llama.cpp
Add dynamic high-resolution image preprocessing for InternVL model
#20847
Merged
Merged
Add dynamic high-resolution image preprocessing for InternVL model #20847
ngxson
merged
13 commits into
ggml-org:master
from
bssrdf:dynamic-resolution-internvl
added support for internvl's dynamic high-resolution (Qianfan-OCR nee…
bd4d9eea
add min/max dynamic patch to gguf meta
d55ce51d
clean up
0c47a82b
ngxson
requested changes
on 2026-03-21
simplified handling min/max dynamic patch
7e08c7f2
ngxson
requested changes
on 2026-03-22
ngxson
commented
on 2026-03-22
ngxson
commented
on 2026-03-22
ngxson
commented
on 2026-03-22
ngxson
commented
on 2026-03-22
ngxson
commented
on 2026-03-22
reuse llava_uhd logic for slice images
ae1099a7
provide default values for older models
2cf9a263
flake8
75c80718
ngxson
approved these changes
on 2026-03-22
prevent writing 0 value to gguf
8e660323
remove duplicated resolution candidates with a better algorithm
b0b435d5
ngxson
commented
on 2026-03-22
fix indentation
109ad435
format
c42b5e7d
add protection from divide by zero
b260e69b
change to 0 to be safe
5f675479
ngxson
approved these changes
on 2026-03-22
Login to write a write a comment.
Login via GitHub