Unverified50% confidenceFactExact time
Needle模型文件仅约100MB，推理时内存占用在数十MB量级，可存储于智能手表闪存
1
Sources
50%
Confidence
Long-term
Relevance
7/6/2026
First Seen
Sources
Needle: Fine-Tuning a 26M Ultralight Tool-Calling Model Locally — Full Walkthrough
bilibili薛定猫AI7/3/2026
Related Claims
UnverifiedGenerated LoRA files are typically only tens to hundreds of MB.69% similarVerifiedTypical installation packages for Electron applications often exceed 100 MB, and runtime memory usage exceeds 200 MB68% similarUnverified对于数十亿参数的SD模型，LoRA文件通常只有10-200MB67% similarUnverifiedEven after quantization, 70B-level models often require more than 40 GB of VRAM or a multi-GPU configuration.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/114483API
curl https://kongchang.com/api/v1/knowledge/claims/114483MCP
get_claim(id=114483)