|
ember.cpp 0.1.0
C++ inference engine for Llama-family and paperkiln models (CPU, CUDA, WebAssembly)
|
3-bit K-quantized block structure More...
#include <quantization.h>

Public Attributes | |
| uint8_t | hmask [GGML_QK_K/8] |
| uint8_t | qs [GGML_QK_K/4] |
| uint8_t | scales [12] |
| uint16_t | d |
| uint16_t | dmin |
3-bit K-quantized block structure
Stores weights quantized to 3 bits with block-wise scaling. Balances compression and precision between Q2_K and Q4_K.
Definition at line 116 of file quantization.h.
| uint16_t block_q3_K::d |
Block scale
Definition at line 120 of file quantization.h.
| uint16_t block_q3_K::dmin |
Block minimum value
Definition at line 121 of file quantization.h.
High bit masks
Definition at line 117 of file quantization.h.
Quantized values
Definition at line 118 of file quantization.h.
| uint8_t block_q3_K::scales[12] |
Sub-block scales
Definition at line 119 of file quantization.h.