AI BriefingUpstage Ranks No. 1 on the HuggingFace LLM Leaderboard
Key point
Upstage's latest LLM surpassed GPT-3.5 to rank No. 1 worldwide on the HuggingFace leaderboard.
Details
Korean AI startup Upstage unveiled its latest LLM, achieving an average score of 72.3 on the HuggingFace Open LLM Leaderboard and ranking No. 1 in the world. This result surpassed GPT-3.5's score (71.9), marking a landmark achievement in which an open-source model outperformed a model from a global tech giant.
The model was trained and fine-tuned based on LLaMA 2 at a 70B (70 billion) parameter scale. The evaluation criteria are as follows.
- Reasoning challenge
- Common sense inference
- Contextual understanding
- Factual accuracy
Upstage achieved this result in just 2 months by leveraging its experience building the Korean NLU dataset KLUE and the prompt engineering technology from its own service AskUp. Going forward, the company plans to focus on targeting the Private AI market, which guarantees security for enterprises.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.