llama.cpp
mtmd: refactor preprocessing + support max/min pixels
#16878

Merged

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

ngxson merged 16 commits into ggml-org:master from ngxson:xsn/mtmd-max-min-pixels

mtmd: refactor preprocessing + support max/min pixels

c1b189ae

github-actions added examples

fix mlp type

13cd2045

implement mix/max pixels

66d5c43d

improve hparams

c53c5663

better image preproc for qwen

68b1507a

fix

7bd1a011

fix out of bound composite

2892e0fb

fix (2)

42744178

fix token calculation

000d1d9d

get_merge_kernel_size()

bfd03fb6

fix llama4 and lfm2

2c0d9607

gonna fix them all

4621d99d

ngxson marked this pull request as ready for review 136 days ago

ngxson requested a review from

copilot-pull-request-reviewer 136 days ago

ngxson requested a review from

ggerganov 136 days ago

copilot-pull-request-reviewer commented on 2025-10-31

ggerganov commented on 2025-10-31

use simple resize for qwen

bae84d44

qwen: increase min tokens

29c72688

no resize if dst size == src size

00ee52e7

restore to initial min/max tokens value for qwen

a834e1ce

ggerganov approved these changes on 2025-11-01

ngxson merged cf659bbb into master 135 days ago

Reviewers

ggerganov

copilot-pull-request-reviewer

Assignees

No one assigned

Labels

examples

Milestone

No milestone

llama.cpp mtmd: refactor preprocessing + support max/min pixels #16878 Merged

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

llama.cpp
mtmd: refactor preprocessing + support max/min pixels
#16878

Merged