README.md

---
license: mit
---


## About llamafile
<a href="https://github.com/Mozilla-Ocho/llamafile" target="_blank">github</a><br/>
<a href="https://www.bingal.com/posts/ai-llamafile-usage/" target="_blank">llamafile 中文说明</a><br/>
<a href="https://www.modelscope.cn/models/bingal/llamafile-models/summary" target="_blank">The llamafile model collection in modelscope.cn</a><br/>
<a href="https://www.modelscope.cn/models/bingal/Qwen1.5-14B-Chat-llamafile/summary" target="_blank">Qwen1.5-14B-Chat-llamafile in modelscope.cn</a>

## Useage

**The Windows system has a limitation where it does not support a single exe file over 4GB, so it is necessary to download the llamafile and gguf models separately and run them individually. In addition, you can also use the Windows Subsystem for Linux (WSL) to run it, which can bypass the 4GB limit as well.**
### Windows
  * Download [llamafile](https://github.com/Mozilla-Ocho/llamafile/releases/download/0.6.2/llamafile-0.6.2).
  * Rename the file to `llamafile-0.6.2.exe`
  * Download GGUF model [qwen1_5-14b-chat-q5_k_m.gguf](https://huggingface.co/Qwen/Qwen1.5-14B-Chat-GGUF/resolve/main/qwen1_5-14b-chat-q5_k_m.gguf)
  * Open terminal window, and run: `\llamafile-0.6.2.exe .\qwen1_5-14b-chat-q5_k_m.gguf -ngl 9999 --host 0.0.0.0 --port 8080`
  * Open browser to http://127.0.0.1:8080 to start chatting

### Linux / macOS
1. Download model: [qwen1.5-14b-chat-q5_k_m.llamafile](https://huggingface.co/bingal/Qwen1.5-14B-Chat-llamafile/resolve/main/qwen1.5-14b-chat-q5_k_m.llamafile)
2. Run the model
  * Add execution permissions: `chmod +x ./qwen1.5-14b-chat-q5_k_m.llamafile`
  * Run in terminal: `./qwen1.5-14b-chat-q5_k_m.llamafile`
  * Open browser to http://127.0.0.1:8080 to start chatting
    
3. Openai api usage
   * api url: `http://127.0.0.1:8080/v1`
   * Python code:
   ```python
   #!/usr/bin/env python3
   from openai import OpenAI
   client = OpenAI(
       base_url="http://127.0.0.1:8080/v1", # "http://<Your api-server IP>:port"
       api_key = "sk-no-key-required"
   )
   completion = client.chat.completions.create(
       model="LLaMA_CPP",
       messages=[
           {"role": "system", "content": "You are an AI assistant."},
           {"role": "user", "content": "Write a story about dragon"}
       ]
   )
   print(completion.choices[0].message)
    ```

## Parameter Description

- `-ngl 9999` indicates how many layers of the model are placed on the GPU to run, with the rest running on the CPU. If there is no GPU available, it can be set to `-ngl 0`. The default is 9999, which means everything runs on the GPU (drivers and CUDA runtime environment must be installed).
- `--host 0.0.0.0` is the hostname for the web service. If only local access is needed, it can be set to `--host 127.0.0.1`. If set to `0.0.0.0`, it can be accessed via IP within the network.
- `--port 8080` is the port for the web service, with the default being `8080`, which can be modified using this parameter.
- `-t 16` is the number of threads. When running on the CPU, you can set the number of cores to run concurrently based on the CPU core count.
- Other parameters can be viewed with `--help`.