Upstage

View Original

Upstage's 70B Language Model Outperforms GPT-3.5, Becomes Global No.1


2023/8/1
  • Upstage's 70-billion-parameter model ranked no. 1 on HuggingFace Open LLM Leaderboard

  • The model's performance surpassed GPT-3.5 to become the world's best open-source language model worldwide

  • With the world's best LLM, the company is poised to solidify its leadership in private AI market


(Seoul, Aug. 1, 2023 /Upstage) Upstage, the leading AI startup in Korea, has recently unveiled its latest generative AI model, surpassing OpenAI's GPT-3.5 to claim the title of the world's best AI model.

On August 1st, Upstage announced the release of its latest version of the large language model (LLM), which achieved an average score of 72.3 on the HuggingFace Open LLM Leaderboard. This score surpasses even that of GPT-3.5, solidifying Upstage's position as a frontrunner in the competitive AI landscape.

The HuggingFace Open LLM Leaderboard serves as a benchmark for evaluating open-source language models, with over 500 LLMs submitted for evaluation. The evaluation is based on four key metrics, namely the model's reasoning challenge, common sense inference, contextual understanding, and factual accuracy (preventing AI hallucinations).

Earlier in July, Upstage's 30-billion-parameter language model already claimed the top spot on the leaderboard with an average score of 64.7, even surpassing Meta's LLaMA 2 70B model, which was released on the same day.

In pursuit of enhancing its model's performance, Upstage fine-tuned it with the latest LLaMA 2 and trained with a bigger 70 billion parameters. The result was a significant boost to the model's average performance, reaching an outstanding 72.3, surpassing the previous leader, Stability AI's ‘Stable Beluga 2’ (71.4), which had maintained its position since the introduction of LLaMA 2.

Further, Upstage's 70B LLM has now etched its name in history, surpassing the benchmark score of GPT-3.5 (71.9). This remarkable achievement represents a groundbreaking milestone, as it marks the first time an open-source model has outperformed a global tech giant, a testament to the local AI startup’s unwavering commitment to technological innovation and excellence.

Behind this groundbreaking accomplishment is Upstage's world-class team of AI experts, whose expertise and dedication culminated in the development of the top-ranked model in just two months. Leveraging their vast experience in crafting a highly successful Korean language NLU dataset named "KLUE" and harnessing prompt engineering and fine-tuning techniques from their flagship application "AskUp," Upstage achieved unparalleled results in an impressively short time frame.

With its innovative AI solutions, Upstage aims to establish itself as a prominent player in the emerging "private AI" market worldwide. Private AI solutions are specifically trained on a company's private data, ensuring the security of sensitive information and minimizing the possibility of generating inaccurate or misleading outputs, such as AI hallucinations.

Upstage's world-class LLM presents an excellent opportunity for companies to integrate cutting-edge generative AI into their business operations. The company also stands out by offering a smaller model compared to global tech giants, providing access to state-of-the-art technology at a more affordable cost. This advantage positions Upstage as a preferred choice for businesses seeking advanced AI solutions.

Sung Kim, Upstage’s CEO, said "I am delighted to unveil Upstage's groundbreaking generative AI model, surpassing even the capabilities of ChatGPT," adding, “Upstage is committed to to expanding its reach in the global private AI market, leveraging unparalleled technological expertise.”

* Image description: Upstage announced the release of its latest version of the large language model (LLM), which achieved an average score of 72.3 on the HuggingFace Open LLM Leaderboard.

← Back to Newsroom