> ## Documentation Index
> Fetch the complete documentation index at: https://docs.monostate.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Conversation

> Chat with your trained models

# Having Conversations

Once a model is loaded, you can start chatting.

## Sending Messages

1. Type your message in the input box
2. Press Enter or click Send
3. Wait for the model to respond
4. Continue the conversation

## Conversation Context

The chat maintains conversation history:

* Each message you send includes previous context
* The model "remembers" what you've discussed
* Longer conversations use more memory

### Context Window

Models have a maximum context length:

| Model             | Context Length |
| ----------------- | -------------- |
| Llama 3.2 (1B/3B) | 128K tokens    |
| Llama 3.1         | 128K tokens    |
| Mistral 7B v0.3   | 32K tokens     |
| Gemma 2           | 8K tokens      |
| Qwen 2.5          | 128K tokens    |

<Note>
  Context lengths vary by model version. Check the model card on Hugging Face for exact specifications.
</Note>

When context fills up, oldest messages may be dropped.

## Conversation Tips

### For Testing Fine-tuned Models

Test with prompts similar to your training data:

```
Training data: Customer support conversations
Test prompt: "I can't log into my account"

Training data: Code generation
Test prompt: "Write a Python function to sort a list"
```

### For Evaluating Quality

Ask questions that reveal model capabilities:

* **Factual**: "What is the capital of France?"
* **Reasoning**: "If A > B and B > C, is A > C?"
* **Creative**: "Write a haiku about programming"
* **Domain-specific**: Questions from your fine-tuning domain

### For Finding Issues

Test edge cases:

* Very short inputs ("Hi")
* Very long inputs
* Unusual characters or formatting
* Questions outside training domain
* Attempts to confuse the model

## Clearing History

To start fresh:

* Look for "Clear" or "New Chat" button
* Or reload the page

This is useful when:

* Testing different scenarios
* Context gets too long
* Starting a new demo

## Multi-turn Conversations

The model sees the full conversation:

```
User: What's 2 + 2?
Assistant: 4

User: And if we add 3 more?
Assistant: That would be 7 (4 + 3 = 7)
```

The second response uses context from the first exchange.

## Common Patterns

### Question-Answer Testing

```
User: [Question]
Assistant: [Answer]
User: Can you explain that differently?
Assistant: [Reformulated answer]
```

### Instruction Following

```
User: Write a poem about cats. Make it exactly 4 lines.
Assistant: [Poem]
User: Now make it about dogs instead
Assistant: [Modified poem]
```

### Role-Playing

```
User: You are a helpful customer service agent. A customer says: "My order is late"
Assistant: [Response in character]
```

## Next Steps

<CardGroup cols={2}>
  <Card title="Parameters" href="/chat/parameters">
    Adjust generation settings
  </Card>

  <Card title="CLI Training" href="/cli/llm-training">
    Train better models
  </Card>
</CardGroup>