Unverified50% confidenceFactExact time
Whisper原始FP32版本的Tiny模型约150MB，经INT8量化后可压缩至约40MB，在树莓派4B上可实现实时转写
1
Sources
50%
Confidence
Long-term
Relevance
7/21/2026
First Seen
Sources
Related Claims
UnverifiedWhisper large-v3 从 FP16 量化至 INT8 后显存需求约从10GB降至5GB，使其可在消费级8GB显卡上运行77% similarUnverified140GB的FP16 70B模型经4-bit量化后可压缩至约35-40GB，通信开销降低约75%；2-bit量化可压缩至约18GB但带来精度损失74% similarUnverifiedWhisper small模型约有244M参数，在消费级CPU上即可流畅运行73% similarUnverified原始FP16格式的视频生成模型可能需要20GB以上显存，经Q4量化后可压缩至5GB以内72% similarVerified32B参数规模的模型在量化后通常可以压缩到约16-20GB显存占用，一张NVIDIA RTX 4090（24GB显存）就能运行72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/582172API
curl https://kongchang.com/api/v1/knowledge/claims/582172MCP
get_claim(id=582172)