HyperGAI

HyperGAI is an applied AI research company, innovating in Multimodal Generative AI technology and multimodal foundation models.

[HyperGAI @hypergai]

🔥HPT 1.5 Edge🔥 - Our Latest Open-Source 4B Multimodal LLM for Edge and Mobile Devices

[Steven Hoi @stevenhoi]

Introducing 🔥HPT 1.5 Edge🔥, HyperGAI’s latest open-source lightweight multimodal LLM for edge and mobile devices, outperforming Microsoft's recent Phi-3-Vision (4B) model and reaching GPT-4V performance on multiple benchmarks.

HPT 1.5 Edge is our latest effort to bring the cutting-edge multimodal LLM more accessible to a wide range of users, especially for on-device GenAI use cases. While the Edge model follows its HPT predecessors’ design, it was trained with an improved data mixture and better training strategies, and powered by a strong lightweight text language model, the Phi-3-mini of only 3.8B parameters. As a result, despite having only 4.2B parameters in total, the Edge model demonstrates impressive performances while being extremely efficient for on-device multimodal GenAI use cases.

HPT 1.5 Edge archives state-of-the-art performances among the open-sourced multimodal LLMs of similar sizes (<5B), even reaching proprietary models GPT-4V and Gemini 1.0 Pro results on multiple benchmarks, like POPE, Seed-I, MMStar, and MathVista. HPT 1.5 Edge is open-sourced and publicly available on Hugging Face and Github, under the Apache 2.0 license.

Main Blog: https://t.co/fWqyrqyJYr

[HyperGAI @hypergai]

Meet the brand new HPT 1.5 Air 🚀

[Steven Hoi @stevenhoi]

Introducing 🔥HPT 1.5 Air🔥 the best open 8B Multimodal LLM with LLama 3!

HPT 1.5 Air, HyperGAI’s best 8B Multimodal Llama 3, follows the similar recipe as its predecessor, HPT 1.0 Air, with a visual encoder, the novel H-Former, and an LLM. Compared with HPT 1.0 Air, we upgraded the visual encoder and changed the LLM to the latest LLaMA 3 8B Instruct version, and trained on an improved larger dataset mixed of image and text data.

HPT 1.5 Air achieves state-of-the-art performance among the open-source multimodal LLM models of similar model sizes (<10B), and demonstrates improved capabilities in multimodal understanding & complex reasoning, and real-world contextual understanding, etc.

HPT 1.5 Air also outperforms GPT-4V and Gemini 1.0 Pro on the SEED-I, ScienceQA, and MMStar benchmarks, despite having significantly fewer parameters. HPT 1.5 Air is now publicly available on Hugging Face and GitHub, with a public demo coming soon.

Main Blog: https://t.co/B39drPi7oH

[HyperGAI @hypergai]

We're only just getting started and if you're
as excited about the future of hyper general
artificial intelligence as we are, apply
at https://t.co/4d2G4F1Lpl

[Stephanie Horbaczewski @iamsteph]

Congratulations to our new founder @stevenhoi! Can’t wait to watch @hypergai journey! Can we invest?

[Marktechpost AI @Marktechpost]

HyperGAI Introduces HPT: A Groundbreaking Family of Leading Multimodal LLMs

Model: https://t.co/klc64XIgR4

Github: https://t.co/rLf71WDLb4

Read: https://t.co/flxgG28MiC

@hypergai #ArtificialIntelligence #LLMs

[EmbeddedLLM @EmbeddedLLM]

MM1 paper from @Apple got you hyped? Wait till you try HPT! This multimodal SOTA is ready for you to play with:
- Outperforms GPT-4V and Gemini Pro
- HPT Air is Apache 2.0 license and outperform all open models
- HPT can either train from scratch or leverage existing pretrained vision and language models.

[Tiezhen WANG @Xianbao_QIAN]

@hypergai released a multi-modal model HPT Air on @huggingface based on Yi-6B-Chat and pretrained visual encoder clip-vit-large-patch14-336. The model seems to be quite strong on benchmarks as below

Technical report: https://t.co/cfY458Sm98
Weights on HF: https://t.co/5kSNhk1du4

[HyperGAI @hypergai]

We've been testing the capabilities of our new Multimodal AI model, HPT. We've been showing it diverse images and asking it to interpret. Here are some of our favourite conversations with HPT. #LLM #MultimodalLLM #HPT #GenAI

[HyperGAI @hypergai]

“Unlike other AI companies focused on applications, we are distinctive in our approach of driving the development of the Multimodal foundation models, and we’ll leverage this to continue delivering breakthrough AI products and solutions." @stevenhoi

[HyperGAI @hypergai]

"HyperGAI believes that true general intelligence requires more than just text-based understanding" says @stevenhoi, Ph.D., Founder and CEO of HyperGAI.
Explore- https://t.co/7gkwHv9lfp

[NewsRank.io @NewsRankIO]

⚡️ @HyperGAI has introduced the #HPT multimodal large language model. The simplified version HPT Air is published immediately with the #source code.

According to the developers, this #LLM surpasses #GPT4 in efficiency.

[Ryan Morrison @RyanMorrisonJer]

HyperGAI is another multimodal AI lab coming out of stealth mode promising best in class models.

Claims HPT Pro outperforms GPT-4V and Gemini Pro on MMBench and SEED-image benchmarks. Come in two versions, Air and Pro.

https://t.co/wX1z5KGTY4

[Steven Hoi @stevenhoi]

I'm excited to announce our new AI company, @HyperGAI, is officially coming out of stealth mode! HyperGAI is an applied AI research company innovating in the Multimodal GenAI to empower everyone to be more creative and productive! Learn more about us at: https://t.co/EMIt7emsrq

[HyperGAI @hypergai]

Introducing🔥HPT🔥- a new family of Open Multimodal LLMs from HyperGAI. HPT (Hyper-Pretrained Transformer) demonstrates strong capabilities on multiple multimodal benchmarks. Read more- https://t.co/REeNvFLGQA

[HyperGAI @hypergai]

Interested in what HyperGAI has to offer? Read more to find out what we're all about- https://t.co/1Wq9epxTgW

[HyperGAI @hypergai]

HyperGAI is an applied AI research company, innovating in Multimodal Generative AI technology and multimodal foundation models. Explore - https://t.co/TMBPnIWRLr

[HyperGAI @hypergai]

Announcing https://t.co/TMBPnIWRLr. HyperGAI emerges from stealth mode to pioneer the next era of Multimodal GenAI technology.