4/ Faster to train! Training with LoRA can easily be 4-8x faster than full fine-tuning.
MACHINE LEARNING
-
LoRAX enables serving hundreds of LoRAs from single GPU
By
–
5/ Unlocks multi-model deployments. Serve 100s of LoRAs from a single #GPU deployment with #LoRAX.
-
LoRA Fine-Tuning with Predibase Speeds Up Inference 3x
By
–
6/ They are only getting better! LoRA fine-tuning with Predibase makes inference 3x faster compared to the base model.
-
How LoRA is Transforming LLM Fine-tuning Techniques
By
–
1/ Curious how #LoRA is changing fine-tuning? Read our latest blog post to learn more! #llm #finetune
-
LoRA Achieves Comparable Performance to Full Fine-Tuning
By
–
2/ No trade-off in performance. A 2021 study demonstrated comparable performance between LoRA and full fine-tuning.
-
LoRA Training Parameters: Minimal Memory Footprint at 0.2%
By
–
3/ Extremely low memory footprint. LoRA trainable parameters are only 0.2% of total amount of model parameters.
-
Teaching GANs: From Evangelist to ML4A Practice
By
–
For the next two years, I became something of an evangelist for GANs, incorporating them into my teaching practice called @ml4a_ and showing people how to train their own GANs using Jupyter notebooks. I wrote about that experience here: https://
medium.com/@genekogan/mac
hine-relearning-8cd57ba242
… Until NVIDIA -
GANs latent space exploration through character animation loops
By
–
GANs biggest contribution was introducing the notion of "latent space" — the space of all possible images the GAN could produce.
— Gene Kogan (@genekogan) 10 juin 2024
Here, I generated short videos making "loops" around the latent space of individual characters, exploring their variations. pic.twitter.com/qOFpCaZvqzGANs biggest contribution was introducing the notion of "latent space" — the space of all possible images the GAN could produce. Here, I generated short videos making "loops" around the latent space of individual characters, exploring their variations.
-
Deep Learning on Handwritten Chinese Characters Dataset
By
–
So I decided to concentrate my efforts on teaching myself deep learning and trying to reproduce their success on different datasets. Looking for novel datasets to train on, I stumbled upon one of almost 1 million images of handwritten Chinese characters, which had been mainly
-
Machine Learning and Creative Coding Convergence Journey
By
–
In 2015, my career looked something like machine learning engineer by day, generative artist by night. A bit dissatisfied by the incrementalist pace of pre-deep machine learning, I found an outlet in the creative coding world. By luck or fate these two worlds converged for me