File size: 4,076 Bytes
dd15a4b |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 |
import os
from .clip_encoder import CLIPVisionTower
from .imagebind import ImageBindWrapper
from .open_clip_encoder import OpenCLIPVisionTower
from .siglip_encoder import SigLipVisionTower
from .clip_encoder import CLIPVisionTower, CLIPVisionTowerS2
from .eva_clip.eva_clip_encoder import EvaClipVisionTower
from .dev_eva_clip.eva_vit import EvaViTWrapper
from blip3o.model.nextdit_crossattn import NextDiTCrossAttnConfig, NextDiTCrossAttn
from blip3o.model.sana_crossattn import SanaCrossAttnConfig, SanaCrossAttn
from diffusers.models import AutoencoderKL
from diffusers.schedulers import FlowMatchEulerDiscreteScheduler, DPMSolverMultistepScheduler
from diffusers import SanaTransformer2DModel
import torch
def build_vision_tower(vision_tower_cfg, **kwargs):
vision_tower = getattr(vision_tower_cfg, 'mm_vision_tower', getattr(vision_tower_cfg, 'vision_tower', None))
is_absolute_path_exists = os.path.exists(vision_tower)
use_s2 = getattr(vision_tower_cfg, 's2', False)
if "siglip" in vision_tower:
return SigLipVisionTower(vision_tower, vision_tower_cfg=vision_tower_cfg, **kwargs)
if "eva" in vision_tower:
return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
if is_absolute_path_exists or vision_tower.startswith("openai") or vision_tower.startswith("laion") or "ShareGPT4V" in vision_tower:
if use_s2:
return CLIPVisionTowerS2(vision_tower, args=vision_tower_cfg, **kwargs)
else:
return CLIPVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
raise ValueError(f'Unknown vision tower: {vision_tower}')
def build_gen_vision_tower(vision_tower_cfg, **kwargs):
vision_tower = getattr(vision_tower_cfg, 'gen_vision_tower')
is_absolute_path_exists = os.path.exists(vision_tower)
use_s2 = getattr(vision_tower_cfg, 's2', False)
if "siglip" in vision_tower:
return SigLipVisionTower(vision_tower, vision_tower_cfg=vision_tower_cfg, **kwargs)
if "eva" in vision_tower:
return EvaClipVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
if is_absolute_path_exists or vision_tower.startswith("openai") or vision_tower.startswith("laion") or "ShareGPT4V" in vision_tower:
if use_s2:
return CLIPVisionTowerS2(vision_tower, args=vision_tower_cfg, **kwargs)
else:
return CLIPVisionTower(vision_tower, args=vision_tower_cfg, **kwargs)
raise ValueError(f'Unknown vision tower: {vision_tower}')
def build_dit(vision_tower_cfg, **kwargs):
vision_tower_cfg.hidden_size = 896
print("="*20, "vision tower config", vision_tower_cfg, "="*20)
# if not hasattr(vision_tower_cfg, "hidden_size"):
# if "3B" in vision_tower_cfg.model_name_or_path:
# vision_tower_cfg.hidden_size = 2048
# elif "7B" in vision_tower_cfg.model_name_or_path:
# vision_tower_cfg.hidden_size = 3584
# else:
# vision_tower_cfg.hidden_size = 3072
print("="*20, "Building SANA with hidden size", vision_tower_cfg.hidden_size, "="*20)
#dit = NextDiTCrossAttn(NextDiTCrossAttnConfig(latent_embedding_size=vision_tower_cfg.hidden_size))
#noise_scheduler = FlowMatchEulerDiscreteScheduler.from_pretrained("Alpha-VLLM/Lumina-Next-SFT-diffusers", subfolder="scheduler")
#dit = SanaTransformer2DModel.from_pretrained("Efficient-Large-Model/Sana_600M_512px_diffusers", subfolder="transformer", torch_dtype=torch.float16,)
# dit = SanaCrossAttn(SanaCrossAttnConfig()) #cross_attention_dim=vision_tower_cfg.hidden_size))
#dit = SanaTransformer2DModel.from_pretrained("Efficient-Large-Model/Sana_600M_512px_diffusers", subfolder="transformer")
dit = SanaTransformer2DModel.from_pretrained("Efficient-Large-Model/Sana_600M_1024px_diffusers",ignore_missing_keys=True, ignore_mismatched_sizes=True,device_map="cpu", subfolder="transformer")
noise_scheduler = DPMSolverMultistepScheduler.from_pretrained("Efficient-Large-Model/Sana_600M_1024px_diffusers",subfolder="scheduler")
return dit, noise_scheduler
|