When I use ai model in situation cannot use jupyter notebook, I write this script for me, because I don't want to copy paste everytime and reload model to continue debug need many time.
pip install https://github.com/wayne931121/python_debug_repl/releases/download/0.0.2/cdebug-0.0.2.tar.gz
or
pip install cdebug
import cdebug
cdebug.main(globals())
PS C:\Users\原神\Desktop> python test.py
$cdebug20251007-080412: print(1)
1
$cdebug20251007-080419: !dir
磁碟區 C 中的磁碟是 Windows
磁碟區序號: B8C5-1580
C:\Users\原神\Desktop 的目錄
2025/10/07 上午 02:55 <DIR> .
2025/10/06 上午 08:39 <DIR> ..
2025/10/06 下午 10:35 11,407 b.jpg
2025/10/07 上午 07:51 370 cdebug.py
2025/10/07 上午 12:01 3,906 cogvideox_i2v_colab.py
2025/09/19 下午 05:27 2,261 Google Chrome.lnk
2025/09/07 下午 10:41 1,241 LINE.lnk
2025/09/22 上午 12:09 904 Lively Wallpaper.lnk
2025/05/22 下午 04:21 2,335 Microsoft Edge.lnk
2024/05/28 下午 06:57 1,077 OBS Studio (64bit).lnk
2025/05/22 下午 04:59 1,431 Roblox Player.lnk
2025/05/22 下午 04:56 1,259 Roblox Studio.lnk
2025/10/06 下午 10:34 <DIR> 神秘資料夾
2025/10/07 上午 12:23 13,833 系統管理員 命令提示字元 - conda activate Daii - python cdebug.py.txt
2025/10/07 上午 01:12 15,933 系統管理員 命令提示字元 - conda activate Daii - python cdebug.py1.txt
2025/10/07 上午 02:08 19,874 系統管理員 命令提示字元 - conda activate Daii - python cdebug.py111.txt
2025/10/07 上午 02:55 20,513 系統管理員 命令提示字元 - conda activate Daii - python cdebug.py3.txt
14 個檔案 96,344 位元組
3 個目錄 45,472,346,112 位元組可用
$cdebug20251007-080423: def a():\br print(2,"\n",end="\n")
$cdebug20251007-080455: a()
2
$cdebug20251007-080500: exit
PS C:\Users\原神\Desktop>
import os
import cdebug
######################################################################################################################
print("import torch")
os.environ["HF_HUB_ENABLE_HF_TRANSFER"] = "1"
import torch
from diffusers import AutoencoderKLCogVideoX, CogVideoXImageToVideoPipeline, CogVideoXTransformer3DModel
from diffusers.utils import export_to_video, load_image
from transformers import T5EncoderModel
"""#### Load models and create pipeline
Note: `bfloat16`, which is the recommended dtype for running "CogVideoX-5b-I2V" will cause OOM errors due to lack of efficient support on Turing GPUs.
Therefore, we must use `float16`, which might result in poorer generation quality. The recommended solution is to use Ampere or above GPUs, which also support efficient quantization kernels from [TorchAO](https://github.com/pytorch/ao) :(
"""
# model_id = "THUDM/CogVideoX-5b-I2V"
##!!!!!!!REMBER USE r"",NOT ""!!!!!!!!!!!!!!!
##WILL CAUSE BUG!!!!!!!!!!##########
path = r"D:\2\hub\models--THUDM--CogVideoX-5b-I2V\snapshots\a6f0f4858a8395e7429d82493864ce92bf73af11"
print("Load Model")
print("transformer")
transformer = CogVideoXTransformer3DModel.from_pretrained(path, subfolder="transformer", torch_dtype=torch.float16)
print("text_encoder")
text_encoder = T5EncoderModel.from_pretrained(path, subfolder="text_encoder", torch_dtype=torch.float16)
print("vae")
vae = AutoencoderKLCogVideoX.from_pretrained(path, subfolder="vae", torch_dtype=torch.float16)
print("START THE SHELL... Begin INTO DEBUG MODE....")
######################################################################################################################
cdebug.main(globals())
import os, time
#######
The code you need to run first
#######
while 1:
try:
cd = input("\033[1;32m$cdebug" + time.strftime("%Y%m%d-%H%M%S", time.localtime()) + ": \033[00m" )
if cd=="exit":
break
cd = cd.replace("\\br","\n")
if cd and cd[0:1]=="!":
os.system(cd[1:])
else:
exec(cd, globals())
except Exception as e:
print(e)
import os, time
######################################################################################################################
print("import torch")
os.environ["HF_HUB_ENABLE_HF_TRANSFER"] = "1"
import torch
from diffusers import AutoencoderKLCogVideoX, CogVideoXImageToVideoPipeline, CogVideoXTransformer3DModel
from diffusers.utils import export_to_video, load_image
from transformers import T5EncoderModel
"""#### Load models and create pipeline
Note: `bfloat16`, which is the recommended dtype for running "CogVideoX-5b-I2V" will cause OOM errors due to lack of efficient support on Turing GPUs.
Therefore, we must use `float16`, which might result in poorer generation quality. The recommended solution is to use Ampere or above GPUs, which also support efficient quantization kernels from [TorchAO](https://github.com/pytorch/ao) :(
"""
# model_id = "THUDM/CogVideoX-5b-I2V"
##!!!!!!!REMBER USE r"",NOT ""!!!!!!!!!!!!!!!
##WILL CAUSE BUG!!!!!!!!!!##########
path = r"D:\2\hub\models--THUDM--CogVideoX-5b-I2V\snapshots\a6f0f4858a8395e7429d82493864ce92bf73af11"
print("Load Model")
print("transformer")
transformer = CogVideoXTransformer3DModel.from_pretrained(path, subfolder="transformer", torch_dtype=torch.float16)
print("text_encoder")
text_encoder = T5EncoderModel.from_pretrained(path, subfolder="text_encoder", torch_dtype=torch.float16)
print("vae")
vae = AutoencoderKLCogVideoX.from_pretrained(path, subfolder="vae", torch_dtype=torch.float16)
print("START THE SHELL... Begin INTO DEBUG MODE....")
######################################################################################################################
while 1:
try:
cd = input("\033[1;32m$cdebug" + time.strftime("%Y%m%d-%H%M%S", time.localtime()) + ": \033[00m" )
if cd=="exit":
break
cd = cd.replace("\\br","\n")
if len(cd) and cd[0:1]=="!":
os.system(cd[1:])
else:
exec(cd, globals())
except Exception as e:
print(e)