On-policy distillation tutorial with Latest models

[snowmanwwg]

Maybe let's do gpt-oss? It is just supported in the mcore path

[coderabbitai]

📝 CodeRabbit Plan Mode

Generate an implementation plan and prompts that you can use with your favorite coding agent.

• Create Plan

Examples

• Example 1
• Example 2

---

---

🧪 Issue enrichment is currently in open beta.

You can configure auto-planning by selecting labels in the issue_enrichment configuration.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord or schedule a call!

[zpqiu]

Since on-policy distillation uses the same backend modules as the GRPO algorithm—including both training and rollout modules—theoretically, any model supported by GRPO will be automatically compatible with the on-policy distillation algorithm.

Users can override the model name following the method provided in the README to experiment with new models.

If users encounter errors when running the on-policy distillation algorithm with a model that is supposedly supported by GRPO, they could open a bug issue.