|
--- |
|
license: apache-2.0 |
|
language: |
|
- zh |
|
tags: |
|
- Chinese |
|
--- |
|
|
|
# Open-Chinese-LLaMA-7B-Patch |
|
|
|
This model is a **Chinese large language model base** generated from the [LLaMA](https://github.com/facebookresearch/llama)-7B model after **secondary pre-training** on Chinese datasets. |
|
|
|
This model is a **patch** model and must be used in conjunction with the official weights. For the installation of the patch and related tutorials, please refer to [OpenLMLab/llama](https://github.com/OpenLMLab/llama). |
|
|
|
## Usage |
|
|
|
Since the official weights for [LLaMA](https://github.com/facebookresearch/llama)-7B have not been open-sourced, the model released this time is of the **patch** type, which needs to be used in combination with the original official weights. |
|
|
|
You can install the **patch** using `tools/patch_model.py`, for example: |
|
|
|
```bash |
|
|
|
python tools/patch_model.py --base_model <path_or_name_to_original_model> |
|
--patch_model openlmlab/open-chinese-llama-7b-patch |
|
--base_model_format <hf_or_raw> |
|
|
|
``` |
|
|
|
The **patch** is installed in place, which means that the installed **patch** is the complete `hf` format weight. You can use `transformers` to load the model. |
|
|
|
## Quick Experience via Command Line |
|
|
|
The **patched** model can be easily loaded by `transformers`. For a quick experience, we provide a console Demo: |
|
|
|
```bash |
|
|
|
python cli_demo.py --model openlmlab/open-chinese-llama-7b-patch |
|
--devices 0 |
|
--max_length 1024 |
|
--do_sample true |
|
--top_k 40 |
|
--top_p 0.8 |
|
--temperature 0.7 |
|
--penalty 1.02 |
|
|
|
``` |
