StellarX 4B V0.2 is an open-source language model by Dampish. Features: 4b LLM, VRAM: 16GB, Context: 2K, License: cc-by-nc-sa-4.0, LLM Explorer Score: 0.11, Arc: 34.6, HellaSwag: 56.7, MMLU: 25.6.
primarily for research purposes, not designed for human-facing interactions
Considerations:
Conduct risk and bias assessments before using for downstream tasks.
Additional Notes
StellarX is not limited to the English language and can sometimes be used for translation and text generation in other languages. It draws inspiration from GPT-NeoX architecture.
Supported Languages
English (primary)
Training Details
Data Sources:
redpajama dataset by togethercumputer
Data Volume:
300B tokens training (goal 810B tokens)
Model Architecture:
GPT-NeoX
Release Notes
Version:
0.2
Notes:
Trained on 300B tokens, goal is 810B tokens, higher future accuracy expected.