roadmap

phase i - launch - Q1 2025

• launch token

• launch socials

• write litepaper

• launch relign framework v1 - post train a small 1b parameter

• launch developer bounty program

phase ii - develop a strong reasoning library - Q1 2025

• implement GRPO (Deepseek r1) + evaluation math benchmark on 1B/8B model

• write comprehensive docs detailing framework architecture and usage

• whitepaper explaining the technical architecture of the framework

• modularize framework/unit testing - enabling quick, collaborative development, as well as faster implementation of new algorithms

• expand on the reasoning stack

  • judge models, allowing one model to teach another model (model distillation)

  • CoT/MCTS strategies

  • implement reasoning trace analytics - gain new insights from training runs

• grow dev community - bounty program, discord, live coding sessions, offline meetups

• launch research program, exploring research collaborations implementing the relign toolkit among:

  • agent framework

  • compute provider

  • data owner

phase iii - scaling up - Q2 2025

• cloud runner for large-scale remote post-training

• 70B+ parameter reasoning model launch - allowing open source access

• launch relign toolkit, allowing ai devs to develop their own task/reward models

phase iv - RaaS - Q3 2025

• build reasoning-as-a-service platform, enabling businesses to train reasoning models on their own specified use cases, burning $RELIGN to access compute resources.