# Continual-Intelligence/SEAL

Self-Adapting Language Models

Repository: https://github.com/Continual-Intelligence/SEAL
Canonical: https://ross.abutalabs.com/products/continual-intelligence-seal
Homepage: https://arxiv.org/abs/2506.10943
Language: Python
License: MIT
License Family: permissive
Last push: 2025-08-01T20:06:35+00:00

## Health v2 (maintenance only)
Score: 34/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 34, release rhythm 35, longevity 31
- inputs: {"age_days": 447, "days_push": 397, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1850, forks 320 (observed 2026-08-28T04:05:44.212018+00:00)

## What it is
SEAL (Self-Adapting LLMs) is a research framework from MIT CSAIL that trains language models via reinforcement learning to generate their own finetuning data and update directives ('self-edits') in response to new inputs. It includes code and experiments for two domains: incorporating new factual knowledge and adapting to new tasks from few-shot examples.

## Use cases
- train an LLM to self-adapt its weights to new knowledge
- research continual learning for language models
- incorporate new factual knowledge into a finetuned model
- adapt a model to new tasks from few-shot examples
- reproduce the Self-Adapting Language Models paper experiments

## When to choose
- you are an ML researcher exploring self-directed model adaptation or continual learning
- you have GPU resources (2+ A100/H100) and want to reproduce or extend the SEAL paper
- you want to experiment with RL-trained self-edit generation for LLM finetuning

## When to avoid
- you need a production-ready finetuning or inference tool
- you lack multi-GPU hardware or SLURM cluster access
- you want a plug-and-play library rather than research code

## Facets
- artifact type: framework
- maturity: experimental
- function: llm-training, machine-learning, reinforcement-learning
- domain: large-language-models, machine-learning
- platform: python
- tags: self-adapting-llms, reinforcement-learning, continual-learning, research-code, mit-csail, slurm, research, linux, gpu

## Member repositories
- Continual-Intelligence/SEAL (main) score 34

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:44.212018+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T03:16:56.473147+00:00, confidence not recorded.
  - readme: https://github.com/Continual-Intelligence/SEAL (fetched 2026-08-28T04:05:44.212018+00:00, sha 68a436bddb36)
  - homepage: https://arxiv.org/abs/2506.10943 (fetched 2026-08-29T10:56:27.311387+00:00, sha ee77ccd8ec44)
  - site_page: https://info.arxiv.org/about/donate.html (fetched 2026-08-29T10:56:27.320824+00:00, sha cca9c3a11c56)
  - site_page: https://info.arxiv.org/about/ourmembers.html (fetched 2026-08-29T10:56:27.324535+00:00, sha 47cbc55ff1de)
  - site_page: https://info.arxiv.org/about (fetched 2026-08-29T10:56:27.326808+00:00, sha a1f16f915a9a)
  - site_page: https://info.arxiv.org/labs/index.html (fetched 2026-08-29T10:56:27.322799+00:00, sha b14a8d05a0ec)
- Data as of 2026-08-30T08:39:29.467469+00:00.