DoorDash's "All the Ads" Super Bowl contest is giving away nearly half a million dollars' worth of items seen in game day advertisements.

DoorDash to gift $50,000 home down payment, BMW in Super Bowl giveaway - CBS News

Thousands of drivers for Uber, Lyft and DoorDash are expected to go on strike in several U.S. cities on Valentine's Day seeking fair pay for their services.

Uber, Lyft, DoorDash drivers to strike on Valentine's Day seeking fair pay | Fox Business

Lyft and DoorDash report financial results this week. But first...

Uber Earnings Boost Gig Economy Ahead of Lyft and DoorDash Results - Bloomberg

The protests come a week after Lyft said it would guarantee weekly earnings for drivers.

Thousands of DoorDash, Uber, Lyft drivers plan to strike on Valentine's Day

A woman whose car was stolen at gunpoint in Richmond last November was able to buy a new set of wheels after nearly 500 people donated a total of almost $20,000.

Community raises $20K for DoorDash driver after getting carjacked at gunpoint in Richmond | WRIC ABC 8News

Thousands of drivers for ride-sharing platforms Uber , Lyft and food delivery app DoorDash are expected to go on strike across the United States on Valentine's Day for fair pay, drivers' groups said on Monday.

Uber, Lyft, DoorDash drivers to strike on Valentine's Day for fair pay | Reuters

Ride-sharing and delivery drivers are planning on striking across several U.S. cities located near airports on Valentine’s Day.

DoorDash, Uber, Lyft drivers plan to strike on Valentine’s Day

No doubt these delivery jobs offer flexible hours. You can work when you want. A Springfield DoorDash driver nearly lost his paycheck to a scammer. It all started with a bogus order. He emailed On Your Side, so a crook doesn’t steal from you.

On Your Side: Delivery driver scam

Thousands will strike ride-sharing platforms and food delivery on Wednesday, drivers' organizations say.

Uber, Lyft and DoorDash drivers set to walk off the job on Valentine's Day - CBS News

DoorDash

doordash

Uber Eats, Grubhub, Postmates, and DoorDash have been locked in heated competition for years now, vying for dominance in the “ ” Food Delivery War. However, DoorDash overtook its competitors with a 55% market share in March 2021 and 56% of all food delivery orders in America. DoorDash has evolved from a small, student-founded startup to the most successful online food delivery platform in America in just seven years. It has even broken into the Australian, Canadian, and Japanese markets⁠. Pretty impressive for a company that is not even a decade old. Founding In 2012, Stanford students Tony Xu, Stanley Tang, Andy Fang, and Evan Moore began working on an app for small business owners. They weren’t sure what they wanted to build, only that they wanted it to be used by small businesses. So they interviewed hundreds of business owners in the area, asking what their ventures were like and if there was anything in particular that they needed help with. They tried setting up iPads by the retail point of sale so that customers could answer a short marketing attribution survey. But, unfortunately, while some customers were willing to answer the survey, it wasn’t conducive to the business. On a subsequent occasion, after interviewing a macaron store owner, they overheard her turning down a delivery order. That was when they realized that food delivery was an issue for restaurants in the area. After asking around again, they found out that very few restaurants deliver due to inconsistent orders. Small businesses can't afford to have their own fleet of delivery drivers when they might receive several delivery orders one day and virtually none the next. Palo Alto Delivery The four students launched PaloAltoDelivery.com, which only had a Google Voice number and menus in PDF format from a few local restaurants. They charged a flat rate of $6 per delivery with no minimum order size, and the four of them personally handled the deliveries. Once business took off, they started receiving more orders than they could handle. So they began hiring people to help with deliveries, hung flyers, and also posted on Craigslist. DoorDash and YCombinator In 2013, after successfully pitching DoorDash to investors in YCombinator, the company received $120k in seed capital and officially renamed the company from Palo Alto Delivery to DoorDash. Growth In 2015, DoorDash was operating in 18 different cities and received $60 million in investments. With the startup's first launch in Canada in 2016, Khosla Ventures and Kleiner Perkins invested another $127 million in the company. Surprisingly though, DoorDash’s valuation went down to $700 million despite the increase in investments. First Acquisition In 2017, DoorDash acquired Rickshaw, a startup that focused on delivery and logistics. At the time, DoorDash was working on its own platform, DoorDash Drive. By acquiring Rickshaw, DoorDash could integrate some aspects of Rickshaw’s software with DoorDash Drive. DoorDash the Unicorn In 2018, when SoftBank invested $535 million into DoorDash, the food delivery service reached a ‘unicorn’ status with a valuation of $1.4 billion. Continued Growth By the end of 2018, DoorDash managed to overtake Uber Eats as the second most popular food delivery service in America. Expansion continued, and by 2019, the company opened DoorDash Kitchen - a ghost kitchen in California. DoorDash also acquired Caviar, a food delivery app with listings from high-end restaurants that didn’t deliver outside of the app. The Delivery Boom When the COVID-19 pandemic hit in 2020, the on-demand food delivery industry boomed. Shelter-in-place orders and a huge reduction in foot traffic meant that restaurants had to rely on services like DoorDash, Uber Eats, Postmates, and Grubhub if they wanted to make any sales at all. The massive surge in orders allowed market leader DoorDash to rake in $1.92 billion in revenue, but despite that, DoorDash wasn't generating profits. Instead, the company was actually losing money. Going Public In December 2020, DoorDash went public with an IPO of $102 but ended the day with $189.51. Despite its lack of profitability, investors seem interested in DoorDash, but . not everyone shares that interest Still, some analysts believe that if DoorDash continues to grow even after the pandemic-induced food delivery business boom ends, it might become profitable in the future. The DoorDash Strategy What made DoorDash stand out from its competitors? The hefty investments it received undoubtedly played a part in its success, but it’s not just because of the money. Simplicity First, DoorDash kept things basic. The app is simple, intuitive, and easy to use—qualities end-users want. They made convenience and speed a priority and provided sellers with a user-friendly app that lets them accept orders without ever speaking to a customer. Using Technology As A Driving Force DoorDash has been able to stay ahead of its competitors by being technologically driven and looking for ways to expand into new markets. For example, one innovative part of their business model was that customers could order food from any restaurant in their service area, which helped boost their market share immensely during the COVID-19 pandemic. The company partnered with both large and small businesses, not just from the inner city but also from the suburbs. Additionally, it provided merchants invaluable data like brand and consumer insights. This kind of data is beneficial to smaller shops that may not have the resources to gather it independently. In its earlier days, DoorDash would also give discount codes or coupons to users who would review it on the app store. This feedback led to improvements in the app. Open For Expansion Lastly, DoorDash is keeping its eyes on the future. Although it’s primarily a food delivery service, the company wants to branch out more in the future. Its goal is to provide an on-demand delivery service, not just for food but for virtually anything that needs delivering. Final Words Even at its best, the economy on which the on-demand delivery industry is based can still be unstable. For now, though, DoorDash maintains its position as a market leader in online food delivery and will likely continue to grow in the future.

