happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Move test-only stuff out of llama-quant.cpp

This commit is contained in:
Colin Kealty 2026-03-12 14:06:23 -04:00
parent 3fe55f1035
commit 8ebfe03f95
3 changed files with 63 additions and 63 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

58
src/llama-quant.cpp
View File

@ -755,64 +755,6 @@ ggml_type llama_ftype_get_default_type(llama_ftype ftype) {
} }
} }
struct ftype_name_entry {
const char * name;
llama_ftype ftype;
};
static const ftype_name_entry ftype_name_table[] = {
{ "F32", LLAMA_FTYPE_ALL_F32 },
{ "F16", LLAMA_FTYPE_MOSTLY_F16 },
{ "BF16", LLAMA_FTYPE_MOSTLY_BF16 },
{ "Q4_0", LLAMA_FTYPE_MOSTLY_Q4_0 },
{ "Q4_1", LLAMA_FTYPE_MOSTLY_Q4_1 },
{ "Q5_0", LLAMA_FTYPE_MOSTLY_Q5_0 },
{ "Q5_1", LLAMA_FTYPE_MOSTLY_Q5_1 },
{ "Q8_0", LLAMA_FTYPE_MOSTLY_Q8_0 },
{ "Q2_K", LLAMA_FTYPE_MOSTLY_Q2_K },
{ "Q2_K_S", LLAMA_FTYPE_MOSTLY_Q2_K_S },
{ "Q3_K_S", LLAMA_FTYPE_MOSTLY_Q3_K_S },
{ "Q3_K_M", LLAMA_FTYPE_MOSTLY_Q3_K_M },
{ "Q3_K_L", LLAMA_FTYPE_MOSTLY_Q3_K_L },
{ "Q4_K_S", LLAMA_FTYPE_MOSTLY_Q4_K_S },
{ "Q4_K_M", LLAMA_FTYPE_MOSTLY_Q4_K_M },
{ "Q5_K_S", LLAMA_FTYPE_MOSTLY_Q5_K_S },
{ "Q5_K_M", LLAMA_FTYPE_MOSTLY_Q5_K_M },
{ "Q6_K", LLAMA_FTYPE_MOSTLY_Q6_K },
{ "IQ1_S", LLAMA_FTYPE_MOSTLY_IQ1_S },
{ "IQ1_M", LLAMA_FTYPE_MOSTLY_IQ1_M },
{ "IQ2_XXS", LLAMA_FTYPE_MOSTLY_IQ2_XXS },
{ "IQ2_XS", LLAMA_FTYPE_MOSTLY_IQ2_XS },
{ "IQ2_S", LLAMA_FTYPE_MOSTLY_IQ2_S },
{ "IQ2_M", LLAMA_FTYPE_MOSTLY_IQ2_M },
{ "IQ3_XXS", LLAMA_FTYPE_MOSTLY_IQ3_XXS },
{ "IQ3_XS", LLAMA_FTYPE_MOSTLY_IQ3_XS },
{ "IQ3_S", LLAMA_FTYPE_MOSTLY_IQ3_S },
{ "IQ3_M", LLAMA_FTYPE_MOSTLY_IQ3_M },
{ "IQ4_NL", LLAMA_FTYPE_MOSTLY_IQ4_NL },
{ "IQ4_XS", LLAMA_FTYPE_MOSTLY_IQ4_XS },
{ "TQ1_0", LLAMA_FTYPE_MOSTLY_TQ1_0 },
{ "TQ2_0", LLAMA_FTYPE_MOSTLY_TQ2_0 },
{ "MXFP4_MOE", LLAMA_FTYPE_MOSTLY_MXFP4_MOE },
};
llama_ftype llama_ftype_from_name(const char * name) {
for (const auto & e : ftype_name_table) {
if (strcmp(name, e.name) == 0) {
return e.ftype;
}
}
return (llama_ftype)-1;
}
const char * llama_ftype_to_name(llama_ftype ftype) {
for (const auto & e : ftype_name_table) {
if (e.ftype == ftype) {
return e.name;
}
}
return nullptr;
}
void init_quantize_state_counters(quantize_state_impl & qs, std::vector<tensor_metadata> & metadata) { void init_quantize_state_counters(quantize_state_impl & qs, std::vector<tensor_metadata> & metadata) {
for (auto & tm : metadata) { for (auto & tm : metadata) {

