diff --git a/python/llm/example/transformers/transformers_int4/dolly/v1/README.md b/python/llm/example/transformers/transformers_int4/dolly/v1/README.md new file mode 100644 index 00000000..fd98688f --- /dev/null +++ b/python/llm/example/transformers/transformers_int4/dolly/v1/README.md @@ -0,0 +1,75 @@ +# Dolly v1 +In this directory, you will find examples on how you could apply BigDL-LLM INT4 optimizations on Dolly v1 models. For illustration purposes, we utilize the [databricks/dolly-v1-6b](https://huggingface.co/databricks/dolly-v1-6b) as a reference Dolly v1 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 a Dolly v1 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 +``` + +### 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 Dolly v1 model to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'databricks/dolly-v1-6b'`. +- `--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 Dolly v1 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 +#### [databricks/dolly-v1-6b](https://huggingface.co/databricks/dolly-v1-6b) +```log +Inference time: xxxx s +-------------------- Prompt -------------------- +Below is an instruction that describes a task. Write a response that appropriately completes the request. + +### Instruction: +What is AI? + +### Response: + +-------------------- Output -------------------- +Below is an instruction that describes a task. Write a response that appropriately completes the request. + +### Instruction: +What is AI? + +### Response: +AI is an umbrella term for a variety of technologies that enable computers to think and act like humans. AI can be used to automate tasks, analyze data, and +``` diff --git a/python/llm/example/transformers/transformers_int4/dolly/v1/generate.py b/python/llm/example/transformers/transformers_int4/dolly/v1/generate.py new file mode 100644 index 00000000..6da4705a --- /dev/null +++ b/python/llm/example/transformers/transformers_int4/dolly/v1/generate.py @@ -0,0 +1,80 @@ +# +# 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 bigdl.llm.transformers import AutoModelForCausalLM +from transformers import AutoTokenizer + +# you could tune the prompt based on your own model, +# here the prompt format refers to https://huggingface.co/databricks/dolly-v1-6b#generate-text +DOLLY_V1_PROMPT_FORMAT = """Below is an instruction that describes a task. Write a response that appropriately completes the request. + +### Instruction: +{prompt} + +### Response: +""" + +if __name__ == '__main__': + parser = argparse.ArgumentParser(description='Transformer INT4 example for Dolly v1 model') + parser.add_argument('--repo-id-or-model-path', type=str, default="databricks/dolly-v1-6b", + help='The huggingface repo id for the Dolly v1 model to be downloaded' + ', or the path to the huggingface checkpoint folder') + 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) + + # Load tokenizer + tokenizer = AutoTokenizer.from_pretrained(model_path) + + # Generate predicted tokens + with torch.inference_mode(): + prompt = DOLLY_V1_PROMPT_FORMAT.format(prompt=args.prompt) + input_ids = tokenizer.encode(prompt, return_tensors="pt") + end_key_token_id=tokenizer.encode("### End")[0] + 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, + pad_token_id=tokenizer.pad_token_id, + eos_token_id=end_key_token_id) + end = time.time() + end_token_position = None + end_token_positions = np.where(output[0] == end_key_token_id)[0] + if len(end_token_positions) > 0: + end_token_position = end_token_positions[0] + output_str = tokenizer.decode(output[0][:end_token_position], skip_special_tokens=False) + 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/mpt/README.md b/python/llm/example/transformers/transformers_int4/mpt/README.md index e8a3f917..69f63159 100644 --- a/python/llm/example/transformers/transformers_int4/mpt/README.md +++ b/python/llm/example/transformers/transformers_int4/mpt/README.md @@ -1,7 +1,4 @@ # 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 @@ -60,8 +57,8 @@ numactl -C 0-47 -m 0 python -u ./generate.py #### [mosaicml/mpt-7b-chat](https://huggingface.co/mosaicml/mpt-7b-chat) ```log Inference time: xxxx s -Prompt: +-------------------- Prompt -------------------- What is AI? -Output: +-------------------- Output -------------------- What is AI? AI is the simulation of human intelligence in machines that are programmed to think and learn like humans. What is machine learning? Machine learning ``` diff --git a/python/llm/example/transformers/transformers_int4/mpt/generate.py b/python/llm/example/transformers/transformers_int4/mpt/generate.py index 7e543219..eae838f5 100644 --- a/python/llm/example/transformers/transformers_int4/mpt/generate.py +++ b/python/llm/example/transformers/transformers_int4/mpt/generate.py @@ -21,7 +21,7 @@ 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 +# you could tune the prompt based on your own model, MPT_PROMPT_FORMAT = "{prompt} " if __name__ == '__main__': @@ -62,5 +62,7 @@ if __name__ == '__main__': 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}') + print('-'*20, 'Prompt', '-'*20) + print(prompt) + print('-'*20, 'Output', '-'*20) + print(output_str)