Tag: News

The StakingRewards contract on the BitTorrent Chain (BTTC) offers users the opportunity to stake tokens and earn rewards, providing a decentralized and automated mechanism for incentivizing participation, according to BitTorrent Inc..

The StakingRewards Contract: Your Gateway to Passive Income

The StakingRewards contract enables users to stake their tokens and earn rewards over a specified period. This smart contract is designed to create incentive mechanisms and encourage long-term participation in the BTTC token ecosystem.

State Variables and Structs: The Foundation

IERC20 public immutable stakingToken;
IERC20 public immutable rewardsToken;

address public owner;

// Duration of rewards to be paid out (in seconds)
uint public duration;
// Timestamp of when the rewards finish
uint public finishAt;
// Minimum of last updated time and reward finish time
uint public updatedAt;
// Reward to be paid out per second
uint public rewardRate;
// Sum of (reward rate * dt * 1e18 / total supply)
uint public rewardPerTokenStored;
// User address => rewardPerTokenStored
mapping(address => uint) public userRewardPerTokenPaid;
// User address => rewards to be claimed
mapping(address => uint) public rewards;

// Total staked
uint public totalSupply;
// User address => staked amount
mapping(address => uint) public balanceOf;

These variables form the backbone of the contract, defining the staking and reward mechanisms.

Events: Keeping Everyone Informed

Including events in the contract is a good practice for transparency and off-chain tracking. Suggested events include:

event Staked(address indexed user, uint amount);
event Withdrawn(address indexed user, uint amount);
event RewardPaid(address indexed user, uint reward);
event RewardDurationUpdated(uint newDuration);
event RewardNotified(uint amount);

Key Functions: The Heart of Staking and Rewards

Constructor

constructor(address _stakingToken, address _rewardToken) {
owner = msg.sender;
stakingToken = IERC20(_stakingToken);
rewardsToken = IERC20(_rewardToken);
}

The constructor initializes the contract with the addresses of the staking and reward tokens, and sets the contract deployer as the owner.

Modifiers

modifier onlyOwner() {
require(msg.sender == owner, "not authorized");
_;
}

modifier updateReward(address _account) {
rewardPerTokenStored = rewardPerToken();
updatedAt = lastTimeRewardApplicable();

if (_account != address(0)) {
rewards[_account] = earned(_account);
userRewardPerTokenPaid[_account] = rewardPerTokenStored;
}

_;
}

These modifiers ensure that only the owner can perform certain actions and that rewards are updated before any state changes.

Reward Calculation Functions

function lastTimeRewardApplicable() public view returns (uint) {
return _min(finishAt, block.timestamp);
}

function rewardPerToken() public view returns (uint) {
if (totalSupply == 0) {
return rewardPerTokenStored;
}
return
rewardPerTokenStored +
(rewardRate * (lastTimeRewardApplicable() - updatedAt) * 1e18) /
totalSupply;
}

These functions calculate the last time rewards were applicable and the reward rate per token.

Staking and Withdrawing Tokens

function stake(uint _amount) external updateReward(msg.sender) {
require(_amount > 0, "amount = 0");
stakingToken.transferFrom(msg.sender, address(this), _amount);
balanceOf[msg.sender] += _amount;
totalSupply += _amount;
emit Staked(msg.sender, _amount);
}

function withdraw(uint _amount) external updateReward(msg.sender) {
require(_amount > 0, "amount = 0");
balanceOf[msg.sender] -= _amount;
totalSupply -= _amount;
stakingToken.transfer(msg.sender, _amount);
emit Withdrawn(msg.sender, _amount);
}

These functions allow users to stake and withdraw tokens, updating their rewards accordingly.

Earning and Claiming Rewards

function earned(address _account) public view returns (uint) {
return
((balanceOf[_account] *
(rewardPerToken() - userRewardPerTokenPaid[_account])) / 1e18) +
rewards[_account];
}

function getReward() external updateReward(msg.sender) {
uint reward = rewards[msg.sender];
if (reward > 0) {
rewards[msg.sender] = 0;
rewardsToken.transfer(msg.sender, reward);
emit RewardPaid(msg.sender, reward);
}
}

These functions calculate the rewards earned by a user and allow them to claim their rewards.

