34 lines
1.1 KiB
Markdown
34 lines
1.1 KiB
Markdown
# LLaMA Chat Runner
|
||
|
||
This project provides a simple way to build and run conversational prompts with [llama.cpp](https://github.com/ggerganov/llama.cpp).
|
||
It uses PHP templates to format system and user messages into the correct structure for inference, and a Makefile to glue everything together.
|
||
|
||
## How it Works
|
||
- **`llm_chat.php`**: Collects the system message and user/assistant messages, then renders them with `llama_chat_template.php`.
|
||
- **`llama_chat_template.php`**: Outputs a structured chat template for LLaMA models.
|
||
- **`Makefile`**:
|
||
- Combines `.sys` files into `consolidated_system_message.txt`
|
||
- Pipes messages into `llama.cpp` (`llama-cli`) for inference
|
||
- Saves the model’s reply in `agent_message.txt`
|
||
|
||
## Usage
|
||
1. Place your `.sys` files (system instructions) in the repo folder.
|
||
2. Create `user_message.txt` with your prompt.
|
||
3. Run:
|
||
```bash
|
||
make agent_message.txt
|
||
```
|
||
4. The model’s response will be saved in ```agent_message.txt.```
|
||
Requirements
|
||
• llama.cpp built with llama-cli
|
||
• PHP CLI
|
||
• A compatible LLaMA model (set in the Makefile)
|
||
|
||
Cleaning Up
|
||
|
||
To remove generated files:
|
||
```
|
||
make clean
|
||
```
|
||
|