Found 1 document(s)
Page 1 of 1
-
Weight-only quantization keeps LLM activations in floating point and stores weights as low-bit integers, turning the dominant workload into floating-point-integer (FP-INT) general matrix multiplication. Recent quantizers...