Unverified50% confidenceFactExact time
CFC基准每个变体运行3次（285÷95=3），通过多次重复运行评估模型输出的稳定性和可复现性
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
8/24/2026
First Seen
Valid until: 11/22/2026
Sources
Related Entities
Related Claims
UnverifiedCFC基准的历史序列涵盖V1至V100，最终发布的可复现测试集包含95个变体、共计285次重复运行65% similarUnverifiedIn mixed-precision training, each FP16 parameter occupies 2 bytes of GPU memory, and each FP32 parameter occupies 4 bytes of GPU memory.60% 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.57% similarUnverifiedThe default CFG value for most models is between 7 and 12; for scenes with rich material representation, a moderate CFG value (7–9) typically achieves the best balance between fidelity and naturalness.56% similarUnverifiedQwen3 27B achieved a 100% pass rate in the 256K long-context memory retrieval test (needle in a haystack), with all 15 runs being accurate.56% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/795252API
curl https://kongchang.com/api/v1/knowledge/claims/795252MCP
get_claim(id=795252)