thek0tyara/llama-cpp-turboquant
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 4

mtmd: refactor preprocessing + support max/min pixels (#16878)

Browse source 
* mtmd: refactor preprocessing + support max/min pixels

* fix mlp type

* implement mix/max pixels

* improve hparams

* better image preproc for qwen

* fix

* fix out of bound composite

* fix (2)

* fix token calculation

* get_merge_kernel_size()

* fix llama4 and lfm2

* gonna fix them all

* use simple resize for qwen

* qwen: increase min tokens

* no resize if dst size == src size

* restore to initial min/max tokens value for qwen
This commit is contained in:
Xuan-Son Nguyen 2025-11-01 15:51:36 +01:00 • committed by GitHub
parent d8b860a219
commit cf659bbb8e
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
2 changed files with 432 additions and 332 deletions
Download patch file Download diff file Expand all files Collapse all files

2
tools/mtmd/clip-impl.h
View file

@ -154,8 +154,8 @@ enum projector_type {
PROJECTOR_TYPE_LFM2,
PROJECTOR_TYPE_KIMIVL,
PROJECTOR_TYPE_LIGHTONOCR,
PROJECTOR_TYPE_UNKNOWN,
PROJECTOR_TYPE_COGVLM,
PROJECTOR_TYPE_UNKNOWN,
};
static std::map<projector_type, std::string> PROJECTOR_TYPE_NAMES = {