vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2025-09-15 03:19:41 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Merge branch 'master' into concedo

Browse source 
# Conflicts:
#	.github/workflows/build.yml
#	.github/workflows/docker.yml
#	CMakeLists.txt
#	Makefile
#	README.md
This commit is contained in:
Concedo 2023-04-19 16:34:51 +08:00
commit f662a9a230
10 changed files with 863 additions and 95 deletions
Download patch file Download diff file Expand all files Collapse all files

10
llama.cpp
View file

@ -485,6 +485,7 @@ struct llama_file_loader {
case GGML_TYPE_F16:
case GGML_TYPE_Q4_0:
case GGML_TYPE_Q4_1:
case GGML_TYPE_Q4_2:
break;
default: {
throw format("unrecognized tensor type %u\n", shard.type);
@ -557,6 +558,7 @@ struct llama_file_saver {
case GGML_TYPE_F16:
case GGML_TYPE_Q4_0:
case GGML_TYPE_Q4_1:
case GGML_TYPE_Q4_2:
break;
default: LLAMA_ASSERT(false);
}
@ -845,6 +847,7 @@ static const char *llama_ftype_name(enum llama_ftype ftype) {
case LLAMA_FTYPE_MOSTLY_Q4_1: return "mostly Q4_1";
case LLAMA_FTYPE_MOSTLY_Q4_1_SOME_F16:
return "mostly Q4_1, some F16";
case LLAMA_FTYPE_MOSTLY_Q4_2: return "mostly Q4_2";
default: return "unknown, may not work";
}
}
@ -1578,6 +1581,7 @@ static void llama_model_quantize_internal(const std::string & fname_inp, const s
switch (ftype) {
case LLAMA_FTYPE_MOSTLY_Q4_0: quantized_type = GGML_TYPE_Q4_0; break;
case LLAMA_FTYPE_MOSTLY_Q4_1: quantized_type = GGML_TYPE_Q4_1; break;
case LLAMA_FTYPE_MOSTLY_Q4_2: quantized_type = GGML_TYPE_Q4_2; break;
default: throw format("invalid output file type %d\n", ftype);
};
@ -1651,6 +1655,10 @@ static void llama_model_quantize_internal(const std::string & fname_inp, const s
{
new_size = ggml_quantize_q4_1(f32_data, new_data, nelements, (int) tensor.ne.at(0), hist_cur.data());
} break;
case GGML_TYPE_Q4_2:
{
new_size = ggml_quantize_q4_2(f32_data, new_data, nelements, (int) tensor.ne.at(0), hist_cur.data());
} break;
default:
LLAMA_ASSERT(false);
}
@ -1962,7 +1970,7 @@ int llama_apply_lora_from_file_internal(struct llama_context * ctx, const char *
base_t = dest_t;
}
if (base_t->type == GGML_TYPE_Q4_0 || base_t->type == GGML_TYPE_Q4_1) {
if (base_t->type == GGML_TYPE_Q4_0 || base_t->type == GGML_TYPE_Q4_1 || base_t->type == GGML_TYPE_Q4_2) {
if (!warned) {
fprintf(stderr, "%s: warning: using a lora adapter with a quantized model may result in poor quality, "
"use a f16 or f32 base model with --lora-base\n", __func__);