Research of DeepSeek Engram Architecture based on Qwen-3 and Stable Diffusion series.
ENGRAFT: write facts into an LLM's n-gram (Engram) memory table as a removable overlay: 100 facts in a 9 MB file, no weights touched, verified on llama.cpp GGUF (Qwen3.8-Flash-Next).
🚀 Engram-PEFT: An unofficial implementation of DeepSeek Engram. Inject high-capacity conditional memory into LLMs via sparse retrieval PEFT without increasing inference FLOPs / DeepSeek Engram 架构的非官方实现。通过参数高效微调 (PEFT) 为大语言模型注入超大规模条件记忆，支持稀疏更新且不增加推理开销。
A user's facts become a few rows in a content-addressed memory table — not a rewrite of the model. Paper, code, data, and interactive site for "User as Engram" (arXiv:2606.19172).
tensorflow-engram: A Python package for Engram Neural Networks, adding biologically-inspired Hebbian memory and engram layers to TensorFlow/Keras models, supporting memory traces, plasticity, attention, and sparsity for neural sequence learning.
PyTorch research scaffold for RetNet, Block Attention Residuals, and hashed Engram memory
dsh-engram-relay 公开镜像（上游：dsh-external/dsh-engram-relay）：跨会话分层记忆 + 哈希×语义×因果超稀疏唤醒，13 个 engram_* 工具，BSD-3-Clause
Weaviate Engram integration plugin for Dify
Unified inference orchestrator: intent classification, adaptive compute routing, quality-gated self-improvement loop, SOTA web search pipeline, and multi-tier Engram memory management.
Hermes Engram — Hermes Agent 原生项目记忆插件：自动召回、可靠存取、会话恢复与压缩归档
Engram conditional-memory sparse LLM framework
A 53M-parameter ISO 20022 payment field extractor that trains from scratch on one consumer GPU, with a nine-check acceptance gate.
Per-lookup LLM telemetry for DeepSeek-style conditional memory, streamed into a single ClickHouse MergeTree table
本项目是一个基于 MuZero 和 Gumbel 搜索构建的三人八子棋（Eight-in-a-Row）AI 训练与对弈系统。该系统采用了 DeepSeek MLA 作为网络骨干，并深度集成了课程学习（Curriculum Learning）、联赛机制（League）、基于种群的训练（PBT）、Engram 情节记忆以及 EfficientZero 的一致性损失等高级强化学习技术。此外，项目支持同步与多进程异步训练，并提供了基于 FastAPI 的实时对弈服务以及基于 WebSocket 的可视化训练监控台。
🌍 Transform seismic data recovery with Promethium, an AI-driven framework for advanced reconstruction and geophysical data enhancement.
Engram-compatible memory provider for Hermes Agent: deterministic recall and capture on every turn. Community project, built with Gentle AI.
Read-only, privacy-filtered MCP server and skill for Engram memory. Works with Grok Bot, Cursor, Claude Code and Codex.
To associate your repository with the engram topic, visit your repo's landing page and select "manage topics."