LoreonLabsPlatform

Overview

Intelligence

Markets
Builders
Research
Ecosystems
Launchpads

Search

Python

noema

Small-scale experiments in latent-space reasoning for language models.

PythonEmerging

Stars

—

Forks

—

Contributors

—

Last push

2mo ago

Recent commits

Latest commits.

fast iteration tools: checkpoint analysis and early-stop training
374db27yhh2mo ago
phase 1: K=2 stop-grad with constant LR to probe decay-kills-grokking hypothesis
9396410yhh2mo ago
phase 1: add K=3 stop-grad config to map the K dependency
926c168yhh2mo ago
phase 1: add K=1 stop-grad config to probe latent chain length
93241f2yhh2mo ago
latent: stop-grad option and param-l2 diagnostic
e463ebdyhh2mo ago

latent: skip pos_emb at filled thought positions, add fixed-K config

fc9d5b4yhh2mo ago

emit run summary: grokking step, final acc, tail trajectory, json file

dbdcd03yhh2mo ago

add n=3 latent 150k config for grokking-speed comparison

9c31aa9yhh2mo ago

Top contributors

Builders behind this project.

No contributor data available.