Qwen3-1.7B-Base

Researchers have developed RLP, an information-driven reinforcement pretraining objective that integrates exploration and chain-of-thought reasoning into the pretraining phase of large language models…