KittyLM
Independent language-model research from tokenizer and architecture through training and evaluation engineering.
A navigable map of KrazyKitty's language-model, dataset, and agent-systems research—built around provenance, reproducibility, and honest evidence boundaries.
Each node distinguishes published artifacts from planned work. Select a node below for its current evidence, limits, and next approval gate.
Independent language-model research from tokenizer and architecture through training and evaluation engineering.
Dataset provenance, permission, preprocessing, quality, and evaluation-resource research.
Tool use, retrieval, task execution, agent evaluation, and sandboxed safety boundaries.
Curated Hugging Face repositories and papers grouped into four public research tracks.
Sources, revisions, licences, permissions, transformations, and integrity hashes remain attached to research artifacts.
Complexity enters only when a controlled experiment or architectural boundary justifies it.
Publishing, external writes, personal data, and consequential actions stay behind explicit approval gates.
Verified, proposed, planned, failed, and unknown are visible states—not interchangeable marketing language.
Collection contents are real Hugging Face repositories and papers. Inclusion supports study; it does not replace each resource's licence, access, or usage review.
Attention, bidirectional pretraining, autoregressive modelling, and compact open baselines.
VERIFIED PUBLICCompute allocation, parameter-efficient adaptation, quantised finetuning, and preference optimisation.
VERIFIED PUBLICOpen corpus construction, transparent filtering, benchmark governance, and contamination awareness.
VERIFIED PUBLICReasoning with actions, tool learning, reflective agents, and executable software-engineering evaluation.