nikola/llama_cpp
1
0
Fork 0
Code Issues Pull Requests Actions 19 Packages Projects Releases Wiki Activity

mtmd: Add DeepSeekOCR Support (#17400)

Browse Source 
* mtmd: llama.cpp DeepSeekOCR support
init commit

* loading sam tensors

* mtmd: fix vision model processing

* deepseek-ocr clip-vit model impl

* mtmd: add DeepSeek-OCR LM support with standard attention

* mtmd: successfully runs DeepSeek-OCR LM in llama-cli

* mtmd: Fix RoPE type for DeepSeek-OCR LM.

* loading LM
testing Vision model loading

* sam warmup working

* sam erroneous return corrected

* clip-vit:  corrected cls_embd concat

* clip-vit: model convert  qkv_proj split

* corrected combining of image encoders' results

* fix: update callback for ffn_moe_weighted and add callback for attn_out in deepseek2 model

* concat image_newline and image_seperator tokens

* visual_model warmup (technically) works

* window partitioning using standard ggml ops

* sam implementation without using CPU only ops

* clip: fixed warnings

* Merge branch 'sf/deepseek-ocr' of github.com:sfallah/llama.cpp into sf/deepseek-ocr

* mtmd: fix get_rel_pos

* mtmd: fixed the wrong scaler for get_rel_pos

* image encoding technically works but the output can't be checked singe image decoding fails

* mtmd: minor changed

* mtmd: add native resolution support

* - image encoding debugged
- issues fixed mainly related wrong config like n_patches etc.
- configs need to be corrected in the converter

* mtmd: correct token order

* - dynamic resizing
- changes are concerning PR https://github.com/sfallah/llama.cpp/pull/4

* mtmd: quick fix token order

* mtmd: fix danling pointer

* mtmd: SAM numerically works

* mtmd: debug CLIP-L (vit_pre_ln)

* mtmd: debug CLIP-L & first working DeepSeek-OCR model

* mtmd : add --dsocr-mode CLI argument for DeepSeek-OCR resolution control & all native resolution modes work

* mtmd: simplify SAM patch embedding

* mtmd: adapt Pillow image resizing function

* mtmd:  simplify DeepSeek-OCR dynamic resolution preprocessing

* mtmd: remove --dsocr-mode argument

* mtmd: refactor code & remove unused helper functions

* mtmd: fix tensor names for image newlines and view separator

* clean up

* reverting automatically removed spaces

* reverting automatically removed spaces

* mtmd: fixed bad ocr check in Deepseek2 (LM)

* mtmd: support combined QKV projection in buid_vit

* using common build_attn in sam

* corrected code-branch when flash-attn disabled
enabling usage of --flash-attn option

* mtmd: minor fix

* minor formatting and style

* fixed flake8 lint issues

* minor editorconfig-check fixes

* minor editorconfig-check fixes

* mtmd: simplify get_rel_pos

* mtmd: make sam hparams configurable

* mtmd: add detailed comments for resize_bicubic_pillow

* mtmd: fixed wrong input setting

* mtmd: convert model in FP16

* mtmd: minor fix

* mtmd: remove tweak to llama-mtmd-cli & deepseek-ocr template

* fix: test-1.jpg ORC issue with small (640) resolution
setting min-resolution base (1024) max large (1280) for dynamic-resolution

* minor: editconfig-check fix

* merge with changes from https://github.com/ggml-org/llama.cpp/pull/17909
added new opt to tests.sh to disable flash-attn

* minor: editconfig-check fix

* testing deepseek-ocr
quick and dirty test script comparing results of Qwen2.5-VL vs DeepSeek-OCR

* quick and (potential) dirty merge with https://github.com/ggml-org/llama.cpp/pull/17909

* refactoring, one single builder function and static helpers

* added deepseek-ocr test to tests.sh

* minor formatting fixes

* check with fixed expected resutls

* minor formatting

* editorconfig-check fix

* merge with changes from https://github.com/ggml-org/llama.cpp/pull/18042

* minor
- added GLM-4.6V to big tests
- added missing deps for python test

* convert: minor fix

* mtmd: format code

* convert: quick fix

* convert: quick fix

* minor python formatting

* fixed merge build issue

* merge resolved
- fixed issues in convert
- tested several deepseek models

* minor fix

* minor

* Update convert_hf_to_gguf.py

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>

* - removed clip_is_deepseekocr
- removed redundant RESIZE_ALGO_BICUBIC_PILLOW resize-algo
- simplified image-preprocessing
- removed/simplified debug functions

* - cleaning commented out code

* fixing instabilities issues reintroducing resize_bicubic_pillow

* - use f16 model for deepseek-ocr test
- ignore llama-arch test for deepseek-ocr

* rename fc_w --> mm_fc_w

* add links to OCR discussion

* cleaner loading code

* add missing .weight to some tensors

* add default jinja template (to be used by server)

