Unverified50% confidenceFactExact time
The research team trained Lafa at three scales—0.5B, 7B, and 8B—each using fewer than 30B tokens.
1
Sources
50%
Confidence
Long-term
Relevance
10/21/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedLoRA使可训练参数量通常仅为全参微调的0.1%-1%，可在单张RTX 3090/4090（显存24GB）上微调7B乃至13B级别的模型69% similarUnverifiedFor personal learning or local testing, it is recommended to use smaller models (such as 1B, 2.5B, or 3.5B).69% similarUnverified一个rank=4的LoRA文件对于SD1.5模型约只有3-6MB，LoRA设计使训练参数量减少95%以上68% similarUnverifiedSmall models of the same scale such as Phi-3-mini, Gemma-2B, and Qwen1.5-1.8B, after training on high-quality data, can achieve performance close to larger models on specific tasks.65% 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、12865% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1078859API
curl https://kongchang.com/api/v1/knowledge/claims/1078859MCP
get_claim(id=1078859)