File size: 6,169 Bytes
a6447a4 |
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 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 |
# Copyright 2024 NVIDIA CORPORATION & AFFILIATES
#
# 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 os
import os.path as osp
from huggingface_hub import repo_exists, snapshot_download
from huggingface_hub.utils import HFValidationError, validate_repo_id
from transformers import AutoConfig, PretrainedConfig
def get_model_config(config):
default_keys = ["llm_cfg", "vision_tower_cfg", "mm_projector_cfg"]
if hasattr(config, "_name_or_path") and len(config._name_or_path) >= 2:
root_path = config._name_or_path
else:
root_path = config.resume_path
# download from huggingface
if root_path is not None and not osp.exists(root_path):
try:
valid_hf_repo = repo_exists(root_path)
except HFValidationError as e:
valid_hf_repo = False
if valid_hf_repo:
root_path = snapshot_download(root_path)
return_list = []
for key in default_keys:
cfg = getattr(config, key, None)
if isinstance(cfg, dict):
try:
return_list.append(os.path.join(root_path, key[:-4]))
except:
raise ValueError(f"Cannot find resume path in config for {key}!")
elif isinstance(cfg, PretrainedConfig):
return_list.append(os.path.join(root_path, key[:-4]))
elif isinstance(cfg, str):
return_list.append(cfg)
return return_list
def get_model_config_fp8(config):
default_keys = ["llm_cfg", "vision_tower_cfg", "mm_projector_cfg"]
if hasattr(config, "_name_or_path") and len(config._name_or_path) >= 2:
root_path = config._name_or_path
else:
root_path = config.resume_path
# download from huggingface
if root_path is not None and not osp.exists(root_path):
try:
valid_hf_repo = repo_exists(root_path)
except HFValidationError as e:
valid_hf_repo = False
if valid_hf_repo:
root_path = snapshot_download(root_path)
return_list = []
for key in default_keys:
cfg = getattr(config, key, None)
if isinstance(cfg, dict):
try:
return_list.append(os.path.join(root_path, key[:-4]))
except:
raise ValueError(f"Cannot find resume path in config for {key}!")
elif isinstance(cfg, PretrainedConfig):
return_list.append(os.path.join(root_path, key[:-4]))
elif isinstance(cfg, str):
return_list.append(cfg)
# fp8_llm
key = "fp8_llm_cfg"
directory_path = os.path.join(root_path, key[:-4])
assert os.path.isdir(directory_path) and os.listdir(
directory_path
), "You need to first convert the model weights to FP8 explicitly."
return_list.append(directory_path)
return return_list
def get_model_config_fp8(config):
default_keys = ["llm_cfg", "vision_tower_cfg", "mm_projector_cfg"]
if hasattr(config, "_name_or_path") and len(config._name_or_path) >= 2:
root_path = config._name_or_path
else:
root_path = config.resume_path
# download from huggingface
if root_path is not None and not osp.exists(root_path):
try:
valid_hf_repo = repo_exists(root_path)
except HFValidationError as e:
valid_hf_repo = False
if valid_hf_repo:
root_path = snapshot_download(root_path)
return_list = []
for key in default_keys:
cfg = getattr(config, key, None)
if isinstance(cfg, dict):
try:
return_list.append(os.path.join(root_path, key[:-4]))
except:
raise ValueError(f"Cannot find resume path in config for {key}!")
elif isinstance(cfg, PretrainedConfig):
return_list.append(os.path.join(root_path, key[:-4]))
elif isinstance(cfg, str):
return_list.append(cfg)
# fp8_llm
key = "fp8_llm_cfg"
directory_path = os.path.join(root_path, key[:-4])
assert os.path.isdir(directory_path) and os.listdir(
directory_path
), "You need to first convert the model weights to FP8 explicitly."
return_list.append(directory_path)
return return_list
def is_mm_model(model_path):
"""
Check if the model at the given path is a visual language model.
Args:
model_path (str): The path to the model.
Returns:
bool: True if the model is an MM model, False otherwise.
"""
config = AutoConfig.from_pretrained(model_path)
architectures = config.architectures
for architecture in architectures:
if "llava" in architecture.lower():
return True
return False
def auto_upgrade(config):
cfg = AutoConfig.from_pretrained(config)
if "llava" in config and "llava" not in cfg.model_type:
assert cfg.model_type == "llama"
print("You are using newer LLaVA code base, while the checkpoint of v0 is from older code base.")
print("You must upgrade the checkpoint to the new code base (this can be done automatically).")
confirm = input("Please confirm that you want to upgrade the checkpoint. [Y/N]")
if confirm.lower() in ["y", "yes"]:
print("Upgrading checkpoint...")
assert len(cfg.architectures) == 1
setattr(cfg.__class__, "model_type", "llava")
cfg.architectures[0] = "LlavaLlamaForCausalLM"
cfg.save_pretrained(config)
print("Checkpoint upgraded.")
else:
print("Checkpoint upgrade aborted.")
exit(1)
|