llama.cpp
mtmd: Add DeepSeekOCR Support
#17400
Merged
Go
Login via GitHub
Home
Pricing
FAQ
Install
Login
via GitHub
Overview
Commits
143
Changes
View On
GitHub
mtmd: Add DeepSeekOCR Support
#17400
ngxson
merged 143 commits into
ggml-org:master
from
sfallah:sf/deepseek-ocr
mtmd: llama.cpp DeepSeekOCR support
43a130b4
loading sam tensors
b6b9f02c
mtmd: fix vision model processing
85c7cda8
Merge pull request #1 from bluebread/sf/deepseek-ocr
578c8d77
deepseek-ocr clip-vit model impl
2aab52e2
mtmd: add DeepSeek-OCR LM support with standard attention
eab28ed3
mtmd: successfully runs DeepSeek-OCR LM in llama-cli
76305878
mtmd: Fix RoPE type for DeepSeek-OCR LM.
2de34367
Merge branch 'sf/deepseek-ocr' of github.com:sfallah/llama.cpp into s…
e8b26102
loading LM
97e0907c
Merge branch 'sf/deepseek-ocr' into sf/deepseek-ocr
13dc6fb3
Merge pull request #2 from bluebread/sf/deepseek-ocr
b32bb5e7
sam warmup working
790bbb97
sam erroneous return corrected
cec9a5c6
clip-vit: corrected cls_embd concat
8b3d319c
clip-vit: model convert qkv_proj split
1e081571
corrected combining of image encoders' results
331cea8f
fix: update callback for ffn_moe_weighted and add callback for attn_o…
6c0715be
Merge branch 'sf/deepseek-ocr' of github.com:sfallah/llama.cpp into s…
a65ddf5b
concat image_newline and image_seperator tokens
63a042f2
visual_model warmup (technically) works
89afda8d
window partitioning using standard ggml ops
88032f46
sfallah
requested a review
from
ngxson
348 days ago
sfallah
requested a review
from
CISC
348 days ago
sfallah
requested a review
from
ggerganov
348 days ago
github-actions
added
model
github-actions
added
examples
github-actions
added
python
sfallah
marked this pull request as draft
348 days ago
Merge branch 'sf/deepseek-ocr' of github.com:sfallah/llama.cpp into s…
1268dc3f
sam implementation without using CPU only ops
68b206b6
clip: fixed warnings
8bce66d5
Merge branch 'sf/deepseek-ocr' of github.com:sfallah/llama.cpp into s…
5e6cf3c6
mtmd: fix get_rel_pos
7e9fbecc
Merge branch 'sf/deepseek-ocr' of github.com:sfallah/llama.cpp into s…
0f5587dc
mtmd: fixed the wrong scaler for get_rel_pos
7b8d735c
image encoding technically works but the output can't be checked sing…
86f111f8
mtmd: minor changed
effe6695
Merge branch 'sf/deepseek-ocr' of github.com:sfallah/llama.cpp into s…
f8f66a15
Merge pull request #3 from bluebread/sf/deepseek-ocr
3fcfc3ac
mtmd: add native resolution support
ee8a1488
- image encoding debugged
4cfa15fc
mtmd: correct token order
3f711883
Merge pull request #5 from bluebread/dsocr-debug
a594990f
Merge branch 'sf/deepseek-ocr' into sf/deepseek-ocr
6dfda99c
Merge pull request #4 from bluebread/sf/deepseek-ocr
7941f5d8
- dynamic resizing
206f8abc
mtmd: quick fix token order
40e7e6e7
mtmd: fix danling pointer
81533e49
Merge pull request #6 from bluebread/sf/deepseek-ocr
88109404
mtmd: SAM numerically works
a488b495
mtmd: debug CLIP-L (vit_pre_ln)
ccb2f238
mtmd: debug CLIP-L & first working DeepSeek-OCR model
841a4a88
Merge remote-tracking branch 'sfallah/master' into sf/deepseek-ocr
ed3b7f10
Merge branch 'sf/deepseek-ocr' of github.com:sfallah/llama.cpp into s…
55430945
mtmd : add --dsocr-mode CLI argument for DeepSeek-OCR resolution cont…
c5f4c64f
mtmd: simplify SAM patch embedding
95239f92
Merge pull request #7 from bluebread/sf/deepseek-ocr
6b0e7cd1
github-actions
added
Nvidia GPU
