Unverified50% confidenceFactExact time
The author's fine-tuning used the QLoRA method, with LoRA rank/alpha of 32/64, a learning rate of 1e-4, batch size 1 with gradient accumulation of 16, AdamW 8-bit optimizer, and training for 1 epoch.
1
Sources
50%
Confidence
Long-term
Relevance
10/11/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe author's publicly shared standard training configuration is: Krea 2 Standard model, Standard LoRA, Network rank 32, Alpha 32, Epochs 64 (saving every 4 epochs), Batch size 1, target resolution 0.25 MP, EMA 0.98.74% similarUnverifiedLoRA训练中常用的学习率为1e-4到5e-4之间，常用的Network Rank值为8、16、32、64、12870% similarUnverified在混合精度Adam训练中，单个参数的完整训练状态高达18字节（FP16权重2字节、FP32主权重4字节、一阶矩4字节、二阶矩4字节，加梯度）69% similarUnverifiedOn the AI Builder model intelligence leaderboard, Claude Opus 5.5 Max topped the rankings with 58 points, Xiaomi MiMo V2.6 Pro ranked 9th with 46 points, and Qwen 3.8 Max ranked 10th with 45 points.66% similarUnverifiedOn the Artificial Analysis Intelligence Index, Qwen3.8 27b scored 34, Qwen3.6 27b scored 21, Gemma 4 26b scored 17, GLM 4.7 Flash scored 15, and Qwen3 Coder Next scored only 9.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/999485API
curl https://kongchang.com/api/v1/knowledge/claims/999485MCP
get_claim(id=999485)