Setting Reward Parameters

function setRewardsDuration(uint _duration) external onlyOwner {
require(finishAt     duration = _duration;
emit RewardDurationUpdated(_duration);
}

function notifyRewardAmount(uint _amount) external onlyOwner updateReward(address(0)) {
if (block.timestamp >= finishAt) {
rewardRate = _amount / duration;
} else {
uint remainingRewards = (finishAt - block.timestamp) * rewardRate;
rewardRate = (_amount + remainingRewards) / duration;
}

require(rewardRate > 0, “reward rate = 0”);
require(
rewardRate * duration “reward amount > balance”
);

finishAt = block.timestamp + duration;
updatedAt = block.timestamp;
emit RewardNotified(_amount);
}

These functions allow the owner to set the reward duration and notify the contract of the reward amount.

Utility Functions

function _min(uint x, uint y) private pure returns (uint) {
return x }

Beyond the Basics: The Power of Decentralized Staking

The StakingRewards contract offers several advantages, including:

• Transparency: All transactions and reward calculations are visible on the blockchain.
• Security: Smart contract logic ensures funds are handled correctly.
• Incentives: Encourages long-term participation and loyalty.
• Automation: Automatically calculates and distributes rewards based on staking activity.

Conclusion: Unlocking the Future of Staking

The StakingRewards smart contract on BTTC is more than just code — it’s a gateway to passive income and decentralized finance. By leveraging blockchain technology, a more transparent, efficient, and accessible staking ecosystem is created.

As users explore the potential of this contract on BTTC, it represents not just staking tokens, but pioneering a new era of decentralized finance and community participation.

Image source: Shutterstock

In the battle to conserve water, leaking pipes form an insidious enemy, accounting for average losses of 30 percent – but in some localities as much as 70 percent – of piped drinking water across the world. The leaks can be very difficult to pinpoint, but a new AI tool is making the process faster and more accurate.

A white X marks a leaking fire hydrant in the San Tan water district, just southeast of metropolitan Phoenix. The leak is a medium one of three to seven gallons per minute. That adds up quickly, with each day of leaking equivalent to the average water consumption of 43 households.

In this case, the typical giveaways of a leak aren’t present – no puddles, no unexpected greenery in the desert landscape, no caved-in street. The water has been seeping away through the bottom of the hydrant.

“That’s something that probably most likely would have leaked until the next fire hydrant maintenance cycle, which could be up to five years,” says Jacob Rogers, division director at EPCOR, the biggest private water utility in Arizona.

But this leak was detected sooner, thanks to an AI acoustic tool developed by FIDO Tech, whose technology not only detects leaks more precisely to their location, but also ranks them by size so that utilities can prioritize which to repair first. Oxford, U.K.-based FIDO is working with several utilities, such as EPCOR in Arizona and the State Water Commission in Querétaro, Mexico, as part of a Microsoft program to act responsibly by replenishing water in the same watersheds around the world where it has operations, including datacenters.

Leaks waste more than water

Stopping municipal water leaks prevents waste and increases water availability for the customers and communities that a utility serves. For the communities involved, leaks represent not only lost water, but also the cost of pumping it, of transporting it – which requires huge pipes and tremendous energy to push it, because it’s very heavy – of filtering it and of the chemicals needed to treat it. This lost water is called non-revenue water because it never arrives at a paying customer. Instead, everybody foots the bill.

“We’ve acquired troubled utilities in Arizona that had 30 to 40 percent water loss,” says Shawn Bradford, senior vice president of regulated water in Arizona and New Mexico for EPCOR, which has been working with FIDO under Microsoft’s program for about a year. “We’re having to pump 40 percent more water than they need just to cover the leaks that it takes to get it from the well or the water treatment plant to homes, and that’s a tremendous cost that all customers face.”

EPCOR has cut non-revenue water to around 10 percent from 27 percent, in part thanks to FIDO. FIDO provides sensors that are placed so they touch a pipe, whether that’s on top of an easily accessible pipe, on a hydrant, in a valve chamber or on a tap, FIDO’s Edwards says. EPCOR has 4,554 such sensors on its pipe network in the 160-square-mile San Tan service area. The sensors record everything, even the lowest, quietest noises.

