From 7a9fdf74df306e22072c139c558cd8f19c5a8a6c Mon Sep 17 00:00:00 2001 From: binbin Deng <108676127+plusbang@users.noreply.github.com> Date: Wed, 19 Jul 2023 18:20:16 +0800 Subject: [PATCH] [LLM] Add more transformers int4 example (Dolly v2) (#8571) * add * add trust_remote_mode --- python/llm/README.md | 1 + .../transformers/transformers_int4/README.md | 1 + .../transformers_int4/dolly_v2/README.md | 71 ++++++++++++++++ .../transformers_int4/dolly_v2/generate.py | 82 +++++++++++++++++++ 4 files changed, 155 insertions(+) create mode 100644 python/llm/example/transformers/transformers_int4/dolly_v2/README.md create mode 100644 python/llm/example/transformers/transformers_int4/dolly_v2/generate.py diff --git a/python/llm/README.md b/python/llm/README.md index 71220572..797426e5 100644 --- a/python/llm/README.md +++ b/python/llm/README.md @@ -24,6 +24,7 @@ We may use any Hugging Face Transfomer models on `bigdl-llm`, and the following | MOSS | [link](example/transformers/transformers_int4/moss) | | Baichuan | [link](example/transformers/transformers_int4/baichuan) | | Dolly-v1 | [link](example/transformers/transformers_int4/dolly_v1) | +| Dolly-v2 | [link](example/transformers/transformers_int4/dolly_v2) | | RedPajama | [link1](example/transformers/native_int4), [link2](example/transformers/transformers_int4/redpajama) | | Phoenix | [link1](example/transformers/native_int4), [link2](example/transformers/transformers_int4/phoenix) | | StarCoder | [link1](example/transformers/native_int4), [link2](example/transformers/transformers_int4/starcoder) | diff --git a/python/llm/example/transformers/transformers_int4/README.md b/python/llm/example/transformers/transformers_int4/README.md index fea6cb53..49bd6db4 100644 --- a/python/llm/example/transformers/transformers_int4/README.md +++ b/python/llm/example/transformers/transformers_int4/README.md @@ -12,6 +12,7 @@ You can use BigDL-LLM to run any Huggingface Transformer models with INT4 optimi | MOSS | [link](moss) | | Baichuan | [link](baichuan) | | Dolly-v1 | [link](dolly_v1) | +| Dolly-v2 | [link](dolly_v2) | | RedPajama | [link](redpajama) | | Phoenix | [link](phoenix) | | StarCoder | [link](starcoder) | diff --git a/python/llm/example/transformers/transformers_int4/dolly_v2/README.md b/python/llm/example/transformers/transformers_int4/dolly_v2/README.md new file mode 100644 index 00000000..7ef54e16 --- /dev/null +++ b/python/llm/example/transformers/transformers_int4/dolly_v2/README.md @@ -0,0 +1,71 @@ +# Dolly v2 +In this directory, you will find examples on how you could apply BigDL-LLM INT4 optimizations on Dolly v2 models. For illustration purposes, we utilize the [databricks/dolly-v2-12b](https://huggingface.co/databricks/dolly-v2-12b) as a reference Dolly v2 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 v2 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. 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 v2 model to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'databricks/dolly-v2-12b'`. +- `--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 v2 model based on the capabilities of your machine. + +#### 2.1 Client +On client Windows machine, it is recommended to run directly with full utilization of all cores: +```powershell +python ./generate.py +``` + +#### 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-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 +``` + +#### 2.3 Sample Output +#### [databricks/dolly-v2-12b](https://huggingface.co/databricks/dolly-v2-12b) +```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: +Artificial Intelligence (AI) is the area of computer science concerned with building machines that can perform tasks normally associated with human intelligence, such as reasoning, learning, +``` diff --git a/python/llm/example/transformers/transformers_int4/dolly_v2/generate.py b/python/llm/example/transformers/transformers_int4/dolly_v2/generate.py new file mode 100644 index 00000000..87298307 --- /dev/null +++ b/python/llm/example/transformers/transformers_int4/dolly_v2/generate.py @@ -0,0 +1,82 @@ +# +# 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 tuning refers to https://huggingface.co/databricks/dolly-v2-12b/blob/main/instruct_pipeline.py#L15 +DOLLY_V2_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='Predict Tokens using `generate()` API for Dolly v2 model') + parser.add_argument('--repo-id-or-model-path', type=str, default="databricks/dolly-v2-12b", + help='The huggingface repo id for the Dolly v2 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, + trust_remote_code=True) + + # Load tokenizer + tokenizer = AutoTokenizer.from_pretrained(model_path, + trust_remote_code=True) + + # Generate predicted tokens + with torch.inference_mode(): + prompt = DOLLY_V2_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)