Unverified50% confidenceBenchmarkExact time
A 4B model trained with RPL outperforms all 8B model baselines.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/20/2026
First Seen
Valid until: 1/18/2027
Sources
Related Entities
Related Claims
UnverifiedOn a contamination-free benchmark, the trained 8B model surpassed Claude Opus 4.8 and GPT-6 Astra on the creative utility metric.67% similarUnverifiedUnder the synthetic settings evaluated, DASA provides a 3.6 to 4.9× speedup compared to GRADMM, with comparable peak GPU memory usage.66% similarUnverified叠加cv=5的5折交叉验证后，144种参数组合需训练144×5=720次模型65% similarUnverified在27B混合架构、4B稠密模型、36B MoE三类模型上以8k上下文测试，Prefill阶段性能无变化，Decode阶段打补丁后快约1.4%64% similarVerifiedThe larger the number of model parameters, the greater the robustness to quantization, allowing 70B-level models to maintain performance close to FP16 after 4-bit quantization.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1074707API
curl https://kongchang.com/api/v1/knowledge/claims/1074707MCP
get_claim(id=1074707)