llama.cpp
Add Gemma3n multimodal support with MobileNetV5 vision encoder
#18256
Merged
Merged
Add Gemma3n multimodal support with MobileNetV5 vision encoder #18256
ngxson
merged
22 commits into
ggml-org:master
from
simrnsingh:feat-gemma3n-vision
Add Gemma3nVisionModel - MobileNetV5 vision encoder convertor to conv…
3e4c8f8f
Add mobilenetv5 impl
ad5ed98d
Fix comments, remove unused vars
f5770547
Fix permute and remove transpose of projection weights
4589d3eb
Merge branch 'master' into feat-gemma3n-vision
28d39cb1
Fix comments, remove debugging prints from hf_to_gguf
47423a29
1. Hard-code image_mean = 0 and image_std = 1
67801e5b
1. Move mobilenetv5 helpers declarations to `clip_graph_mobilenetv5` …
04947c7f
Remove obsolete comments
86618c7c
- convert_hf_to_gguf.py & constants.py & tensor_mapping.py: Use expli…
e2835e9f
- Rename tensors to v.conv..., v.blk..., v.msfa... to better align wi…
632e29f5
Fix stem conv bias name
d37c22b2
ngxson
commented
on 2025-12-26
Remove explicit handling of bias term for stem conv
58667f50
- Change order of addition in "project_per_layer_inputs" to support b…
47b7dd13
Merge branch 'master' into feat-gemma3n-vision
465e888c
clean up conversion script
eea58817
fix code style
bfbb3158
also preserve audio tensors
395d2d41
trailing space
6a68b35e
split arch A and V
e842b931
rm unused gemma3 func
8f6dbbe4
fix alignment
60c23c9a
ngxson
approved these changes
on 2026-01-09
ggerganov
commented
on 2026-01-10
Reviewers
Assignees
Labels
model
examples
python
Milestone
No milestone
Login to write a write a comment.
Login via GitHub