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

llava : support Minicpm-omni (#11289)

Browse source 
* init

* add readme

* update readme

* no use make

* update readme

* update fix code

* fix editorconfig-checker

* no change convert py

* use clip_image_u8_free
This commit is contained in:
tc-mb 2025-01-22 15:35:48 +08:00 • committed by GitHub
parent 6171c9d258
commit 3e3357fd77
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
6 changed files with 100 additions and 15 deletions
Download patch file Download diff file Expand all files Collapse all files

15
examples/llava/minicpmv-convert-image-encoder-to-gguf.py
View file

@ -501,7 +501,7 @@ default_image_mean = [0.48145466, 0.4578275, 0.40821073]
default_image_std = [0.26862954, 0.26130258, 0.27577711]
ap.add_argument('--image-mean', type=float, nargs='+', help='Mean of the images for normalization (overrides processor) ', default=None)
ap.add_argument('--image-std', type=float, nargs='+', help='Standard deviation of the images for normalization (overrides processor)', default=None)
ap.add_argument('--minicpmv_version', type=int, help='minicpmv_version: MiniCPM-V-2 use 1; MiniCPM-V-2.5 use 2; MiniCPM-V-2.6 use 3', default=2)
ap.add_argument('--minicpmv_version', type=int, help='minicpmv_version: MiniCPM-V-2 use 1; MiniCPM-V-2.5 use 2; MiniCPM-V-2.6 use 3; MiniCPM-o-2.6 use 4', default=2)
# with proper
args = ap.parse_args()
@ -545,12 +545,19 @@ if args.use_f32:
minicpmv_version = args.minicpmv_version
emb_dim = 4096
block_count = 26
if minicpmv_version == 1:
emb_dim = 2304
block_count = 26
elif minicpmv_version == 2:
emb_dim = 4096
block_count = 27
elif minicpmv_version == 3:
emb_dim = 3584
block_count = 27
elif minicpmv_version == 4:
emb_dim = 3584
block_count = 27
default_vision_config = {
"hidden_size": 1152,
@ -567,6 +574,9 @@ model = Idefics2VisionTransformer(vision_config)
if minicpmv_version == 3:
vision_config = SiglipVisionConfig(**default_vision_config)
model = SiglipVisionTransformer(vision_config)
elif minicpmv_version == 4:
vision_config = SiglipVisionConfig(**default_vision_config)
model = SiglipVisionTransformer(vision_config)
processor = None
# if model.attn_pool is not None:
@ -587,7 +597,7 @@ elif args.minicpmv_projector is not None:
fname_middle = "mmproj-"
has_text_encoder = False
has_minicpmv_projector = True
minicpmv_version = 3
minicpmv_version = 4
elif args.vision_only:
fname_middle = "vision-"
has_text_encoder = False
@ -625,7 +635,6 @@ if has_vision_encoder:
fout.add_uint32("clip.vision.projection_dim", 0)
fout.add_uint32(add_key_str(KEY_ATTENTION_HEAD_COUNT, VISION), 16)
fout.add_float32(add_key_str(KEY_ATTENTION_LAYERNORM_EPS, VISION), 1e-6)
block_count = 26
fout.add_uint32(add_key_str(KEY_BLOCK_COUNT, VISION), block_count)
if processor is not None: