From db0d12922622b031b4bc35d09f973850452011fa Mon Sep 17 00:00:00 2001 From: Shengsheng Huang Date: Wed, 28 Feb 2024 11:48:31 +0800 Subject: [PATCH] Revert "Add rwkv example (#9432)" (#10264) This reverts commit 6930422b427a70a165f36a63cf3e1156f7de9397. --- README.md | 1 - python/llm/README.md | 1 - .../CPU/PyTorch-Models/Model/rwkv/README.md | 75 ----------------- .../CPU/PyTorch-Models/Model/rwkv/generate.py | 71 ---------------- .../PyTorch-Models/Model/replit/generate.py | 1 - .../GPU/PyTorch-Models/Model/rwkv/README.md | 59 -------------- .../GPU/PyTorch-Models/Model/rwkv/generate.py | 80 ------------------- 7 files changed, 288 deletions(-) delete mode 100644 python/llm/example/CPU/PyTorch-Models/Model/rwkv/README.md delete mode 100644 python/llm/example/CPU/PyTorch-Models/Model/rwkv/generate.py delete mode 100644 python/llm/example/GPU/PyTorch-Models/Model/rwkv/README.md delete mode 100644 python/llm/example/GPU/PyTorch-Models/Model/rwkv/generate.py diff --git a/README.md b/README.md index e7e73a56..5aa58129 100644 --- a/README.md +++ b/README.md @@ -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) | diff --git a/python/llm/README.md b/python/llm/README.md index a8700dd3..4ddced57 100644 --- a/python/llm/README.md +++ b/python/llm/README.md @@ -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) | diff --git a/python/llm/example/CPU/PyTorch-Models/Model/rwkv/README.md b/python/llm/example/CPU/PyTorch-Models/Model/rwkv/README.md deleted file mode 100644 index 1cf6c8b1..00000000 --- a/python/llm/example/CPU/PyTorch-Models/Model/rwkv/README.md +++ /dev/null @@ -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 - -``` diff --git a/python/llm/example/CPU/PyTorch-Models/Model/rwkv/generate.py b/python/llm/example/CPU/PyTorch-Models/Model/rwkv/generate.py deleted file mode 100644 index 162d6b2e..00000000 --- a/python/llm/example/CPU/PyTorch-Models/Model/rwkv/generate.py +++ /dev/null @@ -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) diff --git a/python/llm/example/GPU/PyTorch-Models/Model/replit/generate.py b/python/llm/example/GPU/PyTorch-Models/Model/replit/generate.py index 52ef1a27..57f2c9f6 100644 --- a/python/llm/example/GPU/PyTorch-Models/Model/replit/generate.py +++ b/python/llm/example/GPU/PyTorch-Models/Model/replit/generate.py @@ -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) diff --git a/python/llm/example/GPU/PyTorch-Models/Model/rwkv/README.md b/python/llm/example/GPU/PyTorch-Models/Model/rwkv/README.md deleted file mode 100644 index 7ce117ee..00000000 --- a/python/llm/example/GPU/PyTorch-Models/Model/rwkv/README.md +++ /dev/null @@ -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 - - -``` diff --git a/python/llm/example/GPU/PyTorch-Models/Model/rwkv/generate.py b/python/llm/example/GPU/PyTorch-Models/Model/rwkv/generate.py deleted file mode 100644 index e9ab7407..00000000 --- a/python/llm/example/GPU/PyTorch-Models/Model/rwkv/generate.py +++ /dev/null @@ -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)