Unverified50% confidenceFactExact time
The researchers used Edu-QuRater scores as a reward term for GRPO post-training, with Qwen3-4B as the base model.
1
Sources
50%
Confidence
Long-term
Relevance
9/11/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedQwen3 used a GRPO-like algorithm for reinforcement training of reasoning capabilities.77% similarUnverifiedAfter combining the Edu-QuRater reward with the answer structure reward, the generated responses outperformed the Qwen3-4B base model on both the instructional quality and instruction-following dimensions.69% similarUnverifiedThe success of GRPO has driven the emergence of open-source reasoning models such as the Qwen series68% similarUnverifiedQwen3采用四阶段训练流程，在后期专门针对混合推理能力进行对齐微调66% similarVerifiedQLoRA quantizes the base model to 4-bit precision for storage, while applying LoRA adapters on top of the quantized model for training.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/895911API
curl https://kongchang.com/api/v1/knowledge/claims/895911MCP
get_claim(id=895911)