Exploring einsum
A comprehensive guide to the einsum function covering its syntax and simple use case.
Ashvanth.S Blog
Personal site and research notes by Ashvanth.S
Latest posts
Publish packages to PyPI using Github Actions
Publishing Python packages to PyPI has become more secure and automated thanks to the Trusted Publishers mechanism . I recently adopted this for my package sollu , and it was the second time where I got to use the Trusted Publishers mechanism to publish a Python package, the first being my image caption CLI tool imgcap . The trusted publishers mechanism allows us to publish these Python packages,…
Introducing sollu - AI powered dictionary
I was reading Dostoevsky’s The Brothers Karamazov, and all of a sudden stumbled upon this word called “cupola”, chances are you might know this word. I didn’t and did what I always do, googled it. Ended up knowing that it meant ( roof or part of a roof shaped like a dome ). What began as an innocent dictionary search opened the door to digital distraction and notifications beckoning, headlines…
Refresher on how LoRA works
As the scale of deep learning models grows, so does the computational and memory cost associated with fine-tuning these models for specific tasks. Fine-tuning a large model can involve adjusting billions of parameters, which is resource-intensive and often impractical. Parameter-efficient fine-tuning (PEFT) techniques have been developed to address this issue, allowing for model adaptation with…
Simplicity of Byte Pair Encoding
🌟 Highlight Based on this blog post , got an opportunity to give a talk based on Simplicity & Effectiveness of Byte Pair Encoding (BPE). Here is the link to the talk :🎙️ Cohere for AI Imagine teaching a kid to read and understand text. Chances are we start with individual letters, then combine them individually and let the child read. Then over time, you see the dividends right in front of you.…
Wrapping your head around Self-Attention, Multi-head Attention
The concept of Attention is essential to the working of the Transformer. Through this concept of attention, models focus on the relevant information to give predictions. Take an example of a sentence The sky is ___ and world is round. What words might you use to fill in the blanks for the above sentence? Given the context, it is quite evident to see words like “hesitating”, “destroyed”, and…