Polycoder

  • URL: Polycoder on GitHub
  • Description: An open-source AI code generation tool focused on various programming languages, leveraging a 2.7 billion parameter model.

Orion-14B

  • URL: Orion-14B on Arxiv
  • Description: A 14 billion parameter multilingual LLM, trained on a diverse 2.5 trillion token corpus, enhancing NLP tasks performance.

Cohere Command

  • URL: Cohere Command
  • Description: Known for accuracy and robustness, used by companies like Spotify and Jasper for reliable AI experiences.

PHI-2

  • Developer: Microsoft
  • URL: PHI-2 on Hugging Face
  • Description: A 2.7 billion-parameter model that outperforms larger models across benchmarks, emphasizing “textbook-quality” data.

Stable Beluga 1 and 2

  • Developer: Stability AI’s CarperAI lab
  • URL: Stable Beluga
  • Description: Open-access LLMs for research, leveraging the LLaMA model foundations and aimed at advancing open research.

LLaVA

  • Developer: Microsoft Research
  • URL: LLaVA
  • Description: A general-purpose visual and language understanding model, mimicking GPT-4’s multimodal capabilities and setting new benchmarks in Science QA accuracy.

INTERN-2.5

  • Developer: OpenGVLab
  • URL: INTERN-2.5 on GitHub
  • Description: Achieves over 90% top-1 accuracy on ImageNet, excelling in visual tasks using deformable convolutions.

Tabnine

  • URL: Tabnine
  • Description: An AI-powered code completion tool that provides context-aware recommendations across programming languages.

This roundup of the latest LLM releases highlights the diverse and innovative approaches developers are taking to advance AI technology, offering a wide range of capabilities and specializations.