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

iq3_s: make tests pass

Browse source
This commit is contained in:
Iwan Kawrakow 2024-02-22 18:14:41 +02:00
parent 2730225c5f
commit 47cf30b0ee
2 changed files with 4 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

2
tests/test-backend-ops.cpp
View file

@ -1918,7 +1918,7 @@ static bool test_backend(ggml_backend_t backend, test_mode mode, const char * op
GGML_TYPE_Q6_K,
GGML_TYPE_IQ2_XXS, GGML_TYPE_IQ2_XS,
GGML_TYPE_IQ3_XXS, GGML_TYPE_IQ1_S,
GGML_TYPE_IQ4_NL,
GGML_TYPE_IQ4_NL, GGML_TYPE_IQ3_S,
};
// unary ops

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

@ -151,6 +151,7 @@ int main(int argc, char * argv[]) {
const float max_quantization_error =
type == GGML_TYPE_Q2_K ? 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;
failed = !(total_error < max_quantization_error);
num_failed += failed;
@ -167,7 +168,8 @@ 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 ? MAX_DOT_PRODUCT_ERROR_LOWBIT : MAX_DOT_PRODUCT_ERROR;
type == GGML_TYPE_IQ3_XXS || type == GGML_TYPE_IQ3_S ? MAX_DOT_PRODUCT_ERROR_LOWBIT
: MAX_DOT_PRODUCT_ERROR;
failed = !(vec_dot_error < max_allowed_error);
num_failed += failed;
if (failed || verbose) {