The Nvidia's recent paper encourage the use of small language models (SLM) in agentic frameworks. There are many good agentic frameworks for programming, like Opencode. The good part of the programming tasks is that they provide environments, that can be tested automated, like is the code running or not. (This is the core idea of the AlphaEvolve RL algorithm.) Of course the correctness of the generated code can be challenging. This project's goal is exploring fine-tuning possibilities to get specialized SLM-based AI agents that can write a specific code family very well. An interesting example code family can be music generator python codes, for example the musicpy is a python library that can generate musics. But it is just a suggestion, we can specify other use-cases.
Hivatkozások
https://arxiv.org/abs/2506.02153
https://arxiv.org/abs/2402.03300
https://arxiv.org/abs/2506.13131
https://github.com/sst/opencode
https://github.com/Rainbow-Dreamer/musicpy