FP4

What is FP4?

A 4-bit floating-point number format used for highly compressed low-precision AI computation

In other languages

한국어FP4

4비트 부동소수점 수 형식으로, 매우 낮은 정밀도로 AI 추론과 학습 효율을 높이는 방식이다

日本語FP4

4ビット浮動小数点形式。AIモデルの計算や保存をさらに軽くするための低精度数値表現

Related Terms

• FP4 quantizationA model compression technique that reduces numerical precision to 4 bits to accelerate inference and decrease memory usage
• MXFP4Microscaling 4-bit floating-point format designed to reduce memory use while preserving numerical quality in large model inference
• FP8An 8-bit floating-point format that allows models to train and run faster with lower memory requirements.
• INT4Four-bit integer numeric format that stores values in 16 possible levels, often used to reduce model memory size
• FP8 quantizationA technique that reduces the precision of model weights and activations to 8-bit floating point format to decrease memory usage and accelerate inference
• FP32A computing standard for representing real numbers using 32 bits, widely used in deep learning for numerical calculations where high precision is required.
• MXFP8Microscaling 8-bit floating-point format that uses shared local scaling blocks for lower-precision tensor computation
• BF16Brain Floating Point 16 (bfloat16); a 16-bit floating-point format commonly used for efficient AI training and inference.
• 4-bit quantizationModel compression technique that stores numerical values using 4 bits instead of higher precision formats
• Floating-point arithmeticA method for representing and calculating with real numbers using finite precision, which can produce rounding differences.
• Floating-point operationAn arithmetic calculation involving numbers with fractional parts, used as a fundamental measure of computational effort in computer science.
• FLOPsFloating-point operations per second, a measure of an AI system's computational power and processing scale