AIThe Registerabout 3 hours ago
Changing AI math could reduce the hardware burden, researchers show
5 min read
SEMQ separates semantics from embeddings to reduce AI model memory needs without quantization tradeoffs. A 7B parameter model at FP32 requires 28 GB. SEMQ decouples meaning from representation, preserving relational properties while cutting resource requirements.
Level
Hype check
Tap to vote and see what everyone thinks.
