Merge branch 'upstream' into concedo_experimental

# Conflicts:
#	.github/workflows/winget.yml
#	CMakeLists.txt
#	common/CMakeLists.txt
#	examples/model-conversion/scripts/causal/run-org-model.py
#	ggml/src/ggml-cpu/CMakeLists.txt
#	tools/perplexity/perplexity.cpp
#	tools/server/CMakeLists.txt
This commit is contained in:
Concedo
2026-02-17 19:41:28 +08:00
13 changed files with 80 additions and 54 deletions
+4 -4
View File
@@ -667,7 +667,7 @@ static std::vector<T> string_split(const std::string & str, char delim) {
}
template<>
std::vector<std::string> string_split<std::string>(const std::string & input, char separator)
inline std::vector<std::string> string_split<std::string>(const std::string & input, char separator)
{
std::vector<std::string> parts;
size_t begin_pos = 0;
@@ -682,7 +682,7 @@ std::vector<std::string> string_split<std::string>(const std::string & input, ch
return parts;
}
static bool string_starts_with(const std::string & str,
inline bool string_starts_with(const std::string & str,
const std::string & prefix) { // While we wait for C++20's std::string::starts_with...
return str.rfind(prefix, 0) == 0;
}
@@ -867,11 +867,11 @@ const char * const LLM_KV_SPLIT_TENSORS_COUNT = "split.tensors.count";
const char * const LLM_FFN_EXPS_REGEX = "\\.ffn_(up|down|gate)_(ch|)exps";
static std::string llm_ffn_exps_block_regex(int idx) {
inline std::string llm_ffn_exps_block_regex(int idx) {
return string_format("blk\\.%d%s", idx, LLM_FFN_EXPS_REGEX);
}
static llama_model_tensor_buft_override llm_ffn_exps_cpu_override() {
inline llama_model_tensor_buft_override llm_ffn_exps_cpu_override() {
return { LLM_FFN_EXPS_REGEX, ggml_backend_cpu_buffer_type() };
}