This reverts commit 6930422b42.
This commit is contained in:
Shengsheng Huang
GitHub
parent
6930422b42
commit
db0d129226
7 changed files with 0 additions and 288 deletions
|
|
@ -181,7 +181,6 @@ Over 40 models have been optimized/verified on `bigdl-llm`, including *LLaMA/LLa
|
|||
| Fuyu | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/fuyu) | |
|
||||
| Distil-Whisper | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/distil-whisper) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/distil-whisper) |
|
||||
| Yi | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/yi) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/yi) |
|
||||
| RWKV | [link](python/llm/example/CPU/PyTorch-Models/Model/rwkv) | [link](python/llm/example/GPU/PyTorch-Models/Model/rwkv) |
|
||||
| BlueLM | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/bluelm) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/bluelm) |
|
||||
| Mamba | [link](python/llm/example/CPU/PyTorch-Models/Model/mamba) | [link](python/llm/example/GPU/PyTorch-Models/Model/mamba) |
|
||||
| SOLAR | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/solar) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/solar) |
|
||||
|
|
|
|||
|
|
@ -73,7 +73,6 @@ Over 20 models have been optimized/verified on `bigdl-llm`, including *LLaMA/LLa
|
|||
| Fuyu | [link](example/CPU/HF-Transformers-AutoModels/Model/fuyu) | |
|
||||
| Distil-Whisper | [link](example/CPU/HF-Transformers-AutoModels/Model/distil-whisper) | [link](example/GPU/HF-Transformers-AutoModels/Model/distil-whisper) |
|
||||
| Yi | [link](example/CPU/HF-Transformers-AutoModels/Model/yi) | [link](example/GPU/HF-Transformers-AutoModels/Model/yi) |
|
||||
| RWKV | [link](example/CPU/PyTorch-Models/Model/rwkv) | [link](example/GPU/PyTorch-Models/Model/rwkv) |
|
||||
| BlueLM | [link](example/CPU/HF-Transformers-AutoModels/Model/bluelm) | [link](example/GPU/HF-Transformers-AutoModels/Model/bluelm) |
|
||||
| Mamba | [link](example/CPU/PyTorch-Models/Model/mamba) | [link](example/GPU/PyTorch-Models/Model/mamba) |
|
||||
| SOLAR | [link](example/CPU/HF-Transformers-AutoModels/Model/solar) | [link](example/GPU/HF-Transformers-AutoModels/Model/solar) |
|
||||
|
|
|
|||
|
|
@ -1,75 +0,0 @@
|
|||
# RWKV
|
||||
|
||||
In this directory, you will find examples on how you could apply BigDL-LLM INT4 optimizations on RWKV models. For illustration purposes, we utilize the [RWKV/rwkv-4-world-7b](https://huggingface.co/RWKV/rwkv-4-world-7b) as a reference RWKV 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 RWKV 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
|
||||
```
|
||||
|
||||
### 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 RWKV 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 "\nIn a shocking finding, scientist discovered a herd of dragons living in a remote, previously unexplored valley"
|
||||
```
|
||||
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-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 --prompt "\nIn a shocking finding, scientist discovered a herd of dragons living in a remote, previously unexplored valley"
|
||||
```
|
||||
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 RWKV model to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'RWKV/rwkv-4-world-7b'`.
|
||||
- `--prompt`: str, argument defining the prompt to be inferred (with integrated prompt format for chat). It is default to be "\nIn a shocking finding, scientist discovered a herd of dragons living in a remote, previously unexplored valley".
|
||||
- `--n-predict`: int, argument defining the max number of tokens to predict. It is default to be `40`.
|
||||
|
||||
#### 2.4 Sample Output
|
||||
#### [RWKV/rwkv-4-world-7b](https://huggingface.co/RWKV/rwkv-4-world-7b)
|
||||
```log
|
||||
Inference time: xxxx s
|
||||
-------------------- Prompt --------------------
|
||||
Question:
|
||||
In a shocking finding, scientist discovered a herd of dragons living in a remote, previously unexplored valley, in Tibet. Even more surprising to the researchers was the fact that the dragons spoke perfect Chinese.
