Public/koboldcpp
1
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-09-19 01:05:09 +02:00
Code Issues Packages Projects Releases Wiki Activity

handle glm4.7 flash template

Browse Source
This commit is contained in:
Concedo
2026-01-28 23:29:08 +08:00
parent ef7fe1b5d4
commit 226c79338f
3 changed files with 8 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
gpttype_adapter.cpp
+3 -2
View File
@@ -2758,7 +2758,7 @@ ModelLoadResult gpttype_load_model(const load_model_inputs inputs, FileFormat in
printf("\nThis architecture has explicitly disabled the BOS token - if you need it, you must add it manually.\n");
add_bos_token = false;
}
if (file_format == FileFormat::GGUF_GENERIC && file_format_meta.model_architecture == GGUFArch::ARCH_GLM4) {
if (file_format == FileFormat::GGUF_GENERIC && (file_format_meta.model_architecture == GGUFArch::ARCH_GLM4 || file_format_meta.model_architecture == GGUFArch::ARCH_DEEPSEEK2)) {
std::string temp = gpttype_get_chat_template();
if (temp.find("[gMASK]<sop>") != std::string::npos) {
printf("GLM-4 will have no automatic BOS token.\n");
@@ -3728,7 +3728,8 @@ generation_outputs gpttype_generate(const generation_inputs inputs)
//need to add a cursed hack to improve coherency for GLM4, by ensuring injection for gmask, sop and an extra space
//any complaints please direct them to henky
if (file_format == FileFormat::GGUF_GENERIC && file_format_meta.model_architecture == GGUFArch::ARCH_GLM4) {
//deepseek2 is actually used for glm 4.7 flash
if (file_format == FileFormat::GGUF_GENERIC && (file_format_meta.model_architecture == GGUFArch::ARCH_GLM4 || file_format_meta.model_architecture == GGUFArch::ARCH_DEEPSEEK2)) {
std::string temp = gpttype_get_chat_template();
if (temp.find("[gMASK]<sop>") != std::string::npos) {
if (addedmemory == "") {