datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
matryoshka_ablation_results_0125metrics-outputs-pcfg-matryoshka-layer-03-token-cachemetrics-outputs-pcfg-matryoshka-layer-02-token-cachematryoshka-diffusion-models-paper-examples
Matryoshka Diffusion Models - paper examples
This dataset contains the 1024x1024 images included in the Matryoshka Diffusion Models
paper.
Arxiv: https://arxiv.org/abs/2310.15111
metrics-outputs-pcfg-matryoshka-fmt-2400-s1-token-cachemetrics-outputs-pcfg-matryoshka-layer-01-token-cachemetrics-outputs-pcfg-matryoshka-layer-00-token-cachemetrics-outputs-pcfg-matryoshka-fmt-2308-s2-token-cachemetrics-outputs-pcfg-matryoshka-fmt-2400-s2-token-cachemetrics-outputs-pcfg-matryoshka-fmt-2400-s0-token-cachemetrics-outputs-pcfg-matryoshka-fmt-0000-s2-token-cachenla-qwen36-27b-matryoshka-data
NLA training data — Qwen3.6-27B layer 42 (matryoshka)
EasyNLA-format training data for the matryoshka NLA on
Qwen/Qwen3.6-27B: layer-42 last-token residual
activations (d=5120, raw/unnormalized) of ~440k Ultra-FineWeb text prefixes, paired with
Claude Sonnet 4.6 next-token analyses from
ceselder/nla-matryoshka-warmstart-sonnet46.
file
rows
use
av_sft_shuf.parquet
219,922
verbalizer SFT (prompt messages + bullets response + activation)
ar_sft_shuf.parquet
219,507… See the full description on the dataset page: https://huggingface.co/datasets/ceselder/nla-qwen36-27b-matryoshka-data.business-frameworks
Business Frameworks
Operating judgement for running a business, distilled for agents. Query it like a consultant, pay per answer.
Scope: built for businesses from launch to about $50M in revenue; larger businesses are product two.
One document, written by an operator, for an agent that is running a business — and for an agent advising the human who does. It is structured so an agent reads the free top layer here and pays only for the node it needs; every leg ends in decision… See the full description on the dataset page: https://huggingface.co/datasets/Matryoshka-Paradigms/business-frameworks.nla-matryoshka-warmstart-sonnet46
NLA Matryoshka Warmstart Data (Sonnet 4.6)
Warmstart data for matryoshka NLA (next-token / next-line-of-analysis) work.
For each input text snippet, Claude Sonnet 4.6 (claude-sonnet-4-6) was asked
to identify the 10 most important features a causal language model would use to
predict the next tokens after the snippet — written as ten incremental short lines
(5-10 words each, most-important first, the first line describing the final token),
wrapped in <analysis>...</analysis>.… See the full description on the dataset page: https://huggingface.co/datasets/ceselder/nla-matryoshka-warmstart-sonnet46.matryoshka-sheaf
Paper + runnable code. This repository is the citable record of The Matryoshka Sheaf: An Interaction-First Algebra for Agent Substrates (Tej Desai, Intuition Labs). The formalism is runnable: matryoshka_sheaf.py makes both sheaf axioms self-checking. See also the ELI5 companion: intuitionlabs.tech/codebox/phi/35-the-matryoshka-sheaf.
The Matryoshka Sheaf: An Interaction-First Algebra for Agent Substrates
Tej Desai — Intuition Labs LLC
Abstract
An agent… See the full description on the dataset page: https://huggingface.co/datasets/intuitionlabs/matryoshka-sheaf.metrics-outputs-pcfg-matryoshka-fmt-1667-s0-token-cachenla-qwen2.5-7b-L20-matryoshka-warmstart-sonnet46
NLA Qwen2.5-7B L20 warm-start data — Sonnet-4.6 "matryoshka" explanations + activations
Re-warm-start dataset for the Natural Language Autoencoders
Qwen2.5-7B (layer-20) AV/AR pair. Pairs Qwen2.5-7B-Instruct layer-20 residual-stream
activations with the Claude Sonnet-4.6 explanations from
ceselder/nla-matryoshka-warmstart-sonnet46.
The source dataset is text-only (explanations keyed by custom_id, no vectors). This
dataset adds the missing activations: for each av-*/ar-* row the… See the full description on the dataset page: https://huggingface.co/datasets/syvb/nla-qwen2.5-7b-L20-matryoshka-warmstart-sonnet46.metrics-outputs-pcfg-matryoshka-fmt-2308-s1-token-cachemetrics-outputs-pcfg-matryoshka-fmt-1667-s1-token-cacheDreadPoor__Matryoshka-8B-LINEAR-details
Dataset Card for Evaluation run of DreadPoor/Matryoshka-8B-LINEAR
Dataset automatically created during the evaluation run of model DreadPoor/Matryoshka-8B-LINEAR
The dataset is composed of 38 configuration(s), each one corresponding to one of the evaluated task.
The dataset has been created from 1 run(s). Each run can be found as a specific split in each configuration, the split being named using the timestamp of the run.The "train" split is always pointing to the latest results.… See the full description on the dataset page: https://huggingface.co/datasets/open-llm-leaderboard/DreadPoor__Matryoshka-8B-LINEAR-details.metrics-outputs-pcfg-matryoshka-fmt-0000-s0-token-cachemetrics-outputs-pcfg-matryoshka-fmt-0000-s1-token-cachemetrics-outputs-pcfg-matryoshka-fmt-1667-s2-token-cachemetrics-outputs-pcfg-matryoshka-fmt-2308-s0-token-cache