TIL on Svtter's Blog

Diffusion Model

Sat, 05 Apr 2025 21:51:38 +0800

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17



noise image -> -------------
 | |----> cleared image
step (int) -> | denoiser |
 | |
 --------------|
```The charm of deep learning lies in the fact that once a new task achieves improved performance with a certain architecture, many other tasks can refer to this architecture and benefit from it.

I believe the diffusion model is a typical example. Although I do not conduct research on diffusion models and currently have no related projects, there is no harm in understanding this network architecture.

The diffusion model is one that benefits from the image processing process.

By learning the reverse process of adding noise to images, the diffusion model acquires the ability to generate images from noise.

![noise-dog](noise-dog.png)

To enable the model to achieve better performance, the denoising step of the model is included as one of the inputs.

Using uv

Sun, 30 Mar 2025 14:33:34 +0800

Recently, I’ve started using uv extensively instead of pdm.

knowledge piece

uvx could replace pipx.

The uvx command invokes a tool without installing it.

For example, to run ruff

1

uvx ruff

Python Timezone

Fri, 28 Feb 2025 17:46:29 +0800

Regardless of the current server settings, output the time in Asia/Shanghai.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12


import datetime
import pytz

utc_now = datetime.datetime.utcnow() # Get current time in UTC
utc_timezone = pytz.utc
utc_now = utc_timezone.localize(utc_now) # Localize the time as UTC

# Convert to another timezone, e.g., 'America/New_York'
new_timezone = pytz.timezone('Asia/Shanghai')
new_timezone_time = utc_now.astimezone(new_timezone)

print(new_timezone_time.strftime('%Y-%m-%d %H:%M:%S %Z%z')) # Display time in the new timezone

Cuda and Paddle

Tue, 11 Feb 2025 15:41:18 +0800

Configuring a deep learning environment is a hurdle many struggle to overcome. However, with large models, troubleshooting and pinpointing issues can be significantly faster.

I spent some time adapting an older version of PaddlePaddle and finally got it working. Here, I’ll share an article documenting the process.

In Docker images, many CUDA 11-based images fail to run in a CUDA 12 environment. The exact reason isn’t entirely clear to me. In such cases, you can opt for a CUDA version that matches the major release.

To avoid affecting the environments of others on the server, do not update the NVIDIA driver. Instead, install your own CUDA version and modify the environment variables to change the system’s CUDA.

1
2
3
4


# CUDA_VERSION=11.7
export CUDA_HOME="/usr/local/cuda-$CUDA_VERSION"
export LD_LIBRARY_PATH="$CUDA_HOME/lib64:$LD_LIBRARY_PATH"
export PATH=$CUDA_HOME/bin:$PATH

Apply this environment variable, then check nvidia-smi to see the version change.