(2/2) Researchers at Cerebras, Inception, and MBZUAI developed techniques to:
-Create a custom tokenization scheme optimized for Arabic -Identify the proper mix of English, Arabic, and code data -Augment Arabic tokens with machine-translations Models: https://
huggingface.co/inception-mbzu
ai
…
OPEN SOURCE
-
Cerebras MBZUAI Develop Arabic-Optimized Language Model Techniques
By
–
-
Jais: Top-Performing Open-Source Arabic LLM Model
By
–
(1/2) Jais is the top-performing open-source Arabic LLM in the world. Better than Bloom, Llama, Falcon, etc. It is a 13B parameter model that was trained on a brand-new multilingual dataset. Read the technical paper: https://
inceptioniai.org/jais/docs/Tech
nicalpaper.pdf
… -
Code Llama Now Available on Poe Platform
By
–
Code Llama is now available on Poe! Note that many of the Code Llama demos floating around during the last week on other services had bugs; we took some time to make sure our implementation works accurately to the underlying model.
-
Code Llama Integration Added to Lit-GPT Repository
By
–
And thanks to another kind contribution, Code Llama is making its way into Lit-GPT, too!! https://
github.com/Lightning-AI/l
it-gpt/pull/472
… -
New 13B Bilingual Arabic-English LLM Released Open Source
By
–
(2/3) Jais highlights:
-State-of-the-art 13B parameter bilingual Arabic-English LLM
-Trained on a new data set including 116 billion Arabic tokens and 279 billion English/code tokens -Bidirectional transfer learning
-Open source Model on Hugging Face: https://
huggingface.co/inception-mbzu
ai
… -
Jais: Advanced Arabic LLM Trained on Condor Galaxy 1
By
–
(3/3) Jais was trained on the newly unveiled Condor Galaxy 1 (CG-1) AI supercomputer, built by the G42 – Cerebras strategic partnership. To learn more, check out the Press Release: https://
cerebras.net/press-release/
meet-jais-the-worlds-most-advanced-arabic-large-language-model-open-sourced-by-g42s-inception
… -
Wikipedia Data Usage in NLP Research Papers Analysis
By
–
Asking again since I'm still curious. Has anyone looked into this? (E.g. what % of papers in the @aclanthology use Wikipedia as a data source…)
-
CoTracker: Advanced Pixel Tracking for Video Analysis
By
–
CoTracker can track every pixel in a video, points sampled on a regular grid on any video frame or manually selected points. In our testing it compares favorably against state-of-the-art point tracking methods in both efficiency & accuracy.
— AI at Meta (@AIatMeta) 29 août 2023
Code ➡️ https://t.co/d0igWWYFOu pic.twitter.com/hSO0wbqNu9CoTracker can track every pixel in a video, points sampled on a regular grid on any video frame or manually selected points. In our testing it compares favorably against state-of-the-art point tracking methods in both efficiency & accuracy. Code https://
bit.ly/3KZoLcQ -
CoTracker: Multi-point Video Tracking with Transformer Networks
By
–
New on @huggingface — CoTracker simultaneously tracks the movement of multiple points in videos using a flexible design based on a transformer network — it models correlation of the points in time via specialized attention layers.
— AI at Meta (@AIatMeta) 29 août 2023
🤗 Try CoTracker ➡️ https://t.co/IdaUrH9Xxr pic.twitter.com/WogyQI3F4vNew on @huggingface — CoTracker simultaneously tracks the movement of multiple points in videos using a flexible design based on a transformer network — it models correlation of the points in time via specialized attention layers. Try CoTracker https://
bit.ly/3swQFqt -
Open Source AI: Inevitable Capabilities and Alignment Risks
By
–
Open source is all you need to make inevitable any feat that one group can do and no one else can stop afterwards. This has both good and bad effects on society. The extinction issue, and more generally the present impossibility of aligning good ASIs to counter bad ones, makes