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

Adding IQ2_S and IQ2_M to complete coverage of the 2-3 bit quantization range (#5721)

Browse source 
* Adding IQ2_S and IQ2_M as a single cumulative commit

* Update examples/quantize/quantize.cpp

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

---------

Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
This commit is contained in:
Kawrakow 2024-02-26 18:28:38 +02:00 • committed by GitHub
parent 47bb7b48c7
commit a33e6a0d2a
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
12 changed files with 1754 additions and 38 deletions
Download patch file Download diff file Expand all files Collapse all files

4
tests/test-quantize-fns.cpp
View file

@ -150,6 +150,7 @@ int main(int argc, char * argv[]) {
const float total_error = total_quantization_error(qfns, test_size, test_data.data());
const float max_quantization_error =
type == GGML_TYPE_Q2_K ? MAX_QUANTIZATION_TOTAL_ERROR_2BITS :
type == GGML_TYPE_IQ2_S ? MAX_QUANTIZATION_TOTAL_ERROR_2BITS :
type == GGML_TYPE_Q3_K ? MAX_QUANTIZATION_TOTAL_ERROR_3BITS :
type == GGML_TYPE_IQ3_S ? MAX_QUANTIZATION_TOTAL_ERROR_3BITS :
type == GGML_TYPE_IQ3_XXS ? MAX_QUANTIZATION_TOTAL_ERROR_3BITS_XXS : MAX_QUANTIZATION_TOTAL_ERROR;
@ -168,7 +169,8 @@ int main(int argc, char * argv[]) {
const float vec_dot_error = dot_product_error(qfns, test_size, test_data.data(), test_data2.data());
const float max_allowed_error = type == GGML_TYPE_Q2_K || type == GGML_TYPE_IQ2_XS || type == GGML_TYPE_IQ2_XXS ||
type == GGML_TYPE_IQ3_XXS || type == GGML_TYPE_IQ3_S ? MAX_DOT_PRODUCT_ERROR_LOWBIT
type == GGML_TYPE_IQ3_XXS || type == GGML_TYPE_IQ3_S || type == GGML_TYPE_IQ2_S
? MAX_DOT_PRODUCT_ERROR_LOWBIT
: MAX_DOT_PRODUCT_ERROR;
failed = !(vec_dot_error < max_allowed_error);
num_failed += failed;