• AI Research Insights
  • Posts
  • Marktechpost AI Newsletter: Google Releases Gemma 2 Series Models + NYU Researchers Introduce Cambrian-1 and many more...

Marktechpost AI Newsletter: Google Releases Gemma 2 Series Models + NYU Researchers Introduce Cambrian-1 and many more...

Marktechpost AI Newsletter: BigCodeBMarktechpost AI Newsletter: Google Releases Gemma 2 Series Models + NYU Researchers Introduce Cambrian-1 and many more...ench by BigCode + Together AI Introduces Mixture of Agents (MoA) + many more....

Presented by

Want to get in front of 1.5 Million AI enthusiasts? Work with us here

Featured Research..

Google Releases Gemma 2 Series Models: Advanced LLM Models in 9B and 27B Sizes Trained on 13T Tokens

Google has unveiled two new models in its Gemma 2 series: the 27B and 9B. These models showcase significant advancements in AI language processing, offering high performance with a lightweight structure.

The Gemma 2 27B model is the larger of the two, with 27 billion parameters. This model is designed to handle more complex tasks, providing greater accuracy and depth in language understanding and generation. Its larger size allows it to capture more nuances in language, making it ideal for applications that require a deep understanding of context and subtleties.

On the other hand, the Gemma 2 9B model, with 9 billion parameters, offers a more lightweight option that still delivers high performance. This model is particularly suited for applications where computational efficiency and speed are critical. Despite its smaller size, the 9B model maintains a high level of accuracy and is capable of handling a wide range of tasks effectively.

  • Trained on 13T tokens (27B) and 8T tokens (9B)

  • 9B scores 71.3 MMLU; 52.8 AGIEval; 40.2 HumanEval

  • 27B scores 75.2 MMLU; 55.1 AGIEval; 51.8 HumanEval

  • Used Soft Attention, Distillation, RLHF & Model Merging

 Editor’s Picks…

NYU Researchers Introduce Cambrian-1: Advancing Multimodal AI with Vision-Centric Large Language Models for Enhanced Real-World Performance and Integration

Traditionally, visual representations in AI are evaluated using benchmarks such as ImageNet for image classification or COCO for object detection. These methods focus on specific tasks, and the integrated capabilities of MLLMs in combining visual and textual data need to be fully assessed. NYU Researchers introduced Cambrian-1, a vision-centric MLLM designed to enhance the integration of visual features with language models to address the above concerns. This model includes contributions from New York University and incorporates various vision encoders and a unique connector called the Spatial Vision Aggregator (SVA).

The Cambrian-1 model employs the SVA to dynamically connect high-resolution visual features with language models, reducing token count and enhancing visual grounding. Additionally, the model uses a newly curated visual instruction-tuning dataset, CV-Bench, which transforms traditional vision benchmarks into a visual question-answering format. This approach allows for a comprehensive evaluation & training of visual representations within the MLLM framework.


Meet Gretel Navigator: the first compound AI system built to create, edit, and augment tabular data. 🚀🚀🚀

Get inspired by popular Navigator use cases:

  • Empower frontier AI teams with high-quality datasets to train LLMs.

  • Safeguard sensitive, proprietary datasets when evaluating public ML models

  • Teach LLMs new tasks or domains for new generative AI-powered applications.

  • Augment real-world data to build more performant intelligent applications.

  • Generate synthetic question-truth pairs to evaluate RAG models.

    [Sign Up for Free] Try Gretel Navigator, the first compound AI system built to create, edit, and augment tabular data.

Hugging Face Releases Open LLM Leaderboard 2: A Major Upgrade Featuring Tougher Benchmarks, Fairer Scoring, and Enhanced Community Collaboration for Evaluating Language Models

Hugging Face has announced the release of the Open LLM Leaderboard v2, a significant upgrade designed to address the challenges and limitations of its predecessor. The new leaderboard introduces more rigorous benchmarks, refined evaluation methods, and a fairer scoring system, promising to reinvigorate the competitive landscape for language models.

Over the past year, the original Open LLM Leaderboard became a pivotal resource in the machine learning community, attracting over 2 million unique visitors and engaging 300,000 active monthly users. Despite its success, the escalating performance of models led to benchmark saturation. Models began to reach baseline human performance on benchmarks like HellaSwag, MMLU, and ARC, reducing their effectiveness in distinguishing model capabilities. Additionally, some models exhibited signs of contamination, having been trained on data similar to the benchmarks, which compromised the integrity of their scores.

EvolutionaryScale Introduces ESM3: A Frontier Multimodal Generative Language Model that Reasons Over the Sequence, Structure, and Function of Proteins

Researchers from Evolutionary Scale PBC, Arc Institute, and the University of California have developed ESM3, an advanced generative language model for proteins. ESM3 can simulate evolutionary processes to create functional proteins vastly different from known ones. It integrates sequence, structure, and function to generate proteins following complex prompts. Notably, ESM3 generated a new fluorescent protein, esmGFP, which is 58% different from any known fluorescent proteins—a degree of difference comparable to 500 million years of natural evolution. This breakthrough demonstrates ESM3’s potential in protein engineering, offering creative solutions to biological challenges.