vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2025-09-10 17:14:36 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Merge branch 'master' into concedo

Browse source 
# Conflicts:
#	.devops/full.Dockerfile
#	Makefile
#	flake.nix
This commit is contained in:
Concedo 2023-04-15 11:33:43 +08:00
commit d00b865eb1
6 changed files with 305 additions and 63 deletions
Download patch file Download diff file Expand all files Collapse all files

14
llama.cpp
View file

@ -269,16 +269,6 @@ static std::string llama_format_tensor_shape(const std::vector<uint32_t> & ne) {
return ret;
}
static const char * llama_format_type(enum ggml_type type) {
switch (type) {
case GGML_TYPE_F32: return "f32";
case GGML_TYPE_F16: return "f16";
case GGML_TYPE_Q4_0: return "q4_0";
case GGML_TYPE_Q4_1: return "q4_1";
default: LLAMA_ASSERT(false);
}
}
static size_t llama_calc_tensor_size(const std::vector<uint32_t> & ne, enum ggml_type type) {
size_t size = ggml_type_size(type);
for (uint32_t dim : ne) {
@ -1589,7 +1579,7 @@ static void llama_model_quantize_internal(const std::string & fname_inp, const s
printf("[%zu/%zu] %36s - %s, type = %6s, ",
++idx, model_loader->tensors_map.tensors.size(),
tensor.name.c_str(), llama_format_tensor_shape(tensor.ne).c_str(),
llama_format_type(tensor.type));
ggml_type_name(tensor.type));
// This used to be a regex, but <regex> has an extreme cost to compile times.
bool quantize = tensor.name.rfind("weight") == tensor.name.size() - 6; // ends with 'weight'?
@ -1622,7 +1612,7 @@ static void llama_model_quantize_internal(const std::string & fname_inp, const s
f32_data[i] = ggml_fp16_to_fp32(f16_data[i]);
}
} else {
throw format("type %s unsupported for integer quantization", llama_format_type(tensor.type));
throw format("type %s unsupported for integer quantization", ggml_type_name(tensor.type));
}
printf("quantizing .. ");