I'm a Python Programmer, AI Enthusiast, and a Co-founder of NousResearch.

My work primarily involves AI and Data Engineering, contributing primarily by building open source AI agents, releasing open source Large Language Models (LLMs), datasets, synthetic data pipelines, and RL environments.

My current focus is Hermes Agent — the agent that grows with you. A terminal-native, extensible AI coding & personal agent with persistent memory, skills, and multi-platform gateways (Telegram, Discord, Slack, WhatsApp, Signal, iMessage, and more).

• Hermes Agent - Main repository. Runs locally, supports every major LLM provider, and learns from every session.
• Hermes Agent Docs - Full documentation, quickstart, and provider setup guides.

I've contributed significantly to the development of several opensource LLMs and research projects under Nous Research.

Here are a few of them:

• Hermes-4 Series - Nous' latest most powerful LLM yet, with hybrid reasoning capabilities.
• Atropos - Nous' framework for RL Environments with LLMs — collecting and evaluating LLM trajectories through diverse environments.
• Hermes-3 Series - The third Hermes model series release.
• Nous-Hermes-2-Mixtral-8x7B
• Nous-Hermes-2-Yi-34B - Hermes 2 on Yi-34B.
• Nous-Hermes-Llama2-13b - A Hermes model built on llama 1 and llama 2.
• GPT4-x-Vicuna-13b - A Vicuna model built on GPT-4.

On my personal huggingface, Teknium, I have released several models, including my work on Replit-3b Model & OpenHermes:

• DataForge - Economics - A dataset built by my data synthesis pipeline (not public), DataForge
• OpenHermes 2.5 Mistral 7B - Most powerful Open Hermes, with much improved coding skills than OpenHermes 2
• OpenHermes 2 Mistral 7B - Version 2 of the Open Hermes series.
• OpenHermes 13B - An Open Sourced version of Nous-Hermes!
• OpenHermes Dataset - The publicly available version of Hermes' dataset.
• Replit-Instruct 3B - This model doubled the code performance of the LLM.

I've been part of several intriguing projects on GitHub. Here are a few of them:

• LLM-Benchmark-Logs - A repository full of benchmarks I've done on various LLMs, originally inside of Nous' discord but it became too disorganized, so now lives on Github.
• LLM-Logbook - A temporary project that became too expensive to do, collection of responses for 100 random crowdsourced prompts to various LLMs.
• GPTeacher - A collection of modular datasets generated by GPT-4, for training LLMs.
• RawTransform - A repository of prompts and Python scripts for intelligent transformation of raw text into diverse formats.
• stanford_alpaca-replit - Modified Stanford-Alpaca Trainer for Training Replit's Code Model.
• alpaca-roleplay-discordbot - An LLM discord bot that roleplays!
• alpaca-discord - A Simple Discord Bot for the Alpaca LLM.

Have worked on researching, planning ablations, and cleaning/filtering the dataset for:

• StableBeluga/Free Willy 2 - Orca replication on 70b Llama-2
• StableBeluga/Free-Willy-1 - Orca replication on 65b Llama-1

Both are 10% Orca replications trained on Llama-1 and Llama-2 70B. Also working on domain expert knowledge and task distillation.

Working with the Open Orca team on data cleaning, networking, ablations, and more:

• Open Orca HuggingFace Repo - An Open Sourced Orca paper replication

• Twitter: https://twitter.com/Teknium
• Discord: Teknium