bmax16634/solollm-1.0-152m-base
Text Generation • 0.2B • Updated • 13
From-scratch SoloLLM 1.0 base-model releases packaged for Hugging Face Transformers.
Note SoloLLM-1.0-152M-Base: best completed SoloLLM 1.0 base checkpoint; 151.9M params; trained from scratch on 9.8B tokens; beats GPT-2 small across the fixed project eval suite, with the caveat that it is larger than GPT-2 small.
Note SoloLLM-1.0-123M-Base: smaller-than-GPT-2 SoloLLM 1.0 checkpoint; 123.6M params; wins the fixed multiple-choice suite and most external checks but not every metric.