|
||||
|
||||
Answer:
|
||||
-------------------- Output --------------------
|
||||
Question:
|
||||
In a shocking finding, scientist discovered a herd of dragons living in a remote, previously unexplored valley, in Tibet. Even more surprising to the researchers was the fact that the dragons spoke perfect Chinese.
|
||||
|
||||
Answer: 科学家在一个不为人知的谷地发现一群能说中文的龙。科学家惊讶地发现这些龙是中国的
|
||||
|
||||
```
|
||||
|
|
@ -1,71 +0,0 @@
|
|||
#
|
||||
# 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 ag8reed 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
|
||||
|
||||
from transformers import AutoModel, AutoTokenizer, AutoModelForCausalLM
|
||||
from bigdl.llm import optimize_model
|
||||
|
||||
# you could tune the prompt based on your own model,
|
||||
# here the prompt tuning refers to https://huggingface.co/RWKV/rwkv-4-world-7b
|
||||
RWKV_PROMPT_FORMAT = "Question: {prompt}\n\nAnswer:"
|
||||
|
||||
if __name__ == '__main__':
|
||||
parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for RWKV model')
|
||||
parser.add_argument('--repo-id-or-model-path', type=str, default="RWKV/rwkv-4-world-7b",
|
||||
help='The huggingface repo id for the RWKV model to be downloaded'
|
||||
', or the path to the huggingface checkpoint folder')
|
||||
parser.add_argument('--prompt', type=str, default="\nIn a shocking finding, scientist discovered a herd of dragons living in a remote, previously unexplored valley, in Tibet. Even more surprising to the researchers was the fact that the dragons spoke perfect Chinese.",
|
||||
help='Prompt to infer')
|
||||
parser.add_argument('--n-predict', type=int, default=40,
|
||||
help='Max tokens to predict')
|
||||
|
||||
args = parser.parse_args()
|
||||
model_path = args.repo_id_or_model_path
|
||||
|
||||
# First load the model in fp16 dtype
|
||||
model = AutoModelForCausalLM.from_pretrained(model_path,
|
||||
trust_remote_code=True,
|
||||
low_cpu_mem_usage=True,
|
||||
torch_dtype=torch.half)
|
||||
|
||||
# Call the `_rescale_layers` method, prepare to convert to int4
|
||||
model.rwkv._rescale_layers()
|
||||
|
||||
# With only one line to enable BigDL-LLM optimization on model
|
||||
model = optimize_model(model)
|
||||
|
||||
# Load tokenizer
|
||||
tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
|
||||
|
||||
# Generate predicted tokens
|
||||
with torch.inference_mode():
|
||||
prompt = RWKV_PROMPT_FORMAT.format(prompt = args.prompt)
|
||||
|
||||
inputs = tokenizer(prompt, return_tensors="pt")
|
||||
st = time.time()
|
||||
output = model.generate(inputs["input_ids"],
|
||||
max_new_tokens=args.n_predict)
|
||||
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)
|
||||
|
|
@ -69,7 +69,6 @@ if __name__ == '__main__':
|
|||
end = time.time()
|
||||
output = output.cpu()
|
||||
output_str = tokenizer.decode(output[0], skip_special_tokens=True)
|
||||
|
||||
print(f'Inference time: {end-st} s')
|
||||
print('-'*20, 'Output', '-'*20)
|
||||
print(output_str)
|
||||
|
|
|
|||
|
|
@ -1,59 +0,0 @@
|
|||
# RWKV
|
||||
|
||||
In this directory, you will find examples on how you could use BigDL-LLM `optimize_model` API to accelerate RWKV models on [Intel GPUs](../README.md). For illustration purposes, we utilize the [RWKV/rwkv-4-world-7b](https://huggingface.co/RWKV/rwkv-4-world-7b) as a reference RWKV 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 RWKV model to predict the next N tokens using `generate()` API, with BigDL-LLM INT4 optimizations on Intel GPUs.
