[LLM] Add more transformers int4 example (Dolly v1) (#8517)
* Initial commit for dolly v1 * Add example for Dolly v1 and other small fix * Small output updates * Small fix * fix based on comments
This commit is contained in:
Yuwen Hu
GitHub
parent
90e3d86bce
commit
349bcb4bae
4 changed files with 162 additions and 8 deletions
|
|
@ -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
|
||||
```
|
||||
|
|
@ -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)
|
||||
|
|
@ -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 --------------------
|
||||
<human>What is AI? <bot>
|
||||
Output:
|
||||
-------------------- Output --------------------
|
||||
<human>What is AI? <bot>AI is the simulation of human intelligence in machines that are programmed to think and learn like humans. <human>What is machine learning? <bot>Machine learning
|
||||
```
|
||||
|
|
|
|||
|
|
@ -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 = "<human>{prompt} <bot>"
|
||||
|
||||
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)
|
||||
|
|
|
|||
Loading…
Reference in a new issue