nikola/llama_cpp
1
0
Fork 0
Code Issues Pull Requests Actions 20 Packages Projects Releases Wiki Activity

model : add HunyuanOCR support (#21395)

Browse Source 
* HunyuanOCR: add support for text and vision models

- Add HunyuanOCR vision projector (perceiver-based) with Conv2d merge
- Add separate HUNYUAN_OCR chat template (content-before-role format)
- Handle HunyuanOCR's invalid pad_token_id=-1 in converter
- Fix EOS/EOT token IDs from generation_config.json
- Support xdrope RoPE scaling type
- Add tensor mappings for perceiver projector (mm.before_rms, mm.after_rms, etc.)
- Register HunYuanVLForConditionalGeneration for both text and mmproj conversion

* fix proper mapping

* Update gguf-py/gguf/tensor_mapping.py

Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com>

* Update tools/mtmd/clip.cpp

Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com>

* address comments

* update

* Fix typecheck

* Update convert_hf_to_gguf.py

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

* Update convert_hf_to_gguf.py

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

* Update convert_hf_to_gguf.py

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

* Update convert_hf_to_gguf.py

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

---------

Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com>
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
This commit is contained in:
Richard Davison
2026-04-05 23:32:14 +02:00
committed by GitHub
co-authored by Xuan-Son Nguyen Sigbjørn Skjæret
parent 761797ffdf
commit af76639f72
12 changed files with 273 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files
gguf-py/gguf/constants.py
+10
View File
@@ -734,6 +734,7 @@ class MODEL_TENSOR(IntEnum):
V_LAYER_OUT_SCALE = auto()
V_PRE_NORM = auto()
V_POST_NORM = auto()
V_MM_PRE_NORM = auto() # hunyuanocr
V_MM_POST_NORM = auto()
V_MM_INP_NORM = auto()
V_MM_INP_PROJ = auto() # gemma3
@@ -769,6 +770,8 @@ class MODEL_TENSOR(IntEnum):
V_MM_GATE = auto() # cogvlm
V_TOK_BOI = auto() # cogvlm
V_TOK_EOI = auto() # cogvlm
V_TOK_IMG_BEGIN = auto() # hunyuanocr
V_TOK_IMG_END = auto() # hunyuanocr
V_STD_BIAS = auto() # gemma4
V_STD_SCALE = auto() # gemma4
V_SAM_POS_EMBD = auto() # Deepseek-OCR
@@ -1246,6 +1249,9 @@ TENSOR_NAMES: dict[MODEL_TENSOR, str] = {
MODEL_TENSOR.V_MM_GATE: "mm.gate",
MODEL_TENSOR.V_TOK_BOI: "v.boi",
MODEL_TENSOR.V_TOK_EOI: "v.eoi",
MODEL_TENSOR.V_MM_PRE_NORM: "mm.pre_norm",
MODEL_TENSOR.V_TOK_IMG_BEGIN: "mm.image_begin",
MODEL_TENSOR.V_TOK_IMG_END: "mm.image_end",
MODEL_TENSOR.V_STD_BIAS: "v.std_bias", # gemma4
MODEL_TENSOR.V_STD_SCALE: "v.std_scale", # gemma4
# DeepSeek-OCR SAM
@@ -1393,6 +1399,9 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.V_MM_GATE,
MODEL_TENSOR.V_TOK_BOI,
MODEL_TENSOR.V_TOK_EOI,
MODEL_TENSOR.V_MM_PRE_NORM,
MODEL_TENSOR.V_TOK_IMG_BEGIN,
MODEL_TENSOR.V_TOK_IMG_END,
MODEL_TENSOR.V_STD_BIAS,
MODEL_TENSOR.V_STD_SCALE,
MODEL_TENSOR.V_SAM_POS_EMBD,
@@ -4113,6 +4122,7 @@ class VisionProjectorType:
GLM4V = "glm4v"
YOUTUVL = "youtuvl"
NEMOTRON_V2_VL = "nemotron_v2_vl"
HUNYUANOCR = "hunyuanocr"
# Items here are (block size, type size)