Coder-AN
Abstract
We discovered the underlying physics in Next-token Prediction (NTP). We identified the law of information conservation within NTP and proposed the First Law of Information Capacity (IC-1), demonstrating that the essence of intelligence emergence in auto-regressive models is fundamentally a process of information transfer. We also introduced Landauer's Principle into NTP, formulating the Second Law of Information Capacity (IC-2), which establishes the relationship between auto-regressive model training and energy consumption. Additionally, we presented several corollaries, which hold practical significance for production practices. Finally, we validated the compatibility and complementarity of our findings with existing theories.
Community
This is an automated message from the Librarian Bot. I found the following papers similar to this paper.
The following papers were recommended by the Semantic Scholar API
- ENTP: Encoder-only Next Token Prediction (2024)
- Scaling Laws Across Model Architectures: A Comparative Analysis of Dense and MoE Models in Large Language Models (2024)
- Scaling Optimal LR Across Token Horizons (2024)
- Scaling Parameter-Constrained Language Models with Quality Data (2024)
- A Simple Model of Inference Scaling Laws (2024)
Please give a thumbs up to this comment if you found it helpful!
If you want recommendations for any Paper on Hugging Face checkout this Space
You can directly ask Librarian Bot for paper recommendations by tagging it in a comment:
@librarian-bot
recommend
Hi, thanks for your work. I have 2 questions:
1-) You say that the higher number of tokens at training phase increases the information. The formula indicates that -> for 2 epochs of training, the number of tokens equals 2 times of tokens of dataset. In this situation, In my opinion, seeing same thing for several times cannot increase the "information" linearly. What is your insights about this statement?
2-) Higher entropy indicates better data in the manner of information. In this case, if the dataset augmented as "22 equals to 4" -> "22 equals to 6", the entropy of dataset increases. What do you think about misinformation ?
Models citing this paper 0
No model linking this paper
Datasets citing this paper 0
No dataset linking this paper
Spaces citing this paper 0
No Space linking this paper