ver4a/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

llava : Add Granite Vision Support (#11794)

Browse source 
* Add super wip scripts for multimodal granite gguf

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>

* Add example for converting mmgranite to gguf

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>

* remove hardcoded path

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>

* Add vision feature layer to gguf params

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>

* Clean up llava surgery and remove name substitution hacks

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>

* Add transformers llava next tensor name mapping

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>

* Make siglip / openclip mutuall exclusive

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>

* Fix projector linear substitution

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>

* Fix linear 2 substitution index

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>

* Increase max flattened gridpoints to 64

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>

* Fix hardcoded concat for multiple feature layers

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>

* Pull vision feature layers out of gguf keys

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>

* fix num gridpoints and use all layers

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>

* Avoid dropping last image encoder layer in llava models

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>

* Use 10 for max number of patches

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>

* Standardize vision feature layers

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>

* Cleanup logs

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>

* Update comment for vision feature layer init

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>

* Update notes for alternative to legacy llm conversion script

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>

* Fix notes rendering

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>

* Add v prefix to vision feature layer log

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>

* Use current defaults for feature layer

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>

* Use constant for max gridpoints / feat layers, style fixes

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>

* clarify non-negative feature layers

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>

* Remove CLIP_API from func signature

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>

* USE MAX_IMAGE_FEATURE_LAYERS const in layer calc

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>

* Clarify feature layers are non negative ints and not uint

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>

* Fix condition for reading feature layers

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>

* pop last llava layer when feature layers are unset

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>

* Fix unset vision layer 0

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>

* Update examples/llava/clip.cpp

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

* Reenable assertion for out of bounds get_rows

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>

* Use std vector for gridpoints and feature layers

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>

* Caculate max feature layer at load time

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>

* Include base patch for granite vision allocation

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>

* Fix trailing whitespace

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>

* Add max num patches = 10 back for minicpmv

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>

* Use unordered set to store feature layers

Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com>
Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>

* Use max feature layer for postnorm

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>

* Apply suggestions from code review

---------

Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>
Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com>
This commit is contained in:
Alex Brooks 2025-02-24 09:09:51 -07:00 • committed by GitHub
parent 08d5986290
commit 7a2c913e66
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
6 changed files with 235 additions and 42 deletions
Download patch file Download diff file Expand all files Collapse all files

41
examples/llava/llava_surgery_v2.py
View file

@ -33,6 +33,33 @@ def save_model(model, file_path, file_type):
else:
torch.save(model, file_path)
# Helpers to match weight names from specific components or
# determine if a saved shard contains that component
def is_vision_tower(weight_name):
return (
weight_name.startswith("model.vision_tower") or
weight_name.startswith("vit.") or
weight_name.startswith("vision_tower")
)
def is_newline(weight_name):
return (
weight_name.startswith("model.image_newline") or
weight_name.startswith("image_newline")
)
def is_mm_projector(weight_name):
return (
weight_name.startswith("model.mm_projector") or
weight_name.startswith("vision_proj.") or
weight_name.startswith("multi_modal_projector")
)
def newline_criteria(checkpoint):
return any(is_newline(k) for k in checkpoint.keys())
def proj_criteria(checkpoint):
return any(is_mm_projector(k) for k in checkpoint.keys())
# Adapted function to clean vision tower from checkpoint
def clean_vision_tower_from_checkpoint(checkpoint_path):
@ -40,7 +67,7 @@ def clean_vision_tower_from_checkpoint(checkpoint_path):
# file_type = 'pytorch'
model_path = os.path.dirname(checkpoint_path)
print(f"Searching for vision tower tensors in {checkpoint_path}")
clip_tensors = [k for k, v in checkpoint.items() if (k.startswith("model.vision_tower") or k.startswith("vit."))]
clip_tensors = [k for k, v in checkpoint.items() if is_vision_tower(k)]
if len(clip_tensors) > 0:
print(f"Found {len(clip_tensors)} tensors to extract from {checkpoint_path}")
@ -84,12 +111,6 @@ def find_relevant_checkpoints(checkpoint_paths, newline_criteria, projector):
return newline_checkpoint_path, projector_checkpoint_path
def newline_criteria(checkpoint):
return any(k.startswith("model.image_newline") for k in checkpoint.keys())
def proj_criteria(checkpoint):
return any(k.startswith("model.mm_projector") or k.startswith("vision_proj.") for k in checkpoint.keys())
# Command-line interface setup
ap = argparse.ArgumentParser()
@ -123,14 +144,14 @@ first_checkpoint = None
if newline_checkpoint_path is not None:
print(f"Taking newline from {newline_checkpoint_path}")
first_checkpoint, file_type = load_model(newline_checkpoint_path)
first_mm_tensors = [k for k, v in first_checkpoint.items() if k.startswith("model.image_newline")]
first_mm_tensors = [k for k, v in first_checkpoint.items() if is_newline(k)]
# Load the checkpoint
mm_tensors = []
last_checkpoint = None
if projector_checkpoint_path is not None:
last_checkpoint, file_type = load_model(projector_checkpoint_path)
mm_tensors = [k for k, v in last_checkpoint.items() if k.startswith("model.mm_projector") or k.startswith("vision_proj.")]
mm_tensors = [k for k, v in last_checkpoint.items() if is_mm_projector(k)]
if len(mm_tensors) == 0:
if last_checkpoint is not None:
@ -155,5 +176,5 @@ if len(projector) > 0:
save_model(projector, f"{args.model}/llava.projector", 'pytorch')
print("Done!")
print(f"Now you can convert {args.model} to a a regular LLaMA GGUF file.")
print(f"Now you can convert {args.model} to a regular LLaMA GGUF file.")
print(f"Also, use {args.model}/llava.projector to prepare a llava-encoder.gguf file.")