“Sound propagates differently in different materials, just like it does in a school orchestra. You know, the longer the trombone, the longer the pipe, the deeper the sound compared to the trumpet, with shorter length and higher frequency. The largest leaks are the quietest – the human ear cannot hear them – especially in a plastic pipe,” Edwards says.

Utilities have long relied on acoustic devices to detect leaks. FIDO’s technology goes a step further, running the data its sensors have collected through a deep-learning AI tool that has learned to accurately determine whether a noise is caused by a leak or something else, like machinery or a train rumbling by. It can also assess how big the leak is and pinpoint its location. The other benefit of FIDO’s technology is that utilities’ technicians can easily interact with the AI in natural language.

EPCOR, too, had used acoustic leak detectors elsewhere, but “you have to spend a lot of time trying to figure out exactly where between those two devices the leak is actually located,” Bradford says. “The benefit of FIDO is it takes all that raw data and runs it through AI to help pinpoint where the leak is. FIDO in particular can do this on plastic pipe, which has always been a challenge for the industry – the leak doesn’t resonate like it does in metal pipe.”

The lack of resonance in plastic pipes means many acoustic leak detectors struggle. Yet most new pipe infrastructure is plastic, or PVC, because it’s lighter and easier to maneuver, making it easier to install and repair. On top of that, plastic pipe is better suited to the soil chemistry in the southwestern U.S., Bradford says, and dominates in the San Tan Valley, which has developed in the past three decades and continues to be among the fastest-growing U.S. communities. The utility does also have metal pipes – some are 60 to 70 years old and still in top condition. Pipes can break for many reasons besides age – for example, shifts in the ground from freeze-thaw cycles. In fast-growing San Tan, the leaks tend to occur in areas under construction.

Before FIDO, the utility relied on satellite imagery to spot such giveaways as green vegetation growing in unlikely places, Rogers says. But such greenery doesn’t sprout overnight – it takes days or weeks to grow. FIDO can find leaks much more quickly.

A fast and flexible solution

The AI also analyzes the utility’s network map to advise where to place sensors to avoid deaf spots, Edwards says. Results pour in immediately. “You can leave it overnight or for years at a time,” she says. The sensors are easy to move to monitor different swathes of the pipe infrastructure.

The sensors can also be placed on either side of a leak after it’s repaired, to make sure the fix is successful, or to spot whether the digging during the repair has caused another leak. And then the sensors can be moved to start detecting in a different part of the network. Meanwhile, FIDO’s AI can assess the likely volume of the leaking water so the repair work can be prioritized for maximum impact.

The leaky hydrant marked with an X turned out to have a faulty seal, and the repair was quick. Not all leaks are fixed as easily. Repairing buried pipes may require digging holes. Precision in locating leaks is essential to avoid tearing up streets, sidewalks or yards any more than necessary. It isn’t just a matter of minimizing traffic disruption – the utility is required to repave the street after the repair, which represents a significant cost that grows with the size of the excavation.

Image source: Shutterstock

Bitcoin, long regarded as the pioneer of digital assets, is evolving beyond its original identity as a store of value. While Bitcoin’s core proposition of decentralization, security, and stability remains intact, it is now starting to make its mark in decentralized finance (DeFi), according to BNB Chain Blog.

This new direction promises to unlock Bitcoin’s potential, making it more productive and useful for investors, developers, and institutions alike. The question is no longer if Bitcoin will participate in DeFi but rather how far it will go in shaping its future.

What is Bitcoin DeFi (BTCFi)?

Bitcoin DeFi refers to the integration of decentralized finance applications with the Bitcoin blockchain. DeFi uses blockchain technology and smart contracts to create financial services that do not rely on traditional intermediaries such as banks or brokers. Instead, DeFi platforms operate through autonomous protocols encoded on blockchain networks.

While Ethereum has historically dominated the DeFi space, Bitcoin is now emerging as a viable foundation for decentralized financial applications. This is made possible through layer-2 solutions and sidechains, which extend Bitcoin’s capabilities beyond its base layer to enable smart contracts, decentralized applications (dApps), and other DeFi functionalities.

