DaertML

easy_jepa. Trying to make it easier to use JEPAs :)

context_distillation. Framework to achieve context distillation in LLMs

tinyformer. Code to run inference of Transformer models without any Python dependency, and in a simple manner

daert-agents. Repository with independent LLM based agents to perform tasks given by a prompt.

bert_has_a_mouth. Repository to experiment with the possibility of generating text with BERT models

train_llama2. Repository with code to support the tutorial on the Daert Academy: https://www.youtube.com/watch?v=0Kc2g9GmJYY&t=1091s

scratchml. Machine Learning and Deep Learning framework developed from scratch in Python.

open_alphacode2. Repository for an Open Source implementation of AlphaCode 2 from Google Deepmind. Check the video series: https://www.youtube.com/watch?v=NHKI2L9lqyI

llm-computer. Open source implementation of the ideas depicted in https://www.percepta.ai/blog/can-llms-be-computers

bfn_textgen. Repository for experimenting with Bayesian Flow Networks for text generation

bert_prompt_engineering. Repository to develop great prompt engineering techniques for BERT models

neurons.c. Collection of C codes to run different types of neural nets in plain C.

fatkernels. Collection of fat kernels for deep learning writen in triton

llm_eval. Repository meant for a framework to evaluate LLM generations and generate prompts synthetically using metaprompts

llm_alignment. Repository for code and discussions on how to align LLMs and other findings

som_textgen. Repository for experimenting with Self Organizing Maps (SOMs) for text generation

llm_ctx_win_eval. Framework to evaluate the Context Window of LLMs

zerodep-agi. Developing AGI without dependencies, just python3 :)

databricks_notebooks. List of Databricks notebooks used in the course https://www.youtube.com/watch?v=nuM-wT7IEZ8&list=PLhyOIUHD0Y0YQPMYbXDOSsSd6ZjYhH96P

svm_textgen. Repository for experimenting with Support Vector Machines for text generation

train_llm_multinode. Repository for developing scripts for LLM training in a GPU cluster (multiple nodes).

ai_reading_list. Books and papers to read to get into AI without prior experience :)

openberry. Implementation of different methodologies to get towards the 🍓 behavior

daert-stack. Stack to deploy LLM based applications in production

dato_curado. Repository meant for LLM data curation pipelines.

daert-agent-orgs. Repository with agentic organizations in different frameworks

what_is_next_for_llms. This repo serves as a space for posting on future trends that we are exploring.

langshell. Bringing together the LLMs ecosystem with the Unix shell philosophy and tools.

dafonte-stack. LLM enhanced big data platform. Architecture by @pdafigl (thus the name of the repo ;))

tlcute-attention. Open source implementation of the QiMeng-Attention paper: https://www.arxiv.org/pdf/2506.12355

llms_are_trained_cpus. Repository to experiment with LLMs as if they were CPUs by training and testing them with different embeddings

neurons.py. The same goal as neurons.c but in python; cuz "damn C got hands"

model_fusion. What if we could train tiny models and fuse them with any other model?

llm-bi. Business Intelligence and Data Querying that uses natural language processing with LLMs, so that you can forget about SQL

siliconbrains. Open Source TPUs for the masses

daert-tensorop. Open source implementation of the QiMeng-TensorOp: https://arxiv.org/pdf/2505.06302

idm_video. Training Inverse Dynamics Models from video in an automated manner

DaertGEMM. Open source implementation of the QiMeng-GEMM: https://ojs.aaai.org/index.php/AAAI/article/view/34461/36616

compute_layers. Repo meant to experiment and research how different kinds of AIs and algorithms lay down in the compute layers

evoltriton. Writing Optimized Triton Kernels using DSPy and local LLMs

agigym. Test and train your LLMs, VLMs and other AIs in the AGI Gym

llm_creative. Repo for studying the mechanisms to improve LLMs for creative writing. Trying to discover the magic behind GPT4.5

triops. Tiny environment to train LLMs, that works along CUDA and Apple Silicon

daert-infra. Repo to deploy ML infrastructure with ease at scale

agent_harnesses_over_time. A repo to analyze the evolution of agentic harnesses over time

simple_ai_and_algorithms. Repository that implements genetic algorithms, evolved neural networks, simple algorithms and reinforcement learning to solve games

adk-examples. Repo with examples of ADK agents

nested_learning. Tiny implementation of Nested Learning ideas from Google Deepmind on already pretrained LLMs

daert_cua. Open source Computer Use Agents using different technologies

mech_interp. Mechanistic Interpretability of LLMs experiments and reproductions

data_salad. Make "data salads" and evaluate the mix before you train your LLM

daert-gyms. New ways and ideas of training AI models on RL environments. Original ideas came from John Carmack's talk: https://www.youtube.com/watch?v=iz9lUMSQBfY

poorman_autoresearch. Automated neural networks training guided by LLMs, following the same ideas as the karpathy's "autoresearch" repo, but doing this without a coding agent, and just a harness inside the code that can call tools

llm-bi-multiagent. Multi agent system in which a different data store and catalog is assigned to each agent; easing the process of data querying and integration

ttt-discover-unsloth. An implementation of the TTT Discover paper using unsloth

daert-cloud. Infra to build things in cloud and k8s

ralph-loop-template. Template to run ralph loops on claude code on your repo

llm-gw. Repo to host production/dev environments for enterprise LLM systems