* move test model to ggml-org

* rolling back upscale change

* Update convert_hf_to_gguf.py

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>

---------

Co-authored-by: bluebread <hotbread70127@gmail.com>
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
Co-authored-by: Xuan Son Nguyen <son@huggingface.co>
Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com>
This commit is contained in:
Saba Fallah
2026-03-25 19:57:40 +01:00
committed by GitHub
co-authored by Sigbjørn Skjæret bluebread Xuan Son Nguyen Xuan-Son Nguyen
parent 056b50c319
commit a970515bdb
30 changed files with 1569 additions and 27 deletions
Download Patch File Download Diff File Expand all files Collapse all files
gguf-py/gguf/tensor_mapping.py
+74 -1
View File
@@ -1344,6 +1344,7 @@ class TensorNameMap:
MODEL_TENSOR.V_MMPROJ_FC: (
"model.connector.modality_projection.proj", # SmolVLM
"model.vision.linear_proj.linear_proj", # cogvlm
"model.projector.layers", # Deepseek-OCR
"visual.merger.proj", # glm4v
),
@@ -1364,6 +1365,7 @@ class TensorNameMap:
"vision_model.class_embedding", # llama 4
"model.vision.patch_embedding.cls_embedding", # cogvlm
"vision_model.radio_model.model.patch_generator.cls_token.token", # Nemotron Nano v2 VL
"model.vision_model.embeddings.class_embedding", # Deepseek-OCR
),
MODEL_TENSOR.V_ENC_EMBD_PATCH: (
@@ -1377,6 +1379,7 @@ class TensorNameMap:
"visual.patch_embed.proj", # qwen2vl
"vision_tower.patch_embed.proj", # kimi-vl
"model.vision.patch_embedding.proj", # cogvlm
"model.vision_model.embeddings.patch_embedding", # Deepseek-OCR CLIP
"siglip2.vision_model.embeddings.patch_embedding",
"vision_model.radio_model.model.patch_generator.embedder", # Nemotron Nano v2 VL
),
@@ -1398,10 +1401,19 @@ class TensorNameMap:
"vision_model.radio_model.model.patch_generator.pos_embed", # Nemotron Nano v2 VL
),
MODEL_TENSOR.V_ENC_EMBD_IMGNL: (
"model.image_newline", # Deepseek-OCR
),
MODEL_TENSOR.V_ENC_EMBD_VSEP: (
"model.view_seperator", # Deepseek-OCR
),
MODEL_TENSOR.V_ENC_ATTN_QKV: (
"visual.blocks.{bid}.attn.qkv", # qwen3vl
"model.vision.transformer.layers.{bid}.attention.query_key_value", # cogvlm
"vision_tower.encoder.blocks.{bid}.wqkv", # Kimi-K2.5
"model.vision_model.transformer.layers.{bid}.self_attn.qkv_proj", # Deepseek-OCR CLIP
"vision_tower.encoder.blocks.{bid}.wqkv" # Kimi-K2.5
"vision_model.radio_model.model.blocks.{bid}.attn.qkv", # Nemotron Nano v2 VL
),
@@ -1416,6 +1428,7 @@ class TensorNameMap:
"visual.blocks.{bid}.attn.q", # qwen2vl, generated
"vision_tower.encoder.blocks.{bid}.wq", # kimi-vl, generated
"siglip2.vision_model.encoder.layers.{bid}.self_attn.q_proj", # youtuvl
"model.vision_model.transformer.layers.{bid}.self_attn.q_proj", # Deepseek-OCR CLIP, generated
),
MODEL_TENSOR.V_ENC_ATTN_Q_NORM: (
@@ -1434,6 +1447,7 @@ class TensorNameMap:
"vision_encoder.transformer.layers.{bid}.attention.wk", # pixtral
"visual.blocks.{bid}.attn.k", # qwen2vl, generated
"vision_tower.encoder.blocks.{bid}.wk", # kimi-vl, generated
"model.vision_model.transformer.layers.{bid}.self_attn.k_proj", # Deepseek-OCR CLIP, generated
"siglip2.vision_model.encoder.layers.{bid}.self_attn.k_proj",
),
@@ -1454,6 +1468,7 @@ class TensorNameMap:
"visual.blocks.{bid}.attn.v", # qwen2vl, generated
"vision_tower.encoder.blocks.{bid}.wv", # kimi-vl, generated
"siglip2.vision_model.encoder.layers.{bid}.self_attn.v_proj",
"model.vision_model.transformer.layers.{bid}.self_attn.v_proj", # Deepseek-OCR CLIP, generated
),
MODEL_TENSOR.V_ENC_INPUT_NORM: (
@@ -1468,6 +1483,7 @@ class TensorNameMap:
