# AutoArk/TinyEngram

Research of DeepSeek Engram Architecture based on Qwen-3 and Stable Diffusion series.

Repository: https://github.com/AutoArk/TinyEngram
Canonical: https://ross.abutalabs.com/products/tinyengram
Language: Python
License Family: other
Topics: deepseek, engram, llm, qwen, transformer, peft, fine-tuning, deepseek-ai, llm-memory, lora, pytorch, stable-diffusion, tinyengram, memory-injection
Last push: 2026-05-21T06:01:13+00:00

## Health v2 (maintenance only)
Score: 53/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 83, release rhythm 35, longevity 15
- inputs: {"age_days": 222, "days_push": 104, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1191, forks 81 (observed 2026-08-28T04:03:55.729928+00:00)

## What it is
TinyEngram is an open research project exploring DeepSeek-AI's Engram architecture and memory injection as an alternative to LoRA for parameter-efficient fine-tuning of LLMs like Qwen and diffusion models like Stable Diffusion. It provides code, experiment logs, and technical reports showing Engram-based memory injection outperforms LoRA in parameter efficiency and resistance to catastrophic forgetting.

## Use cases
- inject new concepts into Qwen LLMs without full fine-tuning
- replace LoRA with Engram-based memory injection for PEFT
- add visual concepts to Stable Diffusion models lightweightly
- reduce catastrophic forgetting during fine-tuning
- reproduce Engram architecture experiments
- research parameter-efficient adaptation of transformers

## When to choose
- you want to experiment with Engram-style memory injection instead of LoRA
- you need composable, lightweight concept injection into LLMs or Stable Diffusion
- you are researching catastrophic forgetting and parameter-efficient fine-tuning

## When to avoid
- you need a production-ready, stable fine-tuning framework with support guarantees
- you require a permissively licensed dependency and cannot accept unclear licensing
- you just need standard LoRA/PEFT tooling without experimental architecture changes

## Facets
- artifact type: library
- maturity: experimental
- function: machine-learning, llm-training, llm-inference, image-processing, sdk
- domain: large-language-models, machine-learning, deep-learning, image-processing, artificial-intelligence
- platform: python, cross-platform
- tags: deepseek-engram, memory-injection, lora, peft, fine-tuning, qwen, stable-diffusion, research, transformer, catastrophic-forgetting, gpu

## Member repositories
- AutoArk/TinyEngram (main) score 53

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:03:55.729928+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 2026-08-30T06:22:39.276203+00:00, confidence not recorded.
  - readme: https://github.com/AutoArk/TinyEngram (fetched 2026-08-28T04:03:55.729928+00:00, sha 5ca8115763cc)
- Data as of 2026-08-30T08:39:29.467469+00:00.