Parameter Golf v3: ~1.151 BPB in 16MB — Int6 QAT, BigramHash, SmearGate, SWA, zstd-22
-
Updated
Mar 29, 2026 - Jupyter Notebook
#
parameter-golf
Here are 8 public repositories matching this topic...
OpenAI Parameter Golf — SOTA 0.3958 BPB (3-seed mean). Beat prior best (0.4416) by 0.0458. Causal BackoffNgramMixer. PR #1094.
-
Updated
Mar 30, 2026 - Python
OLYMPUS: The Agora — Community companion for OpenAI Parameter Golf. Live leaderboard, compliance engine, technique map, funding transparency, compute guide.
-
Updated
Mar 31, 2026 - Python
Exploratory compact LM research branch derived from LUMI-Arch, focused on structure-sensitive generalization under tight model constraints.
-
Updated
Mar 26, 2026
Can a language model learn algorithms instead of memorizing patterns? Exploring extreme parameter efficiency in vocabulary space — 350K params, no embedding, no output projection, every intermediate state readable as words.
reinforcement-learning pytorch language-model register-machine meta-learning interpretable-ml parameter-golf
-
Updated
Mar 25, 2026 - Python
Sparse I/O bottleneck language model using C. elegans connectome wiring + JEPA + SigREG. Parameter Golf submission: 1.79 BPB in a 2MB artifact.
-
Updated
Mar 30, 2026 - Python
Byte-level adaptation-native 16MB language model for OpenAI Parameter Golf
-
Updated
Mar 25, 2026 - Python
A practical starter repo and working setup for the OpenAI Parameter Golf competition
-
Updated
Mar 25, 2026 - Python
Improve this page
Add a description, image, and links to the parameter-golf topic page so that developers can more easily learn about it.
Add this topic to your repo
To associate your repository with the parameter-golf topic, visit your repo's landing page and select "manage topics."