Behind The Scenes – How Taproot Enables BTCFi

Taproot, activated on Bitcoin in November 2021, expanded the network’s capabilities by improving privacy, efficiency, and, crucially, programmability. Prior to Taproot, Bitcoin’s smart contract functionality was limited, restricting its ability to support more complex DeFi applications. Here’s how Taproot opened the path for Bitcoin DeFi:

• Enhanced Privacy and Efficiency: Taproot introduced Schnorr signatures, which aggregate multiple digital signatures into one. This reduced the amount of data required for complex transactions, making them cheaper.
• More Complex Smart Contracts: Before Taproot, Bitcoin’s scripting language was limited in its ability to handle advanced smart contracts. Taproot’s upgrade allows for more flexibility by enabling more complex conditions to be coded into transactions. This opens the door for applications like decentralized exchanges (DEXs), lending protocols, and automated market makers (AMMs) to be built on Bitcoin through layer-2 solutions.
• Scalability: Taproot increases Bitcoin’s block efficiency by optimizing how data is stored on the blockchain. This improvement is critical for supporting the execution of smart contracts and DeFi applications without overwhelming the network, allowing more transactions to be processed with less congestion.

Other innovations, such as wrapped tokens (like Wrapped Bitcoin or WBTC) and cross-chain bridges, facilitate DeFi activities by making Bitcoin interoperable with other blockchains, including Ethereum, while keeping the asset secured on Bitcoin’s blockchain.

Bitcoin DeFi Use Cases

Borrowing and Lending: Bitcoin DeFi platforms built on Bitcoin layers allow users to borrow and lend in a decentralized manner. For lenders, it offers opportunities to earn interest on idle Bitcoin, while borrowers can access loans without intermediaries, often using their Bitcoin as collateral.

Atomic Swaps: Bitcoin DeFi enables cross-chain atomic swaps, which allow users to exchange assets between different blockchains without intermediaries. This facilitates broader access to decentralized markets and platforms on other chains while keeping the trust and security benefits of Bitcoin.

Decentralized Exchanges (DEXs): Bitcoin-based decentralized exchanges allow users to trade cryptocurrencies directly without the need for a centralized entity. DEXs are typically built on layer-2 solutions, enabling Bitcoin to function in a peer-to-peer financial market.

Stablecoins and Asset Issuance: Bitcoin DeFi supports the creation of stablecoins and other digital assets, such as tokenized Bitcoin and NFTs. These assets can be issued and traded across decentralized applications, offering new ways for Bitcoin holders to participate in financial markets while maintaining the security of the Bitcoin blockchain.

Bitcoin DeFi on BNB Chain

BTCB – the tokenized version of BTC on BNB Chain – allows Bitcoin holders to access the rich BNB Chain ecosystem without needing to sell their bitcoins.

BTCB is created by wrapping Bitcoin through a 1:1 pegging mechanism, ensuring that each BTCB represents an equivalent amount of Bitcoin securely held by Binance. This opens up Bitcoin to participate in the broader DeFi space, enabling holders to access various services such as lending, borrowing, and yield farming on the BNB Chain, which offers faster transactions and lower fees compared to Bitcoin’s native network.

The integration of BTCB with BNB Chain also addresses security concerns through Binance’s custodial approach, where Bitcoin reserves are held in secure multi-signature cold storage wallets. Binance’s Proof of Reserves system provides transparency, allowing users to verify that all circulating BTCB tokens are fully backed by real Bitcoin.

This combination of trust, security, and access to DeFi services makes BTCB an essential bridge for Bitcoin holders to participate in the rapidly growing decentralized finance ecosystem on the BNB Chain, further enhancing the liquidity and utility of their BTC holdings.

The Future of Bitcoin DeFi

Bitcoin’s fundamental characteristics—robust security, decentralization, and censorship resistance—provide a strong foundation for DeFi applications.

As institutional interest in digital assets grows, Bitcoin’s increasing regulatory acceptance adds another layer of confidence, especially in light of its classification as a commodity in the U.S. This positions Bitcoin as a strong candidate for institutions looking to engage with decentralized financial applications in a compliant manner.

For BNB Chain, integrating Bitcoin DeFi goes a long way in its mission to onboard the next billion users into web3. Bitcoin’s mainstream acceptance and BNB Chain’s extreme utility and diversity could be a match made in heaven.

