- NEC unveils two new large language models (LLMs) boasting over tenfold inference speed compared to GPT-4.
- “NEC cotomi Pro” matches performance with GPT-4 and Anthropic’s Claude 2, responding 87% faster with Nvidia L40 GPUs.
- “NEC cotomi Light” rivals global models like “GPT-3.5-Turbo” while exhibiting remarkable speed.
- Utilizing retrieval-augmented generation (RAG), NEC achieves superior response rates, even surpassing GPT-4 post fine-tuning.
- Innovative training methodologies and architectural advancements drive performance and speed enhancements.
- Integration of a comprehensive Japanese dictionary further enhances model efficacy.
- NEC’s cotomi Pro outperforms competitors like “Gemini 1.0 Pro” and shows performance comparable to Claude 2 and GPT-4.
- cotomi Light operates at a remarkable pace, outstripping GPT-4 by over 15 times.
- Evaluating LLM performance remains contentious due to the absence of standardized testing methodologies and challenges in cloud-based infrastructure comparison.
Main AI News:
NEC Corporation has introduced two groundbreaking large language models (LLMs) heralding a paradigm shift in AI capability. Boasting unrivaled performance akin to global LLMs, NEC’s latest creations operate at over tenfold the inference speed. In particular, the “NEC cotomi Pro” model stands shoulder to shoulder with OpenAI’s GPT-4 and Anthropic’s Claude 2, delivering responses a staggering 87% faster than its GPT-4 counterpart, powered by two Nvidia L40 GPUs. Meanwhile, the swift “NEC cotomi Light” matches the prowess of global models like “GPT-3.5-Turbo.”
Utilizing an in-house document retrieval system bolstered by retrieval-augmented generation (RAG), NEC achieved superior response rates compared to GPT-3.5, even before fine-tuning. Post fine-tuning, the system surpassed GPT-4, all while clocking in at a response time 93% faster. NEC attributes this breakthrough to innovative training methodologies and architectural advancements, which not only enhance performance but also turbocharge speed. The integration of a comprehensive Japanese dictionary for tokenization further amplifies the model’s efficacy.
NEC’s flagship “cotomi Pro” eclipses GPT-4’s output by over fivefold on a standard dual GPU server, surpassing models like “Gemini 1.0 Pro” in speed and rivaling the likes of Claude 2 and GPT-4 in performance. Meanwhile, “cotomi Light” operates at a mind-blowing pace, outstripping GPT-4 by more than 15 times while showcasing performance akin to “LLaMA2-70B” and GPT-3.5-Turbo.
However, evaluating LLM performance remains a contentious subject, given the lack of standardized testing methodologies. While excelling in certain domains, LLMs often falter in others, complicating direct comparisons. Additionally, the cloud-based infrastructure of the benchmark models presents challenges in achieving a fair assessment.
Moving forward, NEC pledges to strengthen collaborations and deliver cutting-edge AI solutions through its expanded cotomi lineup, empowering customers to overcome multifaceted challenges with confidence and reliability.
Conclusion:
NEC’s pioneering LLMs mark a significant advancement in AI technology, redefining the landscape by offering unparalleled speed and performance. This innovation underscores the company’s commitment to delivering cutting-edge solutions and positions NEC as a formidable player in the competitive AI market. Businesses seeking enhanced efficiency and performance in natural language processing can look to NEC’s cotomi lineup as a reliable and high-performing solution.
