SoloLLM

bmax16634 's Collections

updated 15 days ago

SoloLLM project artifacts: final v3 GPT-2-class base model, smaller 123M ablation, public completion demo, and legacy v1 baseline.

bmax16634/sologpt-v3-150m-base

Text Generation • 0.2B • Updated 15 days ago • 117

Note Final best SoloLLM v3 model: 151.9M params, trained from scratch on the curated 10B-token dataset, beats GPT-2 small overall on the fixed suite.
Sleeping

Agents

SoloLLM v3 150M Base Completion

🚀

Generate text completions from your prompts

Note Current public text-completion demo for the final v3 150M base model.
bmax16634/sologpt-v3-123m-base

Text Generation • 0.2B • Updated 15 days ago • 60

Note Smaller-than-GPT-2 ablation: 123.6M params, wins most external checks but does not beat GPT-2 across every metric.
bmax16634/sologpt-base-v1

Updated 15 days ago

Note Legacy first-version baseline kept public as historical context; use v3 for the current result.
Sleeping

SoloGPT Base V1 Legacy Demo

🚀

Run interactive Python apps directly in your browser

Note Legacy v1 demo kept public for historical context; use the v3 demo for the current result.