vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Allow "quantizing" to f16 and f32 (#1787)

Browse source 
* Allow "quantizing" to f16 and f32

Fix an issue where quantizing didn't respect LLAMA_NO_K_QUANTS

Add brief help to the list of quantization types in the quantize tool

Ignore case for quantization type arguments in the quantize tool
This commit is contained in:
Kerfuffle 2023-06-13 04:23:23 -06:00 • committed by GitHub
parent 74a6d922f1
commit 74d4cfa343
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
4 changed files with 154 additions and 48 deletions
Download patch file Download diff file Expand all files Collapse all files

12
ggml.c
View file

@ -16301,6 +16301,18 @@ size_t ggml_quantize_chunk(enum ggml_type type, const float * src, void * dst, i
result = ggml_quantize_q6_K(src + start, block, n, n, hist);
} break;
#endif
case GGML_TYPE_F16:
{
int elemsize = sizeof(ggml_fp16_t);
ggml_fp32_to_fp16_row(src + start, (ggml_fp16_t *)dst + start, n);
result = n * elemsize;
} break;
case GGML_TYPE_F32:
{
int elemsize = sizeof(float);
result = n * elemsize;
memcpy((uint8_t *)dst + start * elemsize, src + start, result);
} break;
default:
assert(false);
}