I think this is the best introduction to evals and how they should be implemented to different types of agents.
https://t.co/qFzJPrcjgF
#LLM#eval#harness
Hitting the VRAM wall in PyTorch? To maximize vision pipeline throughput, you need bare-metal CUDA.
Here is my deep dive on eliminating memory thrashing using Shared Memory tiling, C++ guardrails, and branchless Halo boundary math:
https://t.co/1Nq4PTjXxM
#CUDA#AI
Hitting the VRAM wall in PyTorch? To maximize vision pipeline throughput, you need bare-metal CUDA.
Here is my deep dive on eliminating memory thrashing using Shared Memory tiling, C++ guardrails, and branchless Halo boundary math:
https://t.co/1Nq4PTjXxM
#CUDA#AI
The more I learn about LLMs, I have realised that LLMs are just a tool and as a software developer it becomes our duty as to how efficient we use them to make it more efficient for our specific use-case.
Sex is a very intimate and sacred act
Your body is a temple and you shouldn't share it with someone who is not learning about robotics, edge intelligence, CUDA programming, Verilog, physical ai and TinyML
@tweetsbyapurv What if there was a news aggregator dashboard with a real time dedicated rating engine for responsible entity. Showing sentiment of people.
Journalism with real voice of people not a bought anchor promoting propagandas.
Democracy,
Most citizens only get a voice once every 5 years.
Everything in between?
Laws are passed. Policies are shaped. Decisions are made.
That feels broken.
What if every citizen had a continuous voice — not just to speak, but to be understood at scale?
The fact that this article is still helpful for the version mismatch issue on Google Colab.
Thanks @sshkhr16 for this one-stop solution.
https://t.co/C6iQCB0pta
The hypocrisy of left is so much evident.
What happened to "Hindus" in Bangladesh is communal genocide. There is undeniable evidence showing the situation, yet all those so-called human rights activist stay silent...
#bangladeshihindu#Bangladesh#HindusUnderAttackInBangladesh