Image source: Shutterstock

The U.S. Securities and Exchange Commission (SEC) has issued a Wells Notice to OpenSea, the leading NFT marketplace, signaling a potential lawsuit based on the classification of certain NFTs as securities. According to blog.bitfinex.com, this regulatory move marks a new phase in the SEC’s scrutiny of the NFT market and raises concerns about the future of digital art and collectibles. By targeting OpenSea, the SEC could set a precedent that impacts not only NFT platforms but also creators and developers within this rapidly evolving corner of the digital asset space.

The SEC Targets the NFT Industry

The recent Wells Notice delivered by the SEC to OpenSea marks a pivotal moment for the NFT industry. The notice indicates that the SEC is considering classifying certain NFTs on the platform as securities, which could lead to significant legal challenges for OpenSea and broader implications for the NFT market. This development is notable given the expansive nature of NFTs, which range from digital art and collectibles to in-game assets and even event tickets. If the SEC moves forward with enforcement, it could set a precedent that would impact not only NFT platforms but also the creators, artists, and developers who rely on these platforms to distribute their work.

OpenSea, as one of the largest NFT marketplaces, has been a central hub for digital artists and collectors. The SEC’s potential action could derail the industry by imposing regulatory burdens on creators who may lack the resources to navigate complex securities laws. The broad application of securities law to NFTs could lead to discouraging artists from exploring new technologies and mediums. This is especially troubling in a space where digital art and collectibles have provided new avenues for creative expression and economic empowerment.

The legal uncertainty surrounding NFTs has been a growing concern, and the SEC’s approach to regulation by enforcement rather than providing clear guidelines adds to this. By targeting OpenSea, the SEC is venturing into territory that could extend beyond the NFT space to other forms of digital and physical collectibles. The industry is now faced with the challenge of defining the legal status of NFTs, which could have far-reaching consequences for the future of digital art and decentralized platforms.

In response to the SEC’s actions, OpenSea has pledged $5 million to support the legal defense of NFT creators and developers who may also face similar challenges. This move underscores the platform’s commitment to protecting the creative community and ensuring that innovation in the NFT space can continue. However, the outcome of this legal battle could shape the trajectory of the NFT industry for years to come, making it a critical issue for all stakeholders in the digital art and crypto communities.

What is a Wells Notice and What Kind of Ramifications Does This Have?

A Wells Notice is a formal communication from the SEC indicating that the agency is considering bringing an enforcement action against a company or individual for potential violations of securities laws. It outlines the charges being contemplated and gives the recipient an opportunity to respond before any formal charges are filed.

The ramifications of a Wells Notice can be significant, particularly for a company like OpenSea, which operates in the emerging NFT market. If the SEC proceeds with enforcement, it could lead to legal action that not only impacts OpenSea but also sets a broader precedent for how NFTs are regulated. This could potentially reclassify many NFTs as securities, subjecting them to strict regulatory requirements, which might deter innovation, affect market dynamics, and create costly legal challenges and regulatory barriers for artists, creators, and platforms operating within the NFT space.

Sweeping enforcement measures like the SEC’s potential action against OpenSea could have far-reaching ramifications beyond just cryptocurrency and NFTs, potentially extending into other types of collectibles such as physical art, trading cards, and even memorabilia. If the SEC successfully argues that NFTs should be classified as securities due to their potential for investment and resale, it could set a precedent for regulating a wide range of collectibles that have historically been treated as commodities or personal property, not financial securities.

Traditionally, securities have been defined as financial instruments like stocks, bonds, and investment contracts, which are tied directly to the financial performance of a company or enterprise. Extending this definition to include collectibles simply because they might appreciate in value or be resold could impose burdensome regulatory requirements on a vast array of goods that are currently outside the purview of securities law.

Such an interpretation could stifle markets, hinder innovation, and create legal uncertainties across industries that deal in collectibles. Artists, creators, and collectors might be forced to navigate complex legal frameworks designed for financial products, potentially dampening the appeal and accessibility of these markets. The broad application of securities law in this manner could have a deleterious effect on creativity and innovation, as creators might avoid exploring new forms of digital and physical art out of fear of regulatory repercussions. This may also impact traditional collectibles in unforeseen ways with sweeping negative impacts on a wide array of collectible products that were not under the SEC’s purview, previously.

Do Collectibles Suddenly Become Regulated Securities Just Because They’re Digital?