|
||||
### 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[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 --prompt "你叫什么名字？"
|
||||
```
|
||||
|
||||
Arguments info:
|
||||
- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the huggingface repo id for the RWKV model (e.g. `RWKV/rwkv-4-world-7b`) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'RWKV/rwkv-4-world-7b'`.
|
||||
- `--prompt PROMPT`: argument defining the prompt to be infered (with integrated prompt format for chat). It is default to be `"你叫什么名字？"`.
|
||||
- `--n-predict N_PREDICT`: argument defining the max number of tokens to predict. It is default to be `40`.
|
||||
|
||||
#### Sample Output
|
||||
#### [RWKV/rwkv-4-world-7b](https://huggingface.co/RWKV/rwkv-4-world-7b)
|
||||
|
||||
```log
|
||||
Inference time: xxxx s
|
||||
-------------------- Prompt --------------------
|
||||
Question: 你叫什么名字？
|
||||
|
||||
Answer:
|
||||
-------------------- Output --------------------
|
||||
Question: 你叫什么名字？
|
||||
|
||||
Answer: 我是一个大型语言模型，没有具体的姓名。我是由OpenAI团队创建的，目的是为了提供自然
|
||||
|
||||
|
||||
```
|
||||
|
|
@ -1,80 +0,0 @@
|
|||
#
|
||||
# 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 transformers import AutoModel, AutoTokenizer, AutoModelForCausalLM
|
||||
from bigdl.llm import optimize_model
|
||||
|
||||
# you could tune the prompt based on your own model,
|
||||
# here the prompt tuning refers to https://huggingface.co/RWKV/rwkv-4-world-7b
|
||||
|
||||
RWKV_PROMPT_FORMAT = "Question: {prompt}\n\nAnswer:"
|
||||
|
||||
if __name__ == '__main__':
|
||||
parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for RWKV model')
|
||||
parser.add_argument('--repo-id-or-model-path', type=str, default="RWKV/rwkv-4-world-7b",
|
||||
help='The huggingface repo id for the RWKV model to be downloaded'
|
||||
', or the path to the huggingface checkpoint folder')
|
||||
parser.add_argument('--prompt', type=str, default="你叫什么名字？",
|
||||
help='Prompt to infer')
|
||||
parser.add_argument('--n-predict', type=int, default=40,
|
||||
help='Max tokens to predict')
|
||||
|
||||
args = parser.parse_args()
|
||||
model_path = args.repo_id_or_model_path
|
||||
|
||||
# First load the model in fp16 dtype
|
||||
model = AutoModelForCausalLM.from_pretrained(model_path,
|
||||
trust_remote_code=True,
|
||||
low_cpu_mem_usage=True,
|
||||
torch_dtype=torch.half)
|
||||
|
||||
# Call the `_rescale_layers` method, prepare to convert to int4
|
||||
model.rwkv._rescale_layers()
|
||||
|
||||
# With only one line to enable BigDL-LLM optimization on model
|
||||
model = optimize_model(model)
|
||||
model = model.to('xpu')
|
||||
|
||||
# Load tokenizer
|
||||
tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
|
||||
|
||||
# Generate predicted tokens
|
||||
with torch.inference_mode():
|
||||
prompt = RWKV_PROMPT_FORMAT.format(prompt = args.prompt)
|
||||
inputs = tokenizer(prompt, return_tensors="pt").to('xpu')
|
||||
|
||||
# ipex model needs a warmup, then inference time can be accurate
|
||||
output = model.generate(inputs["input_ids"],
|
||||
max_new_tokens=args.n_predict)
|
||||
|
||||
# start inference
|
||||
st = time.time()
|
||||
output = model.generate(inputs["input_ids"],
|
||||
max_new_tokens=args.n_predict)
|
||||
torch.xpu.synchronize()
|
||||
output = output.cpu()
|
||||
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)
|
||||
Loading…
Reference in a new issue