Expired50% confidenceFactExact time
The tester's initial configuration was CFG 1, Steps 40, resolution 2MP, aspect ratio 4:3, sampler Euler, scheduler Simple, with the model and text encoder using int8 quantization and the VAE at bf16 precision.
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
9/20/2026
First Seen
Valid until: 10/4/2026(expired)
Sources
Related Entities
Related Claims
UnverifiedThe tester found that raising CFG from 1 to 4–4.5 yielded some improvement in output image quality.73% similarUnverifiedThe MiniCPM series achieves performance close to 7B and even 13B models on multiple benchmarks with a parameter count at the 2B–4B level67% similarVerifiedCompared to FP16, BF16 has the same 8-bit exponent width as FP32 and therefore a larger dynamic range, making it the mainstream precision choice for current large model training.66% similarUnverifiedAfter testing 30 layer combinations, the best pairing was SenseNova L32 mapped to MiniMax H3 L49, using a 4096→128→5120 projector that achieved a cosine similarity of 0.9042 on the validation set.66% similarUnverified实测参数建议：测试阶段用480P分辨率+8步速度最快但有明显瑕疵；生产阶段建议768标准或480P、步数不低于24、性价比约30步、正式产出建议50步64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/937393API
curl https://kongchang.com/api/v1/knowledge/claims/937393MCP
get_claim(id=937393)