Flagships
My flagship models that actually work or are the best I have capable from a category currently.
Text Generation • 0.2B • UpdatedNote The improved and structurally validated variant of mini-beatrix utilizing a specialized delta attention mechanism through the entire structure. The upgraded and stronger variant of mini-beatrix-1, trained from scratch again.
AbstractPhil/mini-beatrix-1
Text Generation • 0.1B • Updated • 1.2k • 1Note A fully trained version of AlephLM specifically curated for LLM next token prediction behavior.
AbstractPhil/captionbert-8192-v2-b
Feature Extraction • 58.3M • Updated • 118 • 1Note Captionbert v2 b, the stronger of the twins. A distilled bert from 5 models using cc12m in 3 caption lengths extracted from 5 bert teachers.
AbstractPhil/captionbert-8192-v2
Feature Extraction • 58.3M • Updated • 171 • 1Note A distilled bert from 5 models using cc12m in 3 caption lengths extracted from 5 bert teachers.
-
AbstractPhil/Qwen3.5-0.8B-json-captioner
Image-Text-to-Text • 0.9B • Updated • 13
AbstractPhil/geovocab-patch-maker
Updated • 4Note The heavily evolved geovocab form - meant to target a large array of dense enriched geometric vocabulary tokens with useful behavioral information.
AbstractPhil/gated-david
Image Classification • UpdatedNote Prototype for the entire geometric systems.
AbstractPhil/sd15-flow-lune
Text-to-Image • Updated • 42Note An SD15 rectified flow matching model nearly fully perfectly converted from ddpm to rectified flow matching.
AbstractPhil/david-shared-space
Image Classification • UpdatedNote A David variant that supports shared latent space between multiple expert datas simultaneously.
AbstractPhil/vit-beatrix
Image Classification • 9.53M • Updated • 11Note An early prototype vit with low accuracy but understands basic geometry.
AbstractPhil/penta-vit-experiments
Zero-Shot Classification • Updated • 1Note Directed 5simpled pentachoron vit experiments.
AbstractPhil/bert-beatrix-2048
0.1B • Updated • 11 • 1Note A symbolic categorical encoder.
AbstractPhil/beeper-rose-v4
Text Generation • Updated • 10Note A prototype geometric llm.
AbstractPhil/pentachora-greyscale-frequency-encoded
Zero-Shot Classification • UpdatedNote The earliest prototype from a notebook, specifically designed to encode using complexity.
AbstractPhil/pentachora-multi-channel-frequency-encoded
Zero-Shot Classification • UpdatedNote The evolved form of the greyscale version.
AbstractPhil/clips
5B • Updated • 1 • 18Note A large array of clips and text encoders from many sd15, sdxl, flux, and other diffusion models. Primarily packaged for use out of the box in comfyui.
AbstractPhil/T5-Small-Human-Attentive-Try2-Pass3
60.5M • Updated • 5Note A burned early attempt at a t5-small with symbolic representation, pre-beatrix 2048.