mirror of
https://github.com/ml-explore/mlx-examples.git
synced 2025-08-30 02:53:41 +08:00
42 lines
989 B
Markdown
42 lines
989 B
Markdown
# Qwen
|
|
|
|
Qwen (通义千问) is a language model developed by Alibaba Cloud.[^1] The
|
|
architecture of Qwen is similar to Llama except for the bias in the attention
|
|
layers.
|
|
|
|
## Setup
|
|
|
|
First download and convert the model with:
|
|
|
|
```sh
|
|
python convert.py
|
|
```
|
|
The script downloads the model from Hugging Face. The default model is
|
|
`Qwen/Qwen-1_8B`. Check out the [Hugging Face page](https://huggingface.co/Qwen) to see a list of available models.
|
|
|
|
The conversion script will make the `weights.npz` and `config.json` files in
|
|
the working directory.
|
|
|
|
## Generate
|
|
|
|
To generate text with the default prompt:
|
|
|
|
```sh
|
|
python qwen.py
|
|
```
|
|
|
|
If you change the model, make sure to pass the corresponding tokenizer. E.g.,
|
|
for Qwen 7B use:
|
|
|
|
```
|
|
python qwen.py --tokenizer Qwen/Qwen-7B
|
|
```
|
|
|
|
To see a list of options, run:
|
|
|
|
```sh
|
|
python qwen.py --help
|
|
```
|
|
|
|
[^1]: For more details on the model see the official repo of [Qwen](https://github.com/QwenLM/Qwen) and the [Hugging Face](https://huggingface.co/Qwen).
|