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

gemma3 vision works, but is using more tokens than expected - may need resizing

Browse Source
This commit is contained in:
Concedo
2025-03-13 00:31:16 +08:00
parent eb1809c105 80a02aa858
commit 77debb1b1b
39 changed files with 2140 additions and 814 deletions
Download Patch File Download Diff File Expand all files Collapse all files
gpttype_adapter.cpp
+6 -2
View File
@@ -2876,11 +2876,15 @@ generation_outputs gpttype_generate(const generation_inputs inputs)
bool llava_images_changed = false;
bool add_bos_token = true;
if(file_format == FileFormat::GGUF_GENERIC)
bool add_bos_token = true; //if set to false, mmproj handling breaks
if(file_format == FileFormat::GGUF_GENERIC && mmproj_filename == "")
{
const llama_vocab * tmpvocab = llama_model_get_vocab(&(llama_ctx_v4->model));
add_bos_token = llama_vocab_get_add_bos(tmpvocab);
if(!add_bos_token)
{
printf("\nBOS token prefix was disabled for this model.");
}
}
for(int x=0;x<inputs.stop_sequence_len;++x)