ayo/ipex-llm
1
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 1

Update NPU example readme (#11931)

Browse source
This commit is contained in:
binbin Deng 2024-08-27 12:44:58 +08:00 committed by GitHub
parent 6c3eb1e1e8
commit 14dddfc0d6
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 16 additions and 51 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

65
python/llm/example/NPU/HF-Transformers-AutoModels/LLM/README.md
View file

@ -9,7 +9,7 @@ In this directory, you will find examples on how you could apply IPEX-LLM INT4 o
| Llama3 | [meta-llama/Meta-Llama-3-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct) | | Llama3 | [meta-llama/Meta-Llama-3-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct) |
| Chatglm3 | [THUDM/chatglm3-6b](https://huggingface.co/THUDM/chatglm3-6b) | | Chatglm3 | [THUDM/chatglm3-6b](https://huggingface.co/THUDM/chatglm3-6b) |
| Chatglm2 | [THUDM/chatglm2-6b](https://huggingface.co/THUDM/chatglm2-6b) | | Chatglm2 | [THUDM/chatglm2-6b](https://huggingface.co/THUDM/chatglm2-6b) |
| Qwen2 | [Qwen/Qwen2-7B-Instruct](https://huggingface.co/Qwen/Qwen2-7B-Instruct) | | Qwen2 | [Qwen/Qwen2-7B-Instruct](https://huggingface.co/Qwen/Qwen2-7B-Instruct), [Qwen/Qwen2-1.5B-Instruct](https://huggingface.co/Qwen/Qwen2-1.5B-Instruct) |
| MiniCPM | [openbmb/MiniCPM-2B-sft-bf16](https://huggingface.co/openbmb/MiniCPM-2B-sft-bf16) | | MiniCPM | [openbmb/MiniCPM-2B-sft-bf16](https://huggingface.co/openbmb/MiniCPM-2B-sft-bf16) |
| Phi-3 | [microsoft/Phi-3-mini-4k-instruct](https://huggingface.co/microsoft/Phi-3-mini-4k-instruct) | | Phi-3 | [microsoft/Phi-3-mini-4k-instruct](https://huggingface.co/microsoft/Phi-3-mini-4k-instruct) |
| Stablelm | [stabilityai/stablelm-zephyr-3b](https://huggingface.co/stabilityai/stablelm-zephyr-3b) | | Stablelm | [stabilityai/stablelm-zephyr-3b](https://huggingface.co/stabilityai/stablelm-zephyr-3b) |
@ -23,10 +23,8 @@ Go to https://www.intel.com/content/www/us/en/download/794734/intel-npu-driver-w
Then go to **Device Manager**, find **Neural Processors** -> **Intel(R) AI Boost**. Then go to **Device Manager**, find **Neural Processors** -> **Intel(R) AI Boost**.
Right click and select **Update Driver**. And then manually select the folder unzipped from the driver. Right click and select **Update Driver**. And then manually select the folder unzipped from the driver.
## Example 1: Predict Tokens using `generate()` API ## 1. Install
In the example [generate.py](./generate.py), we show a basic use case for a Llama2 model to predict the next N tokens using `generate()` API, with IPEX-LLM INT4 optimizations on Intel NPUs. ### 1.1 Installation on Windows
### 1. Install
#### 1.1 Installation on Windows
We suggest using conda to manage environment: We suggest using conda to manage environment:
```bash ```bash
conda create -n llm python=3.10 conda create -n llm python=3.10
@ -36,9 +34,9 @@ conda activate llm
pip install --pre --upgrade ipex-llm[npu] pip install --pre --upgrade ipex-llm[npu]
``` ```
### 2. Runtime Configurations ## 2. Runtime Configurations
For optimal performance, it is recommended to set several environment variables. Please check out the suggestions based on your device. For optimal performance, it is recommended to set several environment variables. Please check out the suggestions based on your device.
#### 2.1 Configurations for Windows ### 2.1 Configurations for Windows
> [!NOTE] > [!NOTE]
> For optimal performance, we recommend running code in `conhost` rather than Windows Terminal: > For optimal performance, we recommend running code in `conhost` rather than Windows Terminal:
@ -54,19 +52,20 @@ For optimal performance, it is recommended to set several environment variables.
set BIGDL_USE_NPU=1 set BIGDL_USE_NPU=1
``` ```
### 3. Running examples ## 3. Run models
In the example [generate.py](./generate.py), we show a basic use case for a Llama2 model to predict the next N tokens using `generate()` API, with IPEX-LLM INT4 optimizations on Intel NPUs.
``` ```
python ./generate.py python ./generate.py
``` ```
Arguments info: Arguments info:
- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the huggingface repo id for the Llama2 model (e.g. `meta-llama/Llama-2-7b-chat-hf` and `meta-llama/Llama-2-13b-chat-hf`) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'meta-llama/Llama-2-7b-chat-hf'`, and more verified models please see the list in [Verified Models](#verified-models). - `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the huggingface repo id for the Llama2 model (e.g. `meta-llama/Llama-2-7b-chat-hf`) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'meta-llama/Llama-2-7b-chat-hf'`, and more verified models please see the list in [Verified Models](#verified-models).