The question of whether collectibles suddenly become securities because they’re digital touches on a critical issue in the evolving landscape of digital assets. Traditionally, collectibles, whether physical or digital, have been treated as commodities or personal property, not securities. Their value is typically driven by factors like rarity, demand, and the reputation of the creator or brand, rather than by the expectation of profit tied to the efforts of a third party, which is a key criterion in defining securities under the Howey Test. However, as digital collectibles, particularly NFTs, have gained prominence, regulatory bodies like the SEC are beginning to scrutinize whether these assets should be classified differently, given their potential for speculation and resale in secondary markets.

The argument that digital collectibles could be considered securities often hinges on their perceived investment potential. NFTs, for example, are often bought with the expectation that their value will increase over time, especially if they are part of a popular collection or associated with a well-known artist. This expectation of profit could, in theory, bring them under the umbrella of securities law. However, this interpretation is problematic because it conflates the nature of collectibles with that of financial instruments designed specifically for investment purposes. Just because an item can appreciate in value and be resold does not necessarily make it a security; otherwise, many traditional forms of collectibles, from baseball cards to fine art, would also fall under this category, which they historically have not, and would previously have been considered ludicrous.

The application of securities law to digital collectibles could create significant legal and practical challenges. The art market, for instance, has operated for centuries without being subject to securities regulation, even though art pieces are frequently bought and sold as investments. Extending this regulatory framework to digital art and collectibles simply because they are traded on blockchain platforms could impose unnecessary burdens on creators and collectors, have a negative impact on the industry as a whole and potentially limit the accessibility of these markets. It could also lead to inconsistent and confusing regulatory environments.

Many would argue that digital collectibles should not automatically be classified as securities simply because they exist in a digital format. While the potential for investment and resale might be more apparent in the digital realm, such products are also viewed as collectibles. Opponents of any classification of NFTs as securities argue that regulatory bodies need to carefully consider the implications of broadening the definition of securities to include digital assets, as doing so could have unintended consequences for a wide range of markets. Instead, it has been suggested that a more nuanced approach that distinguishes between genuine investment products and collectibles, whether digital or physical, would better serve the interests of both consumers and creators.

Image source: Shutterstock

The advent of large language models (LLMs) has significantly benefited the AI industry, offering versatile tools capable of generating human-like text and handling a wide range of tasks. However, while LLMs demonstrate impressive general knowledge, their performance in specialized fields, such as veterinary science, is limited when used out of the box. To enhance their utility in specific areas, two primary strategies are commonly adopted in the industry: fine-tuning and retrieval-augmented generation (RAG).

Fine-Tuning vs. RAG

Fine-tuning involves training the model on a carefully curated and structured dataset, demanding substantial hardware resources, as well as the involvement of domain experts, a process that is often time-consuming and costly. Unfortunately, in many fields, it’s incredibly challenging to access domain experts in a way that is compatible with business constraints.

Conversely, RAG involves building a comprehensive corpus of knowledge literature, alongside an effective retrieval system that extracts relevant text chunks to address user queries. By adding this retrieved information to the user query, LLMs can produce better answers. Although this approach still requires subject matter experts to curate the best sources for the dataset, it is more tractable and business-compatible than fine-tuning. Also, since extensive training of the model isn’t necessary, this approach is less computationally intensive and more cost-effective.

NVIDIA NIM and NLP Pipelines

NVIDIA NIM streamlines the design of NLP pipelines using LLMs. These microservices simplify the deployment of generative AI models across platforms, allowing teams to self-host LLMs while offering standard APIs to build applications.

NIM abstracts model inference internals like execution engines and runtime operations, ensuring optimal performance with TensorRT-LLM, vLLM, and others. Key features include:

• Scalable deployment
• Support for diverse LLM architectures with optimized engines
• Flexible integration into existing workflows
• Enterprise-grade security with safetensors and constant CVE monitoring

Developers can run NIM microservices with Docker and perform inference using APIs. Specialized trained model weights can also be used for specific tasks, such as document parsing, by modifying container commands.

Reimagining Veterinary Care with AI

At AITEM, a member of the NVIDIA Inception Program for startups, collaboration with NVIDIA has focused on AI-based solutions across multiple fields, including industrial and life sciences. In the veterinary sector, AITEM is working on LAIKA, an innovative AI copilot designed to assist veterinarians by processing patient data and offering diagnostic suggestions, guidance, and clarifications.

