AgentSea’s Post

Trellis Research breaks down the effects of various optimizers on memory and helps you cram more model into less space. Brilliant tutorial that makes it super easy and simple to understand the differences between Adam, Adam 8 Bit, Adafactor and Galore. https://lnkd.in/d6yZwFPp

QLoRA for efficient finetuning of Quantized LLMs, in a single GPU. https://lnkd.in/eApRTzm5

Our new paper on quantum error correction is up on Arxiv https://lnkd.in/egBAyigt. It illustrates new circuit design algorithms for Clifford logical operations on the family of Hypergraph Product codes, which are one of the most popular code families of interest in near term intermediate scale quantum devices.

My team and I have developed a method to efficiently extract Hamiltonians from aBN and hBN structures using GNNs, which drastically cuts computational costs compared to traditional DFT. Enter Hamiltonian Magic!  We owe a huge thanks to Constructor for optimizing our workflow for better efficiency and reproducibility.  Reproducibility is key. Stop struggling with research papers, and let Constructor simplify the process: https://lnkd.in/d7CxjXJJ  In this video, Andrei Voicu Tomut will demonstrate how to setup and use Constructor, import a project from Github, and construct workflows that will allow you to iterate faster. 

My latest blog is now out! It discusses two GPU-friendly methods for sampling of Gaussian distribution, including source code and performance evaluations. Get it here: https://lnkd.in/dDad4tWc

MYTH: vector databases only perform ANN search. FACT: vector databases support dense ANN search, sparse vector search, metadata filtering, multi-vector support, RBAC, GPU acceleration, etc. #mythvsfact #vectordatabases #vectorsearch

Time to find out the limits of what 1TB of memory can do in the homelab.

MYTH: vector databases only perform ANN search. FACT: vector databases support dense ANN search, sparse vector search, metadata filtering, multi-vector support, RBAC, GPU acceleration, etc. #mythvsfact #vectordatabases #vectorsearch

Admittedly, IMHO, compute-bound optimizations are something that even the most advanced programmers don't necessarily consider. If I have to multiply two 1024 matrices together, I will have just over 1 million multiplications. How long my computer will take depends on the order of our multiplications!? Forget how your textbook told you to multiply; we have more important things to consider than getting the correct answer (speed). The video below explains the memory access patterns of your RAM and CPU and how/why that can be leveraged to optimize matrix multiplication.

SECOND PART:-> oscillates for these toplocial shapes to prduice gertaors to difefernt veiws tat i ahave disccues from strat post to end post means lensing comptuing tunenling to fiber comuting tunneling for computing we have numebrs oscillations then with that works their with help of totploigcal oepration of mathamtcals