feat: MLflow experiment tracking for Crusoe Managed Inference

[Sakshi3027]

What this adds

MLflow integration for tracking LLM experiments on Crusoe Managed Inference.

Logs 3 experiment runs out of the box:

• summarization
• reasoning
• code generation

What gets tracked per run

• Parameters — model, temperature, max_tokens, prompt
• Metrics — latency_seconds, output_word_count, words_per_second
• Artifacts — full model response saved as a text file

Why it's useful

Teams running LLM evaluations on Crusoe need a way to compare runs across prompts, temperatures, and models. This gives them a local MLflow tracking server with zero infrastructure setup just python train_and_log.py then mlflow ui.

Testing

Tested locally using Groq as a drop-in replacement. All 3 runs logged successfully with latency and throughput metrics visible in the MLflow UI.

To run on Crusoe:
export CRUSOE_API_KEY="your-api-key"
python train_and_log.py
mlflow ui