LAIKA integrates multiple LLMs and RAG pipelines. The RAG component retrieves relevant information from a curated dataset of veterinary resources. During preparation, each resource is divided into chunks, with embeddings calculated and stored in the RAG database. During inference, the query is pre-processed and its embeddings are computed and compared with those in the RAG database using geometric distance metrics. The closest matches are selected as the most relevant and used to generate responses.

Due to potential redundancy in the RAG database, multiple retrieved chunks might contain the same information, limiting the diversity of concepts provided to the answer system. To address this, LAIKA employs the Maximal Marginal Relevance (MMR) algorithm to minimize chunk redundancy and ensure a broader range of relevant information.

NVIDIA NeMo Retriever Reranking NIM Microservice

The NVIDIA API Catalog includes NeMo Retriever NIM microservices that enable organizations to seamlessly connect custom models to diverse business data and deliver highly accurate responses. The NVIDIA Retrieval QA Mistral 4B reranking NIM microservice is designed to assess the probability that a given text passage contains relevant information for answering a user query. Integrating this model into the RAG pipeline enables filtering out retrievals that do not pass the reranking model’s evaluation, ensuring that only the most relevant and accurate information is used.

To assess the impact of this step on the RAG pipeline, AITEM designed an experiment:

1. Extract a dataset of ~100 anonymized questions from LAIKA users.
2. Run the current RAG pipeline to retrieve chunks for each question.
3. Sort the retrieved chunks based on probabilities provided by the reranking model.
4. Evaluate each chunk for relevance to the query.
5. Analyze the reranking model’s probability distribution in relation to the relevance determined in Step 4.
6. Compare the ranking of chunks in Step 3 against their relevance from Step 4.

User questions in LAIKA can vary significantly in form. Some queries contain detailed explanations of a situation but lack a specific question. Others contain precise inquiries regarding research, while some seek guidance or differential diagnoses based on clinical cases or analysis documents.

Due to the large number of chunks per question, AITEM used the Llama 3.1 70B Instruct NIM microservice for evaluation, which is also available in the NVIDIA API Catalog.

To better understand the reranking model’s performance, specific queries and model responses were examined in detail. Table 1 highlights the top and bottom reranked chunks for a sample query regarding differential diagnoses for a cat losing weight.

Figure 4 compares the reranking model probability output distribution (in logits) between relevant (good) and irrelevant (bad) chunks. The probabilities for good chunks are higher compared to bad chunks, and a t-test confirmed that this difference is statistically significant, with a p-value lower than 3e-72.

Figure 5 shows the distribution difference in the reranking-induced sorting positions: good chunks are predominantly in top positions, while bad chunks are lower. The Mann-Whitney test confirmed that these differences are statistically significant, resulting in a p-value lower than 9e-31.

Figure 6 shows the ranking distribution and helps define an effective cutoff point. In the top five positions, most chunks are good, while the majority of chunks in positions 11-15 are bad. Thus, retaining only the top five retrievals or another chosen number can serve as one way to effectively exclude most of the bad chunks.

To optimize retrieval pipelines, and minimize ingestion costs while maximizing accuracy, a lightweight embedding model can be paired with the NVIDIA reranking NIM microservice, to boost retrieval accuracy. Execution time can be improved by 1.75x (Figure 7).

Better Answers with the NVIDIA Reranking NIM Microservice

The results demonstrate that adding the NVIDIA reranking NIM microservice to the LAIKA RAG pipeline positively affects the relevance of retrieved chunks. By forwarding more precise, specialized information to the downstream answering LLM, it equips the model with the knowledge that’s necessary for highly specialized fields like veterinary science.

The NVIDIA reranking NIM microservice, available in the NVIDIA API Catalog, simplifies adoption as you can easily pull and run the model and infer its evaluations through APIs. This eliminates stress related to environment settings and manual optimization, as it comes pre-quantized and optimized with NVIDIA TensorRT for almost any platform.

For more information and the latest updates about LAIKA and other AITEM projects, see AITEM Solutions and follow LAIKA and AITEM on LinkedIn.

Image source: Shutterstock