All Updates

All Updates

icon
Filter
Product updates
Tencent introduced VoCo-LLaMA for compression lengthy of vision tokens
Foundation Models
Jun 24, 2024
This week:
Partnerships
Qualcomm and Google partner to develop AI-driven automotive solutions
Auto Tech
Yesterday
Product updates
Meta AI releases LayerSkip to accelerate inference in LLMs
Generative AI Infrastructure
Yesterday
Funding
Freeform secures funding from NVIDIA's NVentures
Additive Manufacturing
Oct 22, 2024
Product updates
Flexxbotics announces compatibility with LMI Technologies for quality inspection
Smart Factory
Oct 22, 2024
Funding
Oxla raises USD 11 million in seed funding to drive commercialization
Data Infrastructure & Analytics
Oct 22, 2024
Product updates
Cohesity enhances Gaia, its AI assistant, with visual data exploration and expanded data sources
Data Infrastructure & Analytics
Oct 22, 2024
Product updates
Finzly launches FedNow service through BankOS platform in AWS marketplace
FinTech Infrastructure
Oct 22, 2024
Product updates
Runway launches Act-One for AI facial expression motion capture
Generative AI Applications
Oct 22, 2024
Product updates
Ideogram launches Canvas for image manipulation and generation
Generative AI Applications
Oct 22, 2024
Partnerships
UiPath partners with Inflection AI to integrate AI solutions for enterprises
Generative AI Applications
Oct 22, 2024
Foundation Models

Foundation Models

Jun 24, 2024

Tencent introduced VoCo-LLaMA for compression lengthy of vision tokens

Product updates

  • Tencent, has introduced VoCo-LLaMA, an LLM for compressing lengthy vision tokens into a single token with minimal loss of visual data.

  • VoCo-LLaMA comprises "Vision Compression" tokens that are charged with compressing and distilling vision tokens in LLMs. The solution reportedly can achieve a compression ratio of 576x while maintaining 83.7% performance on common visual understanding benchmarks. The solution is also claimed to contribute to efficiency gains, enabling a 99.8% reduction in cache storage, a 94.8% decrease in FLOPs, and a 69.6% faster inference time. 

  • However, the solution is claimed to diminish the model's ability to understand uncompressed tokens and face difficulties with diverse fine-grained compression levels.

Contact us

Gain access to all industry hubs, market maps, research tools, and more
Get a demo
arrow
menuarrow

By using this site, you agree to allow SPEEDA Edge and our partners to use cookies for analytics and personalization. Visit our privacy policy for more information about our data collection practices.