How DoorDash Has Been Winning the Food Delivery War

A few years ago, food technology looked like the next big thing. Meatless meat, food delivery platforms, and vegetables you can grow in the city all seemed poised to disrupt the way we get, prepare, and eat our meals. Today, however, you may wonder where all that momentum went. Animal protein and conventionally grown crops still dominate store shelves. Restaurants have largely gone back to the way they operated before the COVID-19 pandemic. So what happened? Why is food tech faltering, and what does that mean for its future? The Fall of Food Tech The COVID-19 pandemic brought a wave of tech-driven disruption across industries, and food was no exception. Plant-based meat sales in the early days of the pandemic, and app-based delivery services like HelloFresh and DoorDash became the new norm. grew a whopping 200% Today, alternative meats make up less than 1% of all meat consumed, and customers have flooded back to restaurants. Food tech startups across all sub-sectors saw a in the second quarter of 2022. Individual deals fell by almost a third. 21.5% decrease in funding If you look through the news, you’ll likely find a lot fewer stories about exciting, up-and-coming food tech disruptors than you used to. Some aspects of food tech haven’t completely sputtered out. Delivery apps may not have replaced older methods of getting food, but they’re still prominent. For the most part, though, food tech has failed to live up to the hype. Why Is This Happening? Food tech’s drop in funding and innovation comes from a few separate factors. Understanding these is key to seeing how the industry could evolve and grow from here. Disappointing Consumer Performance One of the biggest reasons behind food tech’s fall from grace is simply that customers didn’t latch onto it the way early investors expected. After the initial surge in sales, alternative protein failed to generate repeat customers. Many people were excited about the prospect of something new, but when it didn’t match their expectations, they didn’t try it again. According to one survey, who tried plant-based meat once but didn’t buy it again cited its taste as their reason for doing so. 46% of the people Similarly, 35% said it didn’t meet their overall expectations, and 30% said it was too expensive. Whether this failure to match consumer expectations is a matter of marketing or something on the customers’ end is unclear. Regardless of the cause, though, it meant food tech’s early 2020 surge couldn’t sustain itself. As sales dropped, so did investors’ trust in the industry. Regulatory Hurdles As food tech has drawn more attention, it’s also run into some regulatory roadblocks. Much debate has arisen around whether you can call plant-based foods “meat” or “milk” when they don’t contain any actual meat or milk. A lack of clear rules from government agencies further muddies the waters. The FDA has on labeling for plant-based alternatives to animal products. Some local governments have imposed their own guidelines, but these vary from state to state. yet to release formula guidance As a result, it can be hard for new food tech businesses to guarantee regulatory compliance. Plant-based foods aren’t the only food tech sector facing legal challenges. The National Advertising Division has started over its claims that canceling its services is easy. investigating meal kit company Blue Apron Cases like this bring food tech’s dramatic claims under scrutiny and may scare new businesses away from the market. An Overcrowded Market The food tech industry also suffered from too much growth in too short a time frame. Excitement over the sector’s initial growth led to an influx of startups, all offering remarkably similar products and services. As a result, the market quickly became overcrowded. There are more than with more than $500,000 in sales today. That’s 15 more than the older, more established alternative dairy industry. 60 plant-based meat companies A sector that has yet to convince the public that it’s worth its price premium and taste can’t support that many competitors. What’s In Store for the Future of Food Tech? Given these massive challenges, is this the end of the food tech industry? Not necessarily, but the market will likely struggle for a few more years. Food tech will ultimately survive and grow because it’s a necessary step forward. Bioengineered crops , and alternative meat can dramatically reduce greenhouse gas emissions. improve crop yields and resilience As climate change becomes a more severe and publicized threat, those benefits are hard to ignore. Some of food tech’s challenges will also fade with time and development. Prices will fall and tastes will improve as technology advances and these companies go through trial and error. However, that could be a lengthy process, certainly too long to sustain dozens of near-identical companies right now. In the near-term, many of these startups will likely have to close their doors or become part of larger companies. Food tech businesses should also focus on innovation to meet current challenges and set themselves apart. The industry will eventually thrive, but that timeline is longer than investors initially expected. Food Tech Will Survive, But Growth May Be Slow Food tech is a necessary industry. However, many startups misinterpreted early demand peaks and tried to grow the sector too much too quickly. The resulting crash was understandably startling for players in the market. As technology improves and the market thins out, food tech will reach its disruptive potential. That will just take more time, and growth will be less disruptive than some companies would like.

