Unverified50% confidenceFactExact time
W4A4 表示权重量化为 4-bit、激活值也量化为 4-bit，比 W4A16 方案更激进
1
Sources
50%
Confidence
Long-term
Relevance
7/12/2026
First Seen
Sources
INT4 ConvRot W4A4 Quantization: Running Large Image Models on Low-VRAM GPUs in ComfyUI
redditr/comfyui7/11/2026
Related Claims
UnverifiedOllama uses quantization by default to compress model size, converting model weights from 32-bit floating point (FP32) to 4-bit or 8-bit integers.63% similarUnverifiedA 70B model can be compressed to approximately 35GB at INT4 precision.63% similarUnverified4bit quantization compresses model size by approximately 75% compared to 16bit precision62% similarVerifiedThe Qwen2.5:7B model is approximately 4 GB in size after Q4 quantization.62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/489935API
curl https://kongchang.com/api/v1/knowledge/claims/489935MCP
get_claim(id=489935)