github-actions
added
ggml
Merge branch 'master' into sf/deepseek-ocr
66341666
ngxson
commented on 2025-12-02
mtmd: adapt Pillow image resizing function
c914e054
mtmd: simplify DeepSeek-OCR dynamic resolution preprocessing
e20857ba
Merge branch 'sf/deepseek-ocr' of github.com:sfallah/llama.cpp into s…
43dfc0c8
bluebread
commented on 2025-12-03
mtmd: remove --dsocr-mode argument
b696c547
mtmd: refactor code & remove unused helper functions
b26b507c
sfallah
marked this pull request as ready for review
335 days ago
CISC
commented on 2025-12-04
ngxson
requested changes on 2025-12-04
mtmd: fix tensor names for image newlines and view separator
7451b841
clean up
386ba479
Merge branch 'sf/deepseek-ocr' into sf/deepseek-ocr-cleanup
c73748ab
reverting automatically removed spaces
a661c529
reverting automatically removed spaces
0399ddf1
mtmd: fixed bad ocr check in Deepseek2 (LM)
c89171cf
Merge branch 'sf/deepseek-ocr-cleanup' of github.com:sfallah/llama.cp…
2dd99240
mtmd: support combined QKV projection in buid_vit
fc3f625f
Merge pull request #8 from sfallah/sf/deepseek-ocr-cleanup
4d7d9945
using common build_attn in sam
5381b9cf
corrected code-branch when flash-attn disabled
076138a4
mtmd: minor fix
d0c08e36
minor formatting and style
f5bd310a
Merge pull request #9 from sfallah/sf/deepseek-ocr-attn
6687b4e7
CISC
commented on 2025-12-05
Merge branch 'ggml-org:master' into sf/deepseek-ocr
5f2ee1ae
fixed flake8 lint issues
1c88647e
minor editorconfig-check fixes
d981f19e
minor editorconfig-check fixes
705394c2
bluebread
commented on 2025-12-05
bluebread
commented on 2025-12-05
bluebread
commented on 2025-12-05
mtmd: simplify get_rel_pos
15f2ada0
mtmd: make sam hparams configurable
2d918b3e
mtmd: add detailed comments for resize_bicubic_pillow
5dfcc5ab
mtmd: fixed wrong input setting
53273f83
mtmd: convert model in FP16
48c6cf21
rhjdvsgsgks
commented on 2025-12-08
mtmd: minor fix
5174a1e6
ngxson
commented on 2025-12-09
mtmd: remove tweak to llama-mtmd-cli & deepseek-ocr template
01614069
fix: test-1.jpg ORC issue with small (640) resolution
ed944cd2
minor: editconfig-check fix
aaf2fd17
Merge branch 'master' into sf/deepseek-ocr-merge-test
33fabf0b
merge with changes from https://github.com/ggml-org/llama.cpp/pull/17909
d70f171f
minor: editconfig-check fix
4cbbe8ab
testing deepseek-ocr
47f0fee6
Merge remote-tracking branch 'sfallah/master' into sf/deepseek-ocr-me…
e0e69fd3
quick and (potential) dirty merge with https://github.com/ggml-org/ll…
f95a6fe9
refactoring, one single builder function and static helpers
f7736f23
added deepseek-ocr test to tests.sh
fb3bb6aa
Merge pull request #11 from sfallah/sf/deepseek-ocr-merge_#17965
1b38ccf6
minor formatting fixes
6c36c038
check with fixed expected resutls
dc2066e5
Merge pull request #10 from sfallah/sf/deepseek-ocr-test-script
3fc61d48
minor formatting
7f8621c5
Merge remote-tracking branch 'sfallah/master' into sf/deepseek-ocr
b3bf8cba
editorconfig-check fix
8ad98ee6
sfallah
requested a review
from
ngxson
322 days ago
sfallah
requested a review
from
CISC
322 days ago
Merge branch 'ggml-org:master' into sf/deepseek-ocr
4a4f8296
Merge remote-tracking branch 'sfallah/master' into sf/deepseek-ocr
51c3de68
merge with changes from https://github.com/ggml-org/llama.cpp/pull/18042
512b2c8f
Merge remote-tracking branch 'sfallah/master' into sf/deepseek-ocr
00d23570
minor
