Skip to content

Add Qwen2.5VL support #12402

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 20 commits into from
Apr 27, 2025
Merged

Add Qwen2.5VL support #12402

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
20 commits
Select commit Hold shift + click to select a range
d9f2d71
implment vision model architecture, gguf convertor
HimariO Feb 2, 2025
3d5198e
handle window attention inputs
HimariO Feb 2, 2025
69b39ad
add debug utils
HimariO Mar 15, 2025
444e47c
fix few incorrect tensor memory layout
HimariO Mar 15, 2025
2eb3293
move position id remap out of ggml to avoid int32 cuda operations
HimariO Mar 15, 2025
d1af459
cleaning up
HimariO Apr 1, 2025
91fbdd7
ignore transformers Qwen2_5_xxx type check
HimariO Apr 3, 2025
2de5dc3
remove not so often use `qwen2vl-cli` debug functions
HimariO Apr 4, 2025
7f530ac
remove commented-out code blocks
HimariO Apr 4, 2025
a3cd0e5
fix attn weight scaling after rebase
HimariO Apr 7, 2025
caa7e57
add `PROJECTOR_TYPE_QWEN2_5_VL`
HimariO Apr 25, 2025
f69e9fa
remove `KEY_USE_GLU_MLP`, `KEY_USE_RMS_NORM`
HimariO Apr 25, 2025
77b144a
replace `KEY_FULLATTN_BLK_IDX` with `KEY_WIN_ATTN_PATTERN`
HimariO Apr 25, 2025
7e1bb04
remove `attn_window_size` from gguf
HimariO Apr 26, 2025
516735a
fix model conversion
ngxson Apr 26, 2025
5085dbb
Merge branch 'master' into qwen25-vl
ngxson Apr 26, 2025
0c74ea5
clean up
ngxson Apr 26, 2025
82f8e72
Merge branch 'master' into qwen25-vl
ngxson Apr 26, 2025
89be919
fix merging problem
ngxson Apr 26, 2025
53a15d0
add test
ngxson Apr 26, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
11 changes: 6 additions & 5 deletions convert_hf_to_gguf.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2554,11 +2554,12 @@ def set_vocab(self):
except FileNotFoundError:
self._set_vocab_gpt2()

def get_tensors(self) -> Iterator[tuple[str, Tensor]]:
for name, data in super().get_tensors():
if name.startswith("visual."):
continue
yield name, data
def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
del bid # unused
if name.startswith("visual."):
# skip visual tensors
return []
return [(self.map_tensor_name(name), data_torch)]


@ModelBase.register("WavTokenizerDec")
Expand Down
8 changes: 8 additions & 0 deletions examples/llava/clip-impl.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -34,9 +34,14 @@
#define KEY_PROJ_SCALE_FACTOR "clip.vision.projector.scale_factor"
#define KEY_PROJ_TYPE "clip.projector_type"

#define KEY_USE_GLU_MLP "clip.use_glu_mlp" // for qwen2.5vl
#define KEY_USE_RMS_NORM "clip.use_rms_norm" // for qwen2.5vl

#define KEY_MM_PATCH_MERGE_TYPE "clip.vision.mm_patch_merge_type"
#define KEY_IMAGE_GRID_PINPOINTS "clip.vision.image_grid_pinpoints"
#define KEY_IMAGE_CROP_RESOLUTION "clip.vision.image_crop_resolution"
#define KEY_WIN_ATTN_PATTERN "clip.vision.n_wa_pattern"
#define KEY_ATTN_WINDOW_SIZE "clip.vision.window_size"


//
Expand All @@ -55,6 +60,7 @@
#define TN_FFN_DOWN "%s.blk.%d.ffn_down.%s"
#define TN_FFN_GATE "%s.blk.%d.ffn_gate.%s"
#define TN_FFN_UP "%s.blk.%d.ffn_up.%s"
#define TN_FFN_GATE "%s.blk.%d.ffn_gate.%s"
#define TN_LN_1 "%s.blk.%d.ln1.%s"
#define TN_LN_2 "%s.blk.%d.ln2.%s"
#define TN_LN_PRE "%s.pre_ln.%s"
Expand Down Expand Up @@ -95,6 +101,7 @@ enum projector_type {
PROJECTOR_TYPE_GEMMA3,
PROJECTOR_TYPE_IDEFICS3,
PROJECTOR_TYPE_PIXTRAL,
PROJECTOR_TYPE_QWEN25VL,
PROJECTOR_TYPE_UNKNOWN,
};

Expand All @@ -105,6 +112,7 @@ static std::map<projector_type, std::string> PROJECTOR_TYPE_NAMES = {
{ PROJECTOR_TYPE_MINICPMV, "resampler"},
{ PROJECTOR_TYPE_GLM_EDGE, "adapter"},
{ PROJECTOR_TYPE_QWEN2VL, "qwen2vl_merger"},
{ PROJECTOR_TYPE_QWEN25VL, "qwen2.5vl_merger"},
{ PROJECTOR_TYPE_GEMMA3, "gemma3"},
{ PROJECTOR_TYPE_IDEFICS3, "idefics3"},
{ PROJECTOR_TYPE_PIXTRAL, "pixtral"},
Expand Down
Loading
Loading