LLM Reference

Pythia 31M

About

Pythia 31M is an open-source large language model developed by EleutherAI, designed as part of a suite to explore interpretability, learning dynamics, and ethics in LLMs. This model employs the GPT-NeoXForCausalLM architecture with 30.5 million parameters and a context length of 2048 tokens. Trained from random weights associated with the Pythia series, it has several variants; these versions are based on different datasets such as SimpleWiki, SimplePile Lite, and GoodWiki. The model primarily focuses on text generation, with varying performances across different benchmarks compared to more extensive models like GPT-4. However, limitations include insufficient activity for serverless API deployment, and there is a noted lack of extensive documentation on its uses, training methods, and evaluation, hindering a thorough understanding of its full potential.

Capabilities

MultimodalFunction CallingTool UseJSON Mode

Specifications

FamilyPythia
Released2023-05-31
Parameters31M
ArchitectureDecoder Only
Specializationgeneral