPU/starcoder/readme.md b/python/llm/example/transformers/transformers_int4/GPU/starcoder/readme.md new file mode 100644 index 00000000..b733f19d --- /dev/null +++ b/python/llm/example/transformers/transformers_int4/GPU/starcoder/readme.md @@ -0,0 +1,63 @@ +# StarCoder +In this directory, you will find examples on how you could apply BigDL-LLM INT4 optimizations on StarCoder models on any Intel® Arc™ A-Series Graphics. For illustration purposes, we utilize the [bigcode/starcoder](https://huggingface.co/bigcode/starcoder) as a reference StarCoder model. + +## 0. Requirements +To run these examples with BigDL-LLM on Intel® Arc™ A-Series Graphics, 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 StarCoder model to predict the next N tokens using `generate()` API, with BigDL-LLM INT4 optimizations on Intel® Arc™ A-Series Graphics. +### 1. Install +We suggest using conda to manage environment: +```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 --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 StarCoder model to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'bigcode/starcoder'`. +- `--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`. + +#### Sample Output +#### [bigcode/starcoder](https://huggingface.co/bigcode/starcoder) +```log +Loading checkpoint shards: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████| 7/7 [02:07<00:00, 18.23s/it] +The attention mask and the pad token id were not set. As a consequence, you may observe unexpected behavior. Please pass your input's `attention_mask` to obtain reliable results. +Setting `pad_token_id` to `eos_token_id`:0 for open-end generation. +The attention mask and the pad token id were not set. As a consequence, you may observe unexpected behavior. Please pass your input's `attention_mask` to obtain reliable results. +Setting `pad_token_id` to `eos_token_id`:0 for open-end generation. +Inference time: xxxx s +-------------------- Prompt -------------------- +def print_hello_world(): +-------------------- Output -------------------- +def print_hello_world(): + print("Hello World!") + + +def print_hello_name(name): + print(f"Hello {name}!") + + +def print_ +```