Evin Tunador

triton_docs_tutorials. making the official triton tutorials actually comprehensible

minLlama3. a simplified version of Meta's Llama 3 model to be used for learning

minGemma. a simplified version of Google's Gemma model to be used for learning

templateGPT. customizable template GPT code designed for easy novel architecture experimentation

gpt-lab. cheap & easy LLM experiments for amateurs (alpha)

arxiv-summaries-workflow. The automation scripts I use for my weekly arxiv paper summary videos

next-concept-predictor. a WIP architecture designed to allow transformers to think in a manner without tokens

FractalFormer. A GPT with self-similar nested properties

autograd_engine_tutorial. A from-scratch multi-difficulty-level tutorial on how pytorch, tensor flow, Jax, etc work

gpu_bpe. tutorial for training a BPE tokenizer on multiple GPUs

auto-video-editing-suite. All the tools that allow me to never ever open up Final Cut

minGrok. A minimal implementation of the architecture used in XAI's Grok-1 designed for teaching purposes

matryoshkaGPT. GPTs inside of GPTs like Russian nesting dolls

daily-paper-summaries-workflow. My workflow for my Daily AI Paper Summaries on YT https://youtube.com/@Tunadorable\nhttps://youtube.com/playlist?list=PLPefVKO3tDxP7iFzaSOkOZnXQ4Bkhi9YB&si=ndXJfDFfMkE_b8w7

nGPT. My attempt at replicating/teaching Nvidia's Normalized-GPT

micro-GPT-sandbox. A place to experiment with novel GPT architecture edits & compare them to one another

medusas_uglier_sisters. Jupyter Notebook

hierarchical_embeddings. a GPT experiment that hierarchically separates meaning across embedding vectors

fractal-head-attention. an example novel GPT experiment as part of my micro model sandbox project

TikTok_Successor_Proposal. A casual powerpoint describing my proposal for a TikTok successor

FutureFormer. An alternative to next-token prediction that allows the attention mechanism to use its own future predictions

AI-s_non-anthropomorphic_path_to_understanding. A newer version of this paper can be found at https://github.com/evintunador/Strictly_Defining_-Understanding-

SenPer. a computational evolutionary psychology experiment I did relating to sensation, perception, and the truth of reality

nGPT-MoScales. A tiny edit to nGPT and some custom kernels to speed it up

Next-Concept_Prediction_v8.7_conceptual_walkthrough. An old architecture I was working on. There was a BIG fundamental flaw

Pres_Pol_and_US_Econ. paper I wrote for a class on advanced time-series analysis. when I submitted to a journal they said it's good enough but not on-topic enough for them. didn't have the patience to keep submitting

embedding-space-fractal-test. code from my video https://youtu.be/OYKH2xrtd7U

DML-OpenProblem. HTML

Psyc_400_Personality_paper. a paper explaining why there exists variation in personality traits

Subsettable_Enmeshed_Cooperative_Clusters. basically I re-invented the DAOstack without knowing it

Strictly_Defining_-Understanding-. An old philosophy of AI paper I was writing

learningDiffusionModels. Jupyter Notebook

Decentralizing-Firm-Decision-Making. an idea

wiki_graph_extractor. turn wikipedia dumps into a text-attributed directed acyclic graph dataset

Meta-Evolution_-_Decentralized_Symbiotic_Alignment. Old silly paper I was writing. Using as my first test of the overleaf -> GitHub sync

MemoryMosaics. A small replication of the Memory Mosaics model (alternative to transformer)

Patents-AI-Bad. paper that fell apart as I lost interest in econ & learned that simple AI architecture improvements can't be patented. Still some interesting ideas in here

Self-Replicating-Trust-Funds-UBI-as-a-Virus. an idea

tunadorableGPT. creating a model to emulate me in all ways. first version just responds to YouTube comments

tagseq2tagseq. Python