1 articles with this tag
SRPO framework enables LLMs to self-reflect on errors, generating dense training signals that drastically improve data efficiency and achieve SOTA on reasoning and agentic benchmarks.
