The production playbook for open-source LLMs: fine-tuning, distillation & deployment

Fine-tuning open-source LLMs is no longer experimental, it’s the fastest path to production-grade performance, lower costs, and domain-specific accuracy.

In this live session, we’ll show how leading teams go from raw data → fine-tuning → distillation → deployment using Nebius Token Factory’s Post-Training service, powered by Papyrax, the fastest multi-node JAX framework in its class.

See how enterprise partners build custom, high-speed, cost-efficient models ready for real-world use.

What you’ll learnWhat you’ll learn

• When to prompt, fine-tune, or distill and how to decide.
• Which open-source models (DeepSeek V3, GPT-OSS 120B, Qwen3 Coder 480B, Kimi) fine-tune best.
• How to prepare and clean production data for training.
• Running efficient LoRA, QLoRA, and full fine-tuning on multi-node clusters.
• Distillation workflows that turn 100B models into fast, low-cost students.
• 1-click deployment with built-in evals, autoscaling, and zero-retention inference.

You’ll leave with practical templates, dataset examples, benchmark results, and deploy-ready code.

Who should attendWho should attend

• ML engineers optimizing inference & cost
• AI developers building copilots, agents, or RAGs
• Platform teams evaluating open-source LLMs
• Founders scaling vertical AI products