Unverified50% confidenceSolutionExact time
The test used a three-image comparison workflow, running the native 25-step, 4-step LoRA and 8-step LoRA versions simultaneously; operationally, one only needs to load the corresponding LoRA after the model, change CFG to 1, set the steps to 4 or 8, and use Euler simple uniformly as the sampler.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/24/2026
First Seen
Valid until: 1/22/2027
Sources
Related Entities
Related Claims
Unverified2.5-24 took first place in all three instruction-following tests.69% similarUnverifiedThe Rocket Launcher team's tests were based on Claude Code version 2.1.293, with the environment reset before each run and each test run three times with the median taken.66% similarUnverifiedThe limitations of this test include: limited task scale (50 tasks, 3 trials per prompt), style versions written by a single author, length confounding between the florid and normal versions, testing only English and local models below 14B, and the perplexity check covering only three smaller Qwen models and using full-precision weights while actual inference used 4-bit quantization.66% similarUnverifiedThe complete workflow parameters are KREA 2 Turbo, euler/simple sampling, 8 steps, CFG 1, 1024×1024 resolution, and seed 1997.65% 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64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1092258API
curl https://kongchang.com/api/v1/knowledge/claims/1092258MCP
get_claim(id=1092258)