5
src/llama-quant.h
View File

@ -86,11 +86,6 @@ struct quantize_state_impl {
ggml_type llama_tensor_get_type(quantize_state_impl & qs, const llama_model_quantize_params * params, const ggml_tensor * tensor, ggml_type default_type, const tensor_metadata & tm); ggml_type llama_tensor_get_type(quantize_state_impl & qs, const llama_model_quantize_params * params, const ggml_tensor * tensor, ggml_type default_type, const tensor_metadata & tm);
ggml_type llama_ftype_get_default_type(llama_ftype ftype); ggml_type llama_ftype_get_default_type(llama_ftype ftype);
// Ftype name <-> enum conversions.
// Returns (llama_ftype)-1 on failure.
llama_ftype llama_ftype_from_name(const char * name);
const char * llama_ftype_to_name(llama_ftype ftype);
// Initialize quantize_state_impl counters and populate tensor_metadata categories. // Initialize quantize_state_impl counters and populate tensor_metadata categories.
// metadata: vector with name fields already set, will have category field populated. // metadata: vector with name fields already set, will have category field populated.
void init_quantize_state_counters(quantize_state_impl & qs, std::vector<tensor_metadata> & metadata); void init_quantize_state_counters(quantize_state_impl & qs, std::vector<tensor_metadata> & metadata);

63
tests/test-quant-type-selection.cpp
View File

@ -16,6 +16,69 @@
#include <utility> #include <utility>
#include <vector> #include <vector>
// ---------------------------------------------------------------------------
// ftype name <-> enum mapping
// ---------------------------------------------------------------------------
struct ftype_name_entry {
const char * name;
llama_ftype ftype;
};
static const ftype_name_entry ftype_name_table[] = {
{ "F32", LLAMA_FTYPE_ALL_F32 },
{ "F16", LLAMA_FTYPE_MOSTLY_F16 },
{ "BF16", LLAMA_FTYPE_MOSTLY_BF16 },
{ "Q4_0", LLAMA_FTYPE_MOSTLY_Q4_0 },
{ "Q4_1", LLAMA_FTYPE_MOSTLY_Q4_1 },
{ "Q5_0", LLAMA_FTYPE_MOSTLY_Q5_0 },
{ "Q5_1", LLAMA_FTYPE_MOSTLY_Q5_1 },
{ "Q8_0", LLAMA_FTYPE_MOSTLY_Q8_0 },
{ "Q2_K", LLAMA_FTYPE_MOSTLY_Q2_K },
{ "Q2_K_S", LLAMA_FTYPE_MOSTLY_Q2_K_S },
{ "Q3_K_S", LLAMA_FTYPE_MOSTLY_Q3_K_S },
{ "Q3_K_M", LLAMA_FTYPE_MOSTLY_Q3_K_M },
{ "Q3_K_L", LLAMA_FTYPE_MOSTLY_Q3_K_L },
{ "Q4_K_S", LLAMA_FTYPE_MOSTLY_Q4_K_S },
{ "Q4_K_M", LLAMA_FTYPE_MOSTLY_Q4_K_M },
{ "Q5_K_S", LLAMA_FTYPE_MOSTLY_Q5_K_S },
{ "Q5_K_M", LLAMA_FTYPE_MOSTLY_Q5_K_M },
{ "Q6_K", LLAMA_FTYPE_MOSTLY_Q6_K },
{ "IQ1_S", LLAMA_FTYPE_MOSTLY_IQ1_S },
{ "IQ1_M", LLAMA_FTYPE_MOSTLY_IQ1_M },
{ "IQ2_XXS", LLAMA_FTYPE_MOSTLY_IQ2_XXS },
{ "IQ2_XS", LLAMA_FTYPE_MOSTLY_IQ2_XS },
{ "IQ2_S", LLAMA_FTYPE_MOSTLY_IQ2_S },
{ "IQ2_M", LLAMA_FTYPE_MOSTLY_IQ2_M },
{ "IQ3_XXS", LLAMA_FTYPE_MOSTLY_IQ3_XXS },
{ "IQ3_XS", LLAMA_FTYPE_MOSTLY_IQ3_XS },
{ "IQ3_S", LLAMA_FTYPE_MOSTLY_IQ3_S },
{ "IQ3_M", LLAMA_FTYPE_MOSTLY_IQ3_M },
{ "IQ4_NL", LLAMA_FTYPE_MOSTLY_IQ4_NL },
{ "IQ4_XS", LLAMA_FTYPE_MOSTLY_IQ4_XS },
{ "TQ1_0", LLAMA_FTYPE_MOSTLY_TQ1_0 },
{ "TQ2_0", LLAMA_FTYPE_MOSTLY_TQ2_0 },
{ "MXFP4_MOE", LLAMA_FTYPE_MOSTLY_MXFP4_MOE },
};
static llama_ftype llama_ftype_from_name(const char * name) {
for (const auto & e : ftype_name_table) {
if (strcmp(name, e.name) == 0) {
return e.ftype;
}
}
return (llama_ftype)-1;
}
static const char * llama_ftype_to_name(llama_ftype ftype) {
for (const auto & e : ftype_name_table) {
if (e.ftype == ftype) {
return e.name;
}
}
return nullptr;
}
// --------------------------------------------------------------------------- // ---------------------------------------------------------------------------
// Mock tensor construction - may be better to extract this in the future // Mock tensor construction - may be better to extract this in the future
// --------------------------------------------------------------------------- // ---------------------------------------------------------------------------