# Sixpert K1 - Complete Usage Guide

## Quick Start

### Option 1: Ollama (Easiest)

```bash
# Install Ollama
curl -fsSL https://ollama.com/install.sh | sh

# Download and import the model
ollama create sixpert-k1 -f OllamaModelfile

# Or if GGUF is in Ollama library:
# ollama run sixpert-k1

# Chat
ollama run sixpert-k1
```

### Option 2: llama-cpp-python (Python)

```bash
pip install llama-cpp-python
python examples/generate.py --prompt "Hello, who are you?"
```

### Option 3: API Server

```bash
pip install llama-cpp-python
python examples/api_server.py --model SixpertK1.gguf
```

### Option 4: LM Studio

1. Download LM Studio from https://lmstudio.ai
2. Import `SixpertK1.gguf`
3. Start chatting with the Sixpert K1 preset

## Chat Format

Sixpert K1 uses the following chat template:

```
<|im_start|>system
You are a helpful assistant.<|im_end|>
<|im_start|>user
What is quantum computing?<|im_end|>
<|im_start|>assistant
Quantum computing uses quantum mechanical phenomena...<|im_end|>
```

## Recommended Settings

| Parameter | Value | Notes |
|---|---|---|
| temperature | 0.7 | Good balance of creativity and accuracy |
| top_p | 0.8 | Nucleus sampling |
| top_k | 40 | Limit token selection |
| repeat_penalty | 1.05 | Prevent repetition |
| max_tokens | 8192 | Max output length |
| context_size | 131072 | Full context window |

## Function Calling

Sixpert K1 supports native function calling. See `examples/function_calling.py` for a complete implementation.

### Tool Format

```json
{
  "type": "function",
  "function": {
    "name": "search",
    "description": "Search for information",
    "parameters": {
      "type": "object",
      "properties": {
        "query": {"type": "string"}
      },
      "required": ["query"]
    }
  }
}
```

## Vision / Multimodal

Sixpert K1 can understand images. See `examples/vision_example.py` for implementation details.

```python
response = llm.create_chat_completion(
    messages=[{
        "role": "user",
        "content": [
            {"type": "text", "text": "Describe this image"},
            {"type": "image_url", "image_url": {"url": "data:image/png;base64,..."}},
        ]
    }]
)
```

## Integration Examples

### OpenAI-Compatible Client

```python
from openai import OpenAI

client = OpenAI(base_url="http://localhost:8000/v1", api_key="not-needed")

response = client.chat.completions.create(
    model="sixpert-k1",
    messages=[{"role": "user", "content": "Explain recursion"}],
    temperature=0.7,
)
print(response.choices[0].message.content)
```

### LangChain Integration

```python
from langchain.llms import LlamaCpp

llm = LlamaCpp(
    model_path="SixpertK1.gguf",
    temperature=0.7,
    n_ctx=131072,
    n_gpu_layers=-1,
)

result = llm.invoke("What is machine learning?")
print(result)
```

### CrewAI Agent

```python
from crewai import Agent, Task, Crew

agent = Agent(
    role="Research Analyst",
    backstory="You are Sixpert K1, a precision logic engine",
    goal="Provide accurate, detailed analysis",
    llm=LlamaCpp(model_path="SixpertK1.gguf", temperature=0.7),
    allow_delegation=False,
)
```

## Performance Tips

1. **GPU Offloading**: Set `n_gpu_layers=-1` to offload all layers to GPU
2. **Context Pruning**: Use smaller context windows (8192-32768) for faster inference
3. **Batch Processing**: Use the API server for batch inference
4. **Quantization**: Q4_K_M is the sweet spot; upgrade to Q6_K if quality matters more

## Troubleshooting

| Issue | Solution |
|---|---|
| Out of memory | Reduce context size or use CPU-only inference |
| Slow generation | Enable GPU offloading (`n_gpu_layers=-1`) |
| Repetitive output | Increase `repeat_penalty` to 1.1-1.2 |
| Hallucinations | Lower temperature to 0.3-0.5 |
| Context overflow | Use 4096 context for testing, 131072 for production |