87e4a00c
CISC
commented on 2025-12-16
convert: minor fix
f629d02e
CISC
commented on 2025-12-17
mtmd: format code
5a741fda
convert: quick fix
616f009e
convert: quick fix
e5d426be
minor python formatting
c739cf20
Merge branch 'master' into sf/deepseek-ocr
9a05e1d1
fixed merge build issue
4d91711e
ngxson
assigned
ngxson
316 days ago
Merge remote-tracking branch 'sfallah/master' into sf/deepseek-ocr
ded92076
merge resolved
a94c2417
Merge remote-tracking branch 'sfallah/master' into sf/deepseek-ocr
6978c37f
minor fix
05789f56
Merge branch 'ggml-org:master' into sf/deepseek-ocr
7e47aa88
Merge branch 'ggml-org:master' into sf/deepseek-ocr
7ffa23c2
Merge remote-tracking branch 'sfallah/master' into sf/deepseek-ocr
f41d3239
minor
9b1a1b91
Merge remote-tracking branch 'sfallah/master' into sf/deepseek-ocr
52fcb139
Merge remote-tracking branch 'sfallah/master' into sf/deepseek-ocr
0031b41e
sfallah
requested a review
from
CISC
258 days ago
CISC
commented on 2026-02-18
Update convert_hf_to_gguf.py
5f2283bb
CISC
approved these changes on 2026-02-18
ngxson
commented on 2026-02-18
ngxson
commented on 2026-02-18
Merge remote-tracking branch 'sfallah/master' into sf/deepseek-ocr
7856e24c
- removed clip_is_deepseekocr
50c1e15a
Merge remote-tracking branch 'sfallah/master' into sf/deepseek-ocr
3e221cf7
- cleaning commented out code
e037b956
sfallah
requested a review
from
ngxson
256 days ago
Merge remote-tracking branch 'sfallah/master' into sf/deepseek-ocr
0b61c6ae
fixing instabilities issues reintroducing resize_bicubic_pillow
7a53e7e9
Merge remote-tracking branch 'sfallah/master' into sf/deepseek-ocr
c2e6701e
- use f16 model for deepseek-ocr test
49f3ca55
github-actions
added
testing
Merge remote-tracking branch 'sfallah/master' into sf/deepseek-ocr
21243f3d
Merge remote-tracking branch 'sfallah/master' into sf/deepseek-ocr
a493dc15
Merge remote-tracking branch 'sfallah/master' into sf/deepseek-ocr
754061e4
sfallah
requested a review
231 days ago
sfallah
requested a review
from
JohannesGaessler
231 days ago
sfallah
requested a review
231 days ago
Merge remote-tracking branch 'sfallah/master' into sf/deepseek-ocr
77253998
ngxson
commented on 2026-03-25
rename fc_w --> mm_fc_w
3754c324
Merge branch 'master' into sf/deepseek-ocr
d88b88e4
add links to OCR discussion
0ea5fa45
cleaner loading code
edf020df
add missing .weight to some tensors
80998695
add default jinja template (to be used by server)
1d900949
move test model to ggml-org
6faf264d
rolling back upscale change
8dabfe3a
ggerganov
approved these changes on 2026-03-25
github-actions
added
documentation
ngxson
approved these changes on 2026-03-25
CISC
commented on 2026-03-25
Update convert_hf_to_gguf.py
95cc5665
ngxson
requested a review
from
CISC
223 days ago
ngxson
requested a review
from
ngxson
223 days ago
ngxson
requested a review
from
ggerganov
223 days ago
CISC
approved these changes on 2026-03-25
ngxson
removed review request
223 days ago
ngxson
removed review request
from
JohannesGaessler
223 days ago
ggerganov
approved these changes on 2026-03-25
ngxson
merged
a970515b
into master
223 days ago
Login to write a write a comment.
Login via GitHub
Reviewers
ggerganov
CISC
ngxson
bluebread
rhjdvsgsgks
Assignees
ngxson
Labels
documentation
model
testing
Nvidia GPU
examples
python
ggml
Milestone
No milestone
Login to write a write a comment.
Login via GitHub