[ your name ] · reproduction log

I rebuild deep learning papers from scratch and publish every number I get.

[ Two lines about you. Who you are, what you're studying, and why you decided to reimplement papers instead of just reading them. ]

01attention weight
fig. 1 — causal self-attention, synthetic

about

[ Your Name ]

[ First paragraph. Where you study or work, what you focus on, and what you've built. Write it like you'd say it out loud — no buzzwords. ]

[ Second paragraph. Why reproduction specifically: what you learn from rebuilding a paper that you can't learn from reading it. What you want to do next. ]

the log

Reproduction ledger

Every paper I've rebuilt, with the number the authors reported next to the number I actually got. The gap is the interesting part.

paper reportedreproΔ
arXiv:1706.03762 ↗ done

Attention Is All You Need

[ One or two sentences: what you built, what broke, what surprised you. This is the part people actually read. ]

→ code → writeup
reported[ 00.0 ]
repro[ 00.0 ]
Δ[ ±0.0 ]
arXiv:1512.03385 ↗ training

Deep Residual Learning for Image Recognition

[ What you're currently running, and what you're unsure about. ]

→ code → writeup
reported[ 00.0 ]
repro
Δ
arXiv:2010.11929 ↗ queued

An Image is Worth 16×16 Words

[ Why this one is next on the list. ]

→ code
reported[ 00.0 ]
repro
Δ

stack

What I build with

PyTorch Python CUDA C NumPy Weights & Biases [ add yours ]