Hands-On Build Guide: Batched LLM Token Sampler in Pure NumPy
Build a pure‑NumPy batched token sampler with temperature, top‑k, nucleus, and repetition penalty; ready to run and extend for CV demonstrations.
Posts
Building a FlashAttention-Style Tiled Self-Attention Block in NumPy
This post walks through building a FlashAttention-style tiled self-attention block in NumPy, covering online softmax, KV caching, and performance benchmarks.
Building a Production-Grade Byte-Level BPE Tokenizer in Pure Python
A step-by-step guide to building a byte-level BPE tokenizer with regex pre-tokenization, trie encoding, and JSON serialization in pure Python.
User Safety: Safe-by-Design Patterns for Modern Applications
A practical guide to safe-by-design user-facing systems, covering authentication patterns, fault tolerance, and real-world failure modes at scale.
Hands‑On BPE Tokenizer: Pure‑Python Incremental Merge Learning with Trie Lookup
This post walks you through building a from‑scratch BPE tokenizer in pure Python, complete with online incremental merge learning and a trie for fast tokenization. You’ll end up with a runnable library you can showcase on LinkedIn or a résumé.