Updated Tensor Encoding Schemes (markdown)

Brian 2024-05-18 16:14:21 +10:00
parent 3672234405
commit 2289ec20c9

@ -56,3 +56,43 @@ This is not definitive, but is helpful when reading sourcecode or console output
| IQ4_XS | GGML_FTYPE_MOSTLY_IQ4_XS | GGML_TYPE_IQ4_XS | 4.25 | i-quantization | Superblocks has 8 blocks ( 32 weights per block) | w = func(superblock_scale, importance_matrix) | [llama.cpp PR: IQ4_XS: a 4.25 bpw quantization #5747](https://github.com/ggerganov/llama.cpp/pull/5747) |
* All superblocks have fp16 scaling factor and contains up to 256 weights. Number of weights in a block must be divisible by 256. (To be confirmed)
## Where to find the structure of these tensors in the code?
You would find it all usually in `ggml-common.h` where it typically be of this form
### Blocks
```c
#define QK4_0 32
typedef struct {
ggml_half d; // delta
uint8_t qs[QK4_0 / 2]; // nibbles / quants
} block_q4_0;
static_assert(sizeof(block_q4_0) == sizeof(ggml_half) + QK4_0 / 2, "wrong q4_0 block size/padding");
```
### Superblocks
```c
//
// Super-block quantization structures
//
// 2-bit quantization
// weight is represented as x = a * q + b
// 16 blocks of 16 elements each
// Effectively 2.625 bits per weight
typedef struct {
uint8_t scales[QK_K/16]; // scales and mins, quantized with 4 bits
uint8_t qs[QK_K/4]; // quants
union {
struct {
ggml_half d; // super-block scale for quantized scales
ggml_half dmin; // super-block scale for quantized mins
} GGML_COMMON_AGGR;
ggml_half2 dm;
};
} block_q2_K;
static_assert(sizeof(block_q2_K) == 2*sizeof(ggml_half) + QK_K/16 + QK_K/4, "wrong q2_K block size/padding");
```