sologpt-base-v1

SoloGPT Base v1

Legacy first-version checkpoint for the SoloLLM project.

This repo is kept public as historical context for the project, but it is no longer the recommended model to inspect or demo.

Use the final v3 model instead:

• Final model: https://huggingface.co/bmax16634/sologpt-v3-150m-base
• Completion demo: https://huggingface.co/spaces/bmax16634/sologpt-v3-150m-demo

Status

SoloGPT v1 was the original from-scratch GPT-style PyTorch baseline. It proved that the training and deployment pipeline worked, but later versions rebuilt the architecture, data pipeline, evaluation suite, and project framing.

The current headline artifact is SoloLLM v3 150M Base, a GPT-2-class base model trained from scratch on a curated 10B-token dataset and released with a full model card, metrics, config, loading example, and public completion demo.

Recommended Citation

For current portfolio/recruiting review, use:

SoloLLM v3 trains GPT-2-class base LMs from scratch on one RTX 3090. The final 150M model beats GPT-2 small overall on a fixed evaluation suite, while a slightly smaller 123M model beats GPT-2 on most external benchmarks but does not fully clear the strict across-board smaller-than-GPT-2 bar.