Home web3.0 IVG: Integrating Human Values into Large Language Models at Inference Time

IVG: Integrating Human Values into Large Language Models at Inference Time

Oct 03, 2024 pm 03:16 PM
AI Integrated Value Guidance Implicit and Explicit Value Functions Token-Wise Sampling Chunk-Level Beam Search

Researchers developed Inference-time alignment methods to integrate human values after fine-tuning LLMs using the implicit and explicit functions without changing the base model.

IVG: Integrating Human Values into Large Language Models at Inference Time

Integrating human values after training a model with Learning-based algorithms requires fine-tuning LLMs, which is computationally expensive and time-consuming. Moreover, it generates biased and undesirable responses by the user. A model that can efficiently adapt to user preferences in real time by integrating algorithms that can interfere at inference time is needed. This method will avoid retraining the models repeatedly for desired results by freezing the base model and reducing the computational cost of fine-tuning LLMs.

Researchers developed Inference-time alignment methods to integrate human values after fine-tuning LLMs using the implicit and explicit functions without changing the base model. Implicit functions are used for token generation, which conducts word-by-word evaluations and prefers the output with the highest probability. In contrast, explicit functions require a rigid structure to evaluate larger chunks of text and generate the following sequence of words with the highest probability while maintaining overall context. The explicit function is inflexible and computationally expensive, failing to address token-level optimization, while the implicit function faces interpretability issues and requires frequent forward passes, leading to low real-time efficiency.

To tackle the disadvantages of both functions, the proposed method, Integrated Value Guidance (IVG), combines the implicit function’s token-level optimization and the explicit function’s broader perspective. It was able to ward off adaptation challenges and trade-offs in alignment efficacy, leading to decreased performance discrepancies and making it easier to implement. These advantages facilitated better performance on tasks like controlled sentiment generation and summarization. IVG, combined with the smaller models like GPT-2, could compete with higher models.

IVG incorporates the two value functions, the implicit and explicit functions, to align the model with human values. First, token-wise sampling fine-tunes individual tokens to a specific sequence length, generating multiple sequences. Then, chunk-level beam search compares the probabilities of these sequences and selects the one with the highest probability. Although this method ensures that the output is more robust, the computational power increases during the inference time due to frequent forward passes, leading to slower responses.

Researchers have used two experimental set-ups to evaluate IVG: 1. Controlled sentiment generation and Summarization, and 2. Instruction-following. In the first one, the GPT-2 model family is used by leveraging synthetic datasets from a gold-reward model to generate positive movie reviews and summarise Reddit posts. In comparison, the second one requires an instruction-tuned model, AlpacaEval 2.0. It employs Tulu Guidance, which uses specific models for implicit function and trains a reward-based model for the explicit function, and Ultraguidance, which fine-tunes a model with Direct Preference Optimization (DPO) for both functions. GPT-4-turbo was used as a reference to assess responses in the second experiment, and IVG consistently performed well.

In addition to these two experiments, an ablation study proved that Chunk-Level Beam Search (CBS) had higher speed efficiency than Emulator Fine-Tuning (EFT), which uses the implicit function for fine-tuning. These results have proved that CBS is much better to use in practice.

In conclusion, Integrated Value Guidance (IVG) offers a novel and efficient approach to aligning large language models with human preferences purely at inference time, bypassing the complexities of traditional fine-tuning. By leveraging implicit and explicit value functions, IVG enhances performance in both token-wise sampling and chunk-level decoding, as demonstrated through significant improvements in sentiment generation, summarization, and instruction-following tasks. The results showed that IVG is a versatile method, providing strong empirical evidence of its ability to outclass existing approaches, making it a promising solution for fine-tuning large models in real-world applications.

Don’t Forget to join our 50k ML SubReddit

Want to get in front of 1 Million AI Readers? Work with us here

The above is the detailed content of IVG: Integrating Human Values into Large Language Models at Inference Time. For more information, please follow other related articles on the PHP Chinese website!

