Spaces:
Sleeping
Sleeping
File size: 2,675 Bytes
6f0b9f8 | 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 | # Copyright 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
# SPDX-License-Identifier: Apache-2.0
# This file is modified from https://github.com/haotian-liu/LLaVA/
import torch # noqa
import os
from transformers import AutoConfig, PretrainedConfig, PreTrainedModel
from .siglip_encoder import SiglipVisionTower
from .context_provider import ContextProvider, ContextProviderConfig
def build_vision_tower(
model_name_or_path: str, config: PretrainedConfig
) -> PreTrainedModel:
## skip vision tower instantiation
if model_name_or_path is None:
return None
vision_tower_arch = None
if config.resume_path and "radio" not in model_name_or_path:
assert os.path.exists(
model_name_or_path
), f"Resume vision tower path {model_name_or_path} does not exist!"
vision_tower_cfg = AutoConfig.from_pretrained(model_name_or_path, trust_remote_code=True)
vision_tower_arch = vision_tower_cfg.architectures[0].lower()
vision_tower_name = (
vision_tower_arch if vision_tower_arch is not None else model_name_or_path
)
if "siglip" in vision_tower_name:
vision_tower = SiglipVisionTower(model_name_or_path, config)
else:
raise ValueError(f"Unknown vision tower: {model_name_or_path}")
config.mm_hidden_size = vision_tower.config.hidden_size
return vision_tower
def build_context_provider(
model_type_or_path: str, config: PretrainedConfig
) -> PreTrainedModel:
if model_type_or_path is None:
return None
## load from pretrained model
if config.resume_path:
assert os.path.exists(
model_type_or_path
), f"Resume context provider path {model_type_or_path} does not exist!"
return ContextProvider.from_pretrained(
model_type_or_path, config, torch_dtype=eval(config.model_dtype)
)
## build from scratch
else:
mm_projector_cfg = ContextProviderConfig(model_type_or_path)
mm_projector = ContextProvider(mm_projector_cfg, config).to(
eval(config.model_dtype)
)
return mm_projector
|