I am more and more excited about Local ML on Apple Silicon. Core ML in particular is starting to be a super nice stack to build on. Question: Do you want to have a 7B parameter model running at 30+ tokens/second, using less than 4GB of memory on your Mac? Then you need to
LLMS
-
Groq Inc emerges as potential leader in AI infrastructure
By
–
It's possible, @GroqInc might be the best bet here
-
VideoPoet Wins ICML Best Paper Award for Zero-Shot Video
By
–
Congratulations to the authors of "VideoPoet: A Large Language Model for Zero-Shot Generation" for winning one of this year's @icmlconf Best Paper Awards! #ICML2024 Paper: https://
openreview.net/forum?id=LRkJw
PIDuE
… Blog post: https://
goo.gle/4atanoj -
Distilling Billion-Parameter Models Into Efficient Online Versions
By
–
@smritichirps
, Andrew Gilchrist-Scott & Brad Stocks will be back at the #ICML2024 Google Research booth at 4pm CEST to explain how we distill billion+ parameter models into quick and resource-efficient online versions while maintaining their underlying world & language knowledge! -
Distilling Billion-Parameter Models into Efficient Online Versions
By
–
@smritichirps
, Andrew Gilchrist-Scott & Brad Stocks will be back at the #ICML2024 Google Research booth at 4pm CEST to explain how we distill billion+ parameter models into quick and resource-efficient online versions while maintaining their underlying world & language knowledge! -
Upcoming Release Unlikely to Include Multimodal Models
By
–
I don't expect this release to include multimodal models unfortunately
-
Model Performance Scaling: Multilingual Capabilities Across Parameter Sizes
By
–
Yeah exactly, the 70B model looks ideal in terms of perf/parameter. Funny to see how much the multilingual capabilities improve between 8B and 70B, and then between 70B and 405B
-
Google Leverages RAG for Improved Search LLM Controllability
By
–
Today at 1:30pm CEST, @smritichirps, Andrew Gilchrist-Scott & Brad Stocks will be at the #ICML2024 Google Research booth to discuss how we leverage retrieval augmented generation for improved controllability, quality control and developer experience for Google Search LLMs. pic.twitter.com/HqqxJBiWEJ
— Google AI (@GoogleAI) 23 juillet 2024Today at 1:30pm CEST, @smritichirps
, Andrew Gilchrist-Scott & Brad Stocks will be at the #ICML2024 Google Research booth to discuss how we leverage retrieval augmented generation for improved controllability, quality control and developer experience for Google Search LLMs. -
Technical differences between closed and open-source AI models
By
–
Les modèles fermés ont des verrous post et pré traitement pour empêcher d’emmener le modèle n’importe où. Les modèle ouvert non c’est brut.
-
Open Source AI Models Closing the Performance Gap with Closed Models
By
–
Wow, l’open source vient doucement mais sûrement dominer le marché de l’#IA ! Les performances des modèles fermés de Google, Anthropic ou OpenAI suivent une courbe logarithmique, tandis que les performances des modèles ouverts comme Meta ou Mistral AI, bien que parties de