- `--prompt PROMPT`: argument defining the prompt to be infered (with integrated prompt format for chat). It is default to be `'Once upon a time, there existed a little girl who liked to have adventures. She wanted to go to places and meet new people, and have fun'`. - `--prompt PROMPT`: argument defining the prompt to be infered (with integrated prompt format for chat). It is default to be `'Once upon a time, there existed a little girl who liked to have adventures. She wanted to go to places and meet new people, and have fun'`.
- `--n-predict N_PREDICT`: argument defining the max number of tokens to predict. It is default to be `32`. - `--n-predict N_PREDICT`: argument defining the max number of tokens to predict. It is default to be `32`.
- `--load_in_low_bit`: argument defining the `load_in_low_bit` format used. It is default to be `sym_int8`, `sym_int4` can also be used. - `--load_in_low_bit`: argument defining the `load_in_low_bit` format used. It is default to be `sym_int8`, `sym_int4` can also be used.
#### Sample Output ### Sample Output
#### [meta-llama/Llama-2-7b-chat-hf](https://huggingface.co/meta-llama/Llama-2-7b-chat-hf) #### [meta-llama/Llama-2-7b-chat-hf](https://huggingface.co/meta-llama/Llama-2-7b-chat-hf)
```log ```log
@ -77,44 +76,10 @@ Inference time: xxxx s
done done
``` ```
## Example 2: Predict Tokens using `generate()` API using multi processes ## 4. Run Optimized Models (Experimental)
In the example [llama2.py](./llama2.py) and [qwen2.py](./qwen2.py), we show an experimental support for a Llama2 / Qwen2 model to predict the next N tokens using `generate()` API, with IPEX-LLM INT4 optimization and fused decoderlayer optimization on Intel NPUs. The example below shows how to run the **_optimized model implementations_** on Intel NPU, including
- [Llama2-7B](./llama2.py)
> [!IMPORTANT] - [Qwen2-1.5B](./qwen2.py)
> To run Qwen2 and Llama2 with IPEX-LLM on Intel NPUs, we recommend using version **32.0.100.2540** for the Intel NPU.
>
> Go to https://www.intel.com/content/www/us/en/download/794734/825735/intel-npu-driver-windows.html to download and unzip the driver. Then follow the same steps on [Requirements](#0-requirements).
### 1. Install
#### 1.1 Installation on Windows
We suggest using conda to manage environment:
```bash
conda create -n llm python=3.10
conda activate llm
# install ipex-llm with 'npu' option
pip install --pre --upgrade ipex-llm[npu]
```
### 2. Runtime Configurations
For optimal performance, it is recommended to set several environment variables. Please check out the suggestions based on your device.
#### 2.1 Configurations for Windows
> [!NOTE]
> For optimal performance, we recommend running code in `conhost` rather than Windows Terminal:
> - Press <kbd>Win</kbd>+<kbd>R</kbd> and input `conhost`, then press Enter to launch `conhost`.
> - Run following command to use conda in `conhost`. Replace `<your conda install location>` with your conda install location.
> ```
> call <your conda install location>\Scripts\activate
> ```
**Following envrionment variables are required**:
```cmd
set BIGDL_USE_NPU=1
```
### 3. Running examples
``` ```
# to run Llama-2-7b-chat-hf # to run Llama-2-7b-chat-hf
@ -132,7 +97,7 @@ Arguments info:
- `--max-prompt-len MAX_PROMPT_LEN`: Defines the maximum number of tokens that the input prompt can contain. It is default to be `512`. - `--max-prompt-len MAX_PROMPT_LEN`: Defines the maximum number of tokens that the input prompt can contain. It is default to be `512`.
- `--disable-transpose-value-cache`: Disable the optimization of transposing value cache. - `--disable-transpose-value-cache`: Disable the optimization of transposing value cache.
### 4. Troubleshooting ### Troubleshooting
If you encounter output problem, please try to disable the optimization of transposing value cache with following command: If you encounter output problem, please try to disable the optimization of transposing value cache with following command:
```bash ```bash
@ -144,7 +109,7 @@ python qwen2.py --disable-transpose-value-cache
``` ```
#### Sample Output ### Sample Output
#### [meta-llama/Llama-2-7b-chat-hf](https://huggingface.co/meta-llama/Llama-2-7b-chat-hf) #### [meta-llama/Llama-2-7b-chat-hf](https://huggingface.co/meta-llama/Llama-2-7b-chat-hf)
```log ```log