Is post-training needed before running agentic reasoning #19
Description
Many thanks for the excellent work on multi tools agentic reasoning. I was just wondering if specific post training (for example ppo or grpo) is needed in order to let the main reasoning model be more capable of utilising external agents to solve the problem.
As can be shown from your paper, your work is triggering the intrinsic ability of deepseek r1 to invoke external tools like code agent and search agent with few-shot prompt. Am I understanding it correctly