You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
A decoder-only (GPT-style) transformer built from scratch in NumPy, no deep-learning framework. Every component (attention, LayerNorm, AdamW, backprop) is hand-derived and verified with gradient ch…
LoRA-fine-tuned Llama-3.2-3B for customer support, trained on Apple Silicon (MLX). Full pipeline: data prep, multi-metric eval (ROUGE-L, BERTScore), OpenAI-compatible FastAPI server, HuggingFace ad…