Unverified50% confidenceBenchmarkExact time
On the CEHRI 60-question domain exam, CRN v2 corrected 53.3% of the base model's errors, and 43.3% on paraphrase variants.
1
Sources
50%
Confidence
Long-term
Relevance
9/16/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedCRN v2 has approximately 34 million trainable parameters, equivalent to 0.73% of the 4.65 billion-parameter text module.58% similarUnverifiedThe fine-tuned Qwen3.5-4B model improved accuracy on BU Bench V1 from 15% to 53%.58% similarUnverifiedCFC基准中V1–V4无法恢复已被排除，V6是早期不稳定基线无法精确聚合，99.3%是在可重建可评分的子集上得出的58% similarUnverifiedA SWE-bench score of 63.2 means the model can correctly solve approximately 63.2% of real software engineering problems.57% similarUnverifiedIn tests on 2,456 held-out prompts, increasing the sampling steps from 20 to 50 improved FID only from 27.0 to 26.6 (a gain of 0.4), while removing the shift caused FID to rise from 27.0 to 27.3 and FD-DINOv2 to rise from 218 to 228.55% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/924665API
curl https://kongchang.com/api/v1/knowledge/claims/924665MCP
get_claim(id=924665)