Statement of this Website
The content of this article is voluntarily contributed by netizens, and the copyright belongs to the original author. This site does not assume corresponding legal responsibility. If you find any content suspected of plagiarism or infringement, please contact admin@php.cn

Hot AI Tools

Undresser.AI Undress

Undresser.AI Undress

AI-powered app for creating realistic nude photos

AI Clothes Remover

AI Clothes Remover

Online AI tool for removing clothes from photos.

Undress AI Tool

Undress AI Tool

Undress images for free

Clothoff.io

Clothoff.io

AI clothes remover

Video Face Swap

Video Face Swap

Swap faces in any video effortlessly with our completely free AI face swap tool!

Hot Tools

Notepad++7.3.1

Notepad++7.3.1

Easy-to-use and free code editor

SublimeText3 Chinese version

SublimeText3 Chinese version

Chinese version, very easy to use

Zend Studio 13.0.1

Zend Studio 13.0.1

Powerful PHP integrated development environment

Dreamweaver CS6

Dreamweaver CS6

Visual web development tools

SublimeText3 Mac version

SublimeText3 Mac version

God-level code editing software (SublimeText3)

Intel Markets (INTL) Could Be the Crypto of the Year as Cardano (ADA) Holders Diversify Ahead of Hard Fork Intel Markets (INTL) Could Be the Crypto of the Year as Cardano (ADA) Holders Diversify Ahead of Hard Fork Aug 25, 2024 am 03:49 AM

The crypto market is undergoing a significant sentiment shift as sidelined capital starts to take entry. Projects like Near Protocol (NEAR) and Cardano (ADA) are heating up in anticipation of the upcoming rally.

Shytoshi Kusama Hints at Forthcoming Collaboration with AI Project NFA Labs Shytoshi Kusama Hints at Forthcoming Collaboration with AI Project NFA Labs Aug 09, 2024 am 06:27 AM

Shytoshi Kusama, the enigmatic figure leading the Shiba Inu ecosystem, has sparked speculation about a forthcoming collaboration with an AI project.

Coinbase and Tether Unveil AI-Powered Platforms to Empower Developers in the Blockchain Space Coinbase and Tether Unveil AI-Powered Platforms to Empower Developers in the Blockchain Space Oct 29, 2024 am 03:24 AM

Coinbase's “Based Agent” platform and Tether's Local AI SDK aim to simplify the development of AI-driven cryptocurrency agents.

Apple AI Will Be A Game-Changer, AI Coins Rally Likely Next Week Apple AI Will Be A Game-Changer, AI Coins Rally Likely Next Week Sep 09, 2024 am 03:15 AM

Apple is all set for the iPhone 16 launch on Monday, gearing up for a major push to generative AI by introducing it to its consumers of iPhones

Firecoin Raises $1.2M to Bring AI-Powered Token Insights to the TON Ecosystem Firecoin Raises $1.2M to Bring AI-Powered Token Insights to the TON Ecosystem Oct 25, 2024 am 12:12 AM

Investing in the crypto market can be extremely lucrative, with new tokens occasionally making upward of 160,000% in yearly returns for investors.

Launchpool Incubates ONAI, an AI Ecosystem Based on the TON Blockchain Launchpool Incubates ONAI, an AI Ecosystem Based on the TON Blockchain Aug 05, 2024 pm 03:32 PM

This partnership signifies a crucial advancement towards integrating commercial AI agents and automation into the Web3 space.

Sui (SUI) and GoodEgg (GEGG): Two Promising Projects to Watch in September's Cryptocurrency Market Sui (SUI) and GoodEgg (GEGG): Two Promising Projects to Watch in September's Cryptocurrency Market Sep 12, 2024 pm 09:01 PM

As the cryptocurrency market faces fluctuating trends, savvy investors are beginning to shift their attention toward emerging projects that demonstrate resilience and growth potential. With concerns over Bitcoin's (BTC) volatile price trajectory foll

Despite 'Dead Coin” Narrative, Cardano (ADA) Maintains Top-Ten Position, Explores AI Integration Despite 'Dead Coin” Narrative, Cardano (ADA) Maintains Top-Ten Position, Explores AI Integration Aug 17, 2024 am 06:41 AM

In recent months, Cardano [ADA] has faced criticism, with some labeling it a “dead coin” due to its price trends. However, despite this negative