"visual.blocks.{bid}.norm1", # qwen2vl
"vision_tower.encoder.blocks.{bid}.norm0", # kimi-vl (norm0/norm1)
"model.vision.transformer.layers.{bid}.input_layernorm", # cogvlm
"model.vision_model.transformer.layers.{bid}.layer_norm1", # Deepseek-OCR CLIP
"siglip2.vision_model.encoder.layers.{bid}.layer_norm1",
"vision_model.radio_model.model.blocks.{bid}.norm1", # Nemotron Nano v2 VL
),
@@ -1485,6 +1501,7 @@ class TensorNameMap:
"visual.blocks.{bid}.attn.proj", # qwen2vl
"vision_tower.encoder.blocks.{bid}.wo", # kimi-vl
"model.vision.transformer.layers.{bid}.attention.dense", # cogvlm
"model.vision_model.transformer.layers.{bid}.self_attn.out_proj", # Deepseek-OCR CLIP
"siglip2.vision_model.encoder.layers.{bid}.self_attn.out_proj", # youtuvl
"vision_model.radio_model.model.blocks.{bid}.attn.proj", # Nemotron Nano v2 VL
),
@@ -1501,6 +1518,7 @@ class TensorNameMap:
"visual.blocks.{bid}.norm2", # qwen2vl
"vision_tower.encoder.blocks.{bid}.norm1", # kimi-vl (norm0/norm1)
"model.vision.transformer.layers.{bid}.post_attention_layernorm", # cogvlm
"model.vision_model.transformer.layers.{bid}.layer_norm2", # Deepseek-OCR CLIP
"siglip2.vision_model.encoder.layers.{bid}.layer_norm2",
"vision_model.radio_model.model.blocks.{bid}.norm2", # Nemotron Nano v2 VL
),
@@ -1517,6 +1535,7 @@ class TensorNameMap:
"visual.blocks.{bid}.mlp.up_proj", # qwen2.5vl
"visual.blocks.{bid}.mlp.linear_fc1", # qwen3vl
"vision_tower.encoder.blocks.{bid}.mlp.fc0", # kimi-vl (fc0/fc1)
"model.vision_model.transformer.layers.{bid}.mlp.fc1", # Deepseek-OCR CLIP
"model.vision.transformer.layers.{bid}.mlp.fc1", # cogvlm
"siglip2.vision_model.encoder.layers.{bid}.mlp.fc1",
"vision_model.radio_model.model.blocks.{bid}.mlp.fc1", # Nemotron Nano v2 VL
@@ -1541,6 +1560,7 @@ class TensorNameMap:
"visual.blocks.{bid}.mlp.linear_fc2", # qwen3vl
"vision_tower.encoder.blocks.{bid}.mlp.fc1", # kimi-vl (fc0/fc1)
"model.vision.transformer.layers.{bid}.mlp.fc2", # cogvlm
"model.vision_model.transformer.layers.{bid}.mlp.fc2", # Deepseek-OCR CLIP
"siglip2.vision_model.encoder.layers.{bid}.mlp.fc2",
"vision_model.radio_model.model.blocks.{bid}.mlp.fc2", # Nemotron Nano v2 VL
),
@@ -1560,6 +1580,7 @@ class TensorNameMap:
"vision_tower.ln_pre", # pixtral-hf
"vision_encoder.ln_pre", # pixtral
"vision_model.layernorm_pre", # llama4
"model.vision_model.pre_layrnorm", # Deepseek-OCR CLIP
),
MODEL_TENSOR.V_POST_NORM: (
@@ -1662,6 +1683,58 @@ class TensorNameMap:
"model.visual.deepstack_merger_list.{bid}.linear_fc2", # deepstack in qwen3vl
),
MODEL_TENSOR.V_SAM_POS_EMBD: (
"model.sam_model.pos_embed",
),
MODEL_TENSOR.V_SAM_PATCH_EMBD: (
"model.sam_model.patch_embed.proj",
),
MODEL_TENSOR.V_SAM_PRE_NORM: (
"model.sam_model.blocks.{bid}.norm1", # deepstack in qwen3vl
),
MODEL_TENSOR.V_SAM_POST_NORM: (
"model.sam_model.blocks.{bid}.norm2", # deepstack in qwen3vl
),
MODEL_TENSOR.V_SAM_ATTN_POS_H: (
"model.sam_model.blocks.{bid}.attn.rel_pos_h",
),
MODEL_TENSOR.V_SAM_ATTN_POS_W: (
"model.sam_model.blocks.{bid}.attn.rel_pos_w",
),
MODEL_TENSOR.V_SAM_ATTN_QKV: (
"model.sam_model.blocks.{bid}.attn.qkv",
),
MODEL_TENSOR.V_SAM_ATTN_OUT: (
"model.sam_model.blocks.{bid}.attn.proj",
),
MODEL_TENSOR.V_SAM_MLP_LIN_1: (
"model.sam_model.blocks.{bid}.mlp.lin1",
),
MODEL_TENSOR.V_SAM_MLP_LIN_2: (
"model.sam_model.blocks.{bid}.mlp.lin2",
),
MODEL_TENSOR.V_SAM_NECK: (
"model.sam_model.neck.{bid}",
),
MODEL_TENSOR.V_SAM_NET_2: (
"model.sam_model.net_2",
),
MODEL_TENSOR.V_SAM_NET_3: (
"model.sam_model.net_3",
),
MODEL_TENSOR.V_MM_POST_FC_NORM: (
"model.vision.linear_proj.norm1", # cogvlm
),