mirror of
https://github.com/LostRuins/koboldcpp.git
synced 2026-09-18 16:55:14 +02:00
unclamp glm4 in debug
This commit is contained in:
+8
-2
@@ -1908,8 +1908,14 @@ ModelLoadResult gpttype_load_model(const load_model_inputs inputs, FileFormat in
|
||||
clamped_max_context_length = 16384;
|
||||
}
|
||||
if (isGguf && file_format_meta.model_architecture == GGUFArch::ARCH_GLM4 && kcpp_data->n_batch > 16) {
|
||||
printf("GLM-4 is broken on larger batch sizes. Clamping batch size to 16.\n");
|
||||
kcpp_data->n_batch = kcpp_data->n_ubatch = 16;
|
||||
if(debugmode==1)
|
||||
{
|
||||
printf("GLM-4 is broken on larger batch sizes. Clamp ignored in debug.\n");
|
||||
} else {
|
||||
printf("GLM-4 is broken on larger batch sizes. Clamping batch size to 16.\n");
|
||||
kcpp_data->n_batch = kcpp_data->n_ubatch = 16;
|
||||
}
|
||||
|
||||
}
|
||||
|
||||
kcpp_data->n_ctx = clamped_max_context_length;
|
||||
|
||||
Reference in New Issue
Block a user