Unverified50% confidenceBenchmarkExact time
On four models from the Olmo and Qwen series, after distilling 6 selected skills, three of the models match or exceed the effect of distilling the complete skill library on the average avg@12.
1
Sources
50%
Confidence
Long-term
Relevance
10/13/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe researchers used Edu-QuRater scores as a reward term for GRPO post-training, with Qwen3-4B as the base model.67% similarUnverifiedThe course demonstration uses locally deployed Qwen3 series small models (approximately 9B parameters), served via Ollama.64% similarUnverifiedGlimmer在智能体代理能力、工具调用、长程规划方面明显优于同规模的其他本地模型，而Qwen在纯编程、逻辑推理、操作系统级任务方面更胜一筹62% similarUnverifiedThe Jeff project distilled and fine-tuned two base models, Qwen3.5 and Gemma 4, into decision models with 0.8B and 2B parameters respectively.61% 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60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1013304API
curl https://kongchang.com/api/v1/knowledge/claims/1013304MCP
get_claim(id=1013304)