IN TODAY'S SIGNAL
|
Top News: The mysterious gpt2-chatbot
3 Trending Repos: Xtuner, luminal, Deepfacelive
5 Trending Signals
Pytorch Tip: Indexing for in-place operations
|
Read Time: 4 min 45 sec
|
|
|
|
Enjoying this newsletter?
Please forward it to a friend or colleague. It helps us keep this content free.
|
|
|
|
TOP NEWS
|
Language Models
|
Mysterious ‘gpt2-chatbot’ Outperforms GPT-4 and Suddenly Disappears |
⇧ 45,827 Likes
|
 |
A new AI model named "gpt2-chatbot" appeared without announcement on LMSYS Chatbot Arena. It outperformed OpenAI's GPT-4 and Anthropic's Claude Opus in reasoning, coding, and math tasks, stunning researchers with its skills.
Performance
- Solved an International Math Olympiad problem on first attempt - a competition where only top 4 U.S. high school students qualify annually.
- Exceeded GPT-4 and Claude Opus benchmarks on complex coding prompts used for model evaluation at AI startups like CodeGen.
- Demonstrated
iterative dialogue capabilities, self-awareness in refining responses, and ability to plan out solution steps.
- Exhibited rule-breaking behavior and willingness to solve logic puzzles that historically stumped GPT-4.
- Matched or exceeded GPT-4 performance on ASCII art generation.
Uncertain Origins Spark Speculation
With no official documentation, the model's creator remains unidentified, fueling theories:
- Potential OpenAI Release: Exhibits similarities to OpenAI models; self-identifies as created by OpenAI; OpenAI CEO Sam Altman cryptically tweeted "I do have a soft spot for gpt2".
- Independent Group: Some believe an unaffiliated team released it to showcase AI capabilities and generate buzz, similar to GPT-4chan in 2022.
This suggests gpt2-chatbot could be an interim upgrade (GPT-4.5) rather than the full GPT-5.
|
Community Feedback |
AI Breakfast
Most likely explanation for gpt2-chatbot: OpenAI has been working on a more efficient method for fine-tuning language models, and they managed to get GPT-2, a 1.5B parameter model, to perform pretty damn close to GPT-4, which is an order of magnitude larger and more costly to train/run. They’re driving down the cost of operating LLMs by injecting the little models with some fine-tuned steroids. “GPT-5” might have fewer parameters than GPT-4.
|
Brian Roemmele
I have been testing gpt2-chatbot for a few days. Today it seems to have gotten much more attention. It surpassed all of our ChatGPT-4 benchmarks. Hypothesis: A few of us have concluded it is a form of pre-lobotomized ChatGPT-4 or trained heavily on it.
|
Access |
LMSYS - Now unavailable (could go back online)
|
|
READ MORE
|
|
|
|
|
Don’t Miss The AI and ML Developer Conference |
Join mlcon, a hybrid conference that has built a community of over 13,000+ AI & ML developers since 2021.
With the rapid emergence of generative AI technology, LLMs, and AI's continued maturity, mlcon is the central hub for professionals to learn from industry leaders.
Break silos, share insights, and collaborate for collective progress in AI innovation.
|
|
REGISTER NOW
|
partner with us →
|
|
|
|
TOP REPOS
|
Fine-tuning
|
|
☆ 2320
|
XTuner is an efficient, flexible and full-featured toolkit for fine-tuning large models. It supports different kinds of fine-tuning (continuous pre-training, instruction fine-tuning, agent fine-tuning) and various training algorithms (QLoRA, LoRA, full parameter).
It also supports pre-training and fine-tuning on almost all GPUs. It offers a high training throughput by automatically dispatching high-performance operators, such as FlashAttention and Triton kernels.
|
|
|
Development
|
|
☆ 1180
|
Luminal, a deep learning library, helps you achieve high performance by compiling everything ahead-of-time. It enables you to build and execute static computation graphs, optimizing performance. With minimalistic architecture and native support for CUDA/Metal APIs, it ensures efficient execution. Tested against Pytorch, it validates correctness. Luminal aims to be the fastest ML framework, achieving speeds of 15-25 tokens per second on M-series Macbooks.
|
|
|
Video Generation
|
|
☆ 23197
|
LLaMA DeepFaceLive is a library that allows to perform face swaps in real time a from computer streaming using a webcam or directly from a video call.
The repository shows a list of faces the swap can be performed with. It also offers the possibility to train your own face model for better results.
There is also a Face Animator module in the DeepFaceLive repository that allows to control a static face picture using video or your own face from the camera.
|
|
|
|
|
|
|
|
|
Prem Platform - The Smartest Generative AI Enabler
|
Effortlessly integrate generative AI into your applications with full ownership and confidence. Focus on your product; we'll take care of the AI complexities.
Simple. Easy to use and integrate into your product.
No Vendor LockIn: No single provider dependency.
Ready for Production. Ready to scale.
|
Start building now ↗️
|
|
|
|
TRENDING SIGNALS
|
Language Models
|
|
⇧ 1021 Likes
|
|
|
Development
|
|
⇧ 6239 Likes
|
|
|
Gen AI
|
|
⇧ 240 Likes
|
|
|
LLMs
|
|
⇧ 451 Stars
|
|
|
Theory
|
|
⇧ 175 Likes
|
|
|
|
|
|
|
|
|
PYTORCH TIP
|
Optimization
|
Indexing for in-place operations |
This technique leverages PyTorch’s capability to perform operations in-place, which can lead to significant memory savings because it avoids creating unnecessary copies of data. It’s particularly beneficial when dealing with large datasets or models where memory is a constraint.
To do so, you can exploit PyTorch's advanced indexing to perform in-place operations directly on slices of tensors, reducing memory footprint and enhancing computation speed, especially during complex data manipulations.
This approach directly manipulates the tensor's data without the need for intermediate tensors or loops, making the code cleaner, faster, and more memory-efficient.
Here's how you can use PyTorch indexing to find and update only the negative values of a tensor:
|
import torch
x = torch.randn(100, 100, 100) # Large tensor with random values
# Use advanced indexing to update the negative values
x[x < 0] += 0.1
print("Updated tensor with updated values in-place.")
|
|
|
|
|
