\(\rightarrow\) Algorithmic stuff:

• Boyer–Moore majority vote algorithm.

\(\rightarrow\) Transformer stuff:

• Why is it called KV cache: and not QKV cache
• RoPE: details, block expansion, and code
• Scalable Softmax
• [Not Ready!] Some trends to speed up autoregressive inference of LLMs (unifinished).
• Some notes on asyncio

\(\rightarrow\) Some CUDA/C++ learning notes:

• GPU architecture and warp scheduling.
• Occupancy, Compute intensity, and Tiling.
• DRAM banks and why it matters for code optimization.

\(\rightarrow\) Technical:

• Loading MNIST dataset in C++
• Handling checkpoints from terminal - some useful tricks.
• Quick reset of my compute pod.
• Speed up your migration to VIM.

\(\rightarrow\) Math:

• Matrix Differential Calculus.
• Convergence Rates of SGD in convex and non-convex cases.

Some GitHub repos:

• Training ML models with CUDA/C++: \(\rightarrow\) Github link.
• GPT2 factorized with (multiple) Kronecker Factors: \(\rightarrow\) Github link.
• Backpropagation from scratch \(\rightarrow\) Github link.
• Randomized Algorithms \(\rightarrow\) Github link.
• Reinforcement Learning \(\rightarrow\) Github link.