Why Is Food Tech Seeing a Dropoff?

Workers fight for hazard pay, protective equipment, and better sick leave With and similar orders for “social distancing” in place nationwide, more and more of us are inside, relying on so-called to handle our business outside. To pick up our medicine. To shop for our groceries and deliver our takeout food. To drive us to urgent appointments in lieu of public buses and trains. They face far greater exposure to the coronavirus we’re fleeing, often for a low financial return. 250 million Americans now ordered to shelter at home gig workers The pandemic has given rise to the language of war, and gig workers are among those braving the “ ” in our battle with COVID-19. But it’s also clear that these contractors, who work on-demand jobs and who are lacking health insurance, are among the most vulnerable in our workforce, facing financial and health risks with few benefits or safety-net protections. front lines often immigrants This week, workers at and staged protests asking for hazard pay, more access to paid sick leave, and hand sanitizer and disinfectant wipes, among other demands. On the eve of the strike, Instacart announced that it was working with a third-party to and removing from the app the option not to tip (defaulting instead to the amount of the customer’s previous tip). Instacart, Amazon Whole Foods manufacture sanitizer to distribute to its shoppers But workers went on strike anyway, . They said the changes should have happened sooner, that issues of hazard pay and sick leave should have been addressed, and that the change in the default tip amount was not enough. saying the offer was insulting Gig workers’ fight for more protections preceded the arrival of the novel coronavirus; since at least 2013, Lyft and Uber have faced class-action lawsuits filed on behalf of gig workers pushing to win classification as employees rather than contractors in order to gain rights to more benefits. One such in 2017 without reclassifying workers, but others continue. The pandemic has only heightened the stakes of the battle. lawsuit settled for $27 million How Are Companies Trying to Help Gig Workers? The major tech companies have announced they will follow safety guidelines from and other authorities about hand washing, using hand sanitizer, and maintaining social distance. Food delivery apps, like DoorDash and GrubHub, have enacted “no contact” delivery options, allowing delivery people to drop off customers’ food at their doorsteps rather than face-to-face. the Centers for Disease Control For ride-hailing drivers, who sit within an arm’s length of their passengers, close contact is hardly avoidable. One of the first dozen confirmed coronavirus cases in New York City was . a Queens man who drove an Uber And in California, an Uber driver was recently quarantined and is suspected of having COVID-19. A popular blog covering the ride-hailing industry, The Rideshare Guy tells of picking up an Uber Pool passenger in early March who casually admitted he had been infected with the virus after a trip to Taiwan. The driver said he wiped down the seat, washed