Benchmark

# Benchmark Command

Run performance benchmarks for MyMLXServer.

## Available Commands

- `single` - Single request benchmark
- `concurrent` - Concurrent request benchmark
- `compare` - Compare model configurations
- `eval` - Evaluation tasks (use /eval instead)

## Usage

```bash
uv run benchmark $ARGUMENTS
```

## Examples

### Single Request Benchmark
```bash
uv run benchmark single --prompt "Write a story" --max-tokens 256 --iterations 5
```

### Concurrent Benchmark (requires running server)
```bash
uv run benchmark concurrent --requests 10 --concurrency 4 --server http://localhost:1245
```

### Model Comparison
```bash
uv run benchmark compare --iterations 3 --draft-models "mlx-community/gemma-3-4b-it-qat-4bit"
```

## Output Formats

- `--format table` (default) - Rich table output
- `--format json` - JSON output
- `--format csv` - CSV output

Save to file with `--output <path>`.