Deep Learning Weekly: Issue 365
Gemma 2 2B, Open Source Automated Interpretability for Sparse Autoencoder Features, a paper on Alchemist: Parametric Control of Material Properties with Diffusion Models, and many more!
This week in deep learning, we bring you Google releases Gemma 2 2B, ShieldGemma and Gemma Scope, Open Source Automated Interpretability for Sparse Autoencoder Features, and a paper on Alchemist: Parametric Control of Material Properties with Diffusion Models.
You may also enjoy Groq secures $640M to supercharge AI inference with next-gen LPUs, CPU-Optimized Embedding Models with fastRAG and Haystack, a paper on "Do Anything Now": Characterizing and Evaluating In-The-Wild Jailbreak Prompts on Large Language Models, and more!
As always, happy reading and hacking. If you have something you think should be in next week's issue, find us on Twitter: @dl_weekly.
Until next week!
Industry
Google releases Gemma 2 2B, ShieldGemma and Gemma Scope
One month after the release of Gemma 2, Google has expanded their set of Gemma models to include Gemma 2 2B, ShieldGemma, and Gemma Scope.
Groq secures $640M to supercharge AI inference with next-gen LPUs
Groq has raised $640 million to rapidly scale its capacity and accelerate the development of its next-generation Language Processing Unit (LPU).
OpenAI co-founder Schulman leaves for Anthropic, Brockman takes extended leave
John Schulman, one of the co-founders of OpenAI, has left the company for rival AI startup Anthropic.
Placer Labs raises $75M to enhance market research initiatives with foot traffic data and AI
Location data startup Placer Labs has closed a $75 million funding round that brings its valuation to almost $1.5 billion.
Call for Applications: Llama 3.1 Impact Grants
Meta is accepting proposals from eligible organizations across the world with an idea for how to use Llama 3.1 to address social challenges in their communities.
NVIDIA Researchers Harness Real-Time Gen AI to Build Immersive Desert World
NVIDIA researchers used NVIDIA Edify, a multimodal architecture for visual generative AI, to build a detailed 3D desert landscape within a few minutes in a live demo at SIGGRAPH’s Real-Time Live event.
SearchGPT is a prototype of new AI search features
OpenAI is testing SearchGPT, a temporary prototype of new AI search features that give you fast and timely answers with clear and relevant sources.
MLOps & LLMOps
Securing Generative AI Deployments with NVIDIA NIM and NVIDIA NeMo Guardrails
A post that showcases how to deploy two NIM microservices, an NVIDIA NeMo Retriever embedding NIM and an LLM NIM.
CPU-Optimized Embedding Models with fastRAG and Haystack
An article on how to use optimized embedding models on CPUs to reduce latency, and improve throughput of retrieval and indexing.
Deploy open LLMs with Terraform and Amazon SageMaker
A blog on using Terraform LLM SageMaker Module to simplify the process of deploying open LLMs from Hugging Face to Amazon SageMaker.
Learning
Mapping the misuse of generative AI - Google DeepMind
New research from DeepMind analyzes the misuse of multimodal generative AI in order to help build safer and more responsible systems.
Quantization-Aware Training for Large Language Models with PyTorch
The PyTorch team presents an end-to-end Quantization-Aware Training (QAT) flow for large language models in PyTorch.
Memory-efficient Diffusion Transformers with Quanto and Diffusers
A technical blog post on how to quantize Transformer models from Diffusers and optimize their memory consumption.
Open Source Automated Interpretability for Sparse Autoencoder Features
Researchers affiliated with EleutherAI investigate different techniques for generating and scoring arbitrary text explanations of SAE features.
Libraries & Code
SGLang is a fast serving framework for large language models and vision language models.
A modular, primitive-first, python-first PyTorch library for Reinforcement Learning.
Papers & Publications
Alchemist: Parametric Control of Material Properties with Diffusion Models
Abstract:
We propose a method to control material attributes of objects like roughness, metallic, albedo, and transparency in real images. Our method capitalizes on the generative prior of text-to-image models known for photorealism, employing a scalar value and instructions to alter low-level material properties. Addressing the lack of datasets with controlled material attributes, we generated an object-centric synthetic dataset with physically-based materials. Fine-tuning a modified pre-trained text-to-image model on this synthetic dataset enables us to edit material properties in real-world images while preserving all other attributes. We show the potential application of our model to material edited NeRFs.
MindSearch: Mimicking Human Minds Elicits Deep AI Searcher
Abstract:
Information seeking and integration is a complex cognitive task that consumes enormous time and effort. Inspired by the remarkable progress of Large Language Models, recent works attempt to solve this task by combining LLMs and search engines. However, these methods still obtain unsatisfying performance due to three challenges: (1) complex requests often cannot be accurately and completely retrieved by the search engine once (2) corresponding information to be integrated is spread over multiple web pages along with massive noise, and (3) a large number of web pages with long contents may quickly exceed the maximum context length of LLMs. Inspired by the cognitive process when humans solve these problems, we introduce MindSearch to mimic the human minds in web information seeking and integration, which can be instantiated by a simple yet effective LLM-based multi-agent framework. The WebPlanner models the human mind of multi-step information seeking as a dynamic graph construction process: it decomposes the user query into atomic sub-questions as nodes in the graph and progressively extends the graph based on the search result from WebSearcher. Tasked with each sub-question, WebSearcher performs hierarchical information retrieval with search engines and collects valuable information for WebPlanner. The multi-agent design of MindSearch enables the whole framework to seek and integrate information parallelly from larger-scale (e.g., more than 300) web pages in 3 minutes, which is worth 3 hours of human effort. MindSearch demonstrates significant improvement in the response quality in terms of depth and breadth, on both close-set and open-set QA problems.
Abstract:
The misuse of large language models (LLMs) has drawn significant attention from the general public and LLM vendors. One particular type of adversarial prompt, known as jailbreak prompt, has emerged as the main attack vector to bypass the safeguards and elicit harmful content from LLMs. In this paper, employing our new framework JailbreakHub, we conduct a comprehensive analysis of 1,405 jailbreak prompts spanning from December 2022 to December 2023. We identify 131 jailbreak communities and discover unique characteristics of jailbreak prompts and their major attack strategies, such as prompt injection and privilege escalation. We also observe that jailbreak prompts increasingly shift from online Web communities to prompt-aggregation websites and 28 user accounts have consistently optimized jailbreak prompts over 100 days. To assess the potential harm caused by jailbreak prompts, we create a question set comprising 107,250 samples across 13 forbidden scenarios. Leveraging this dataset, our experiments on six popular LLMs show that their safeguards cannot adequately defend jailbreak prompts in all scenarios. Particularly, we identify five highly effective jailbreak prompts that achieve 0.95 attack success rates on ChatGPT (GPT-3.5) and GPT-4, and the earliest one has persisted online for over 240 days. We hope that our study can facilitate the research community and LLM vendors in promoting safer and regulated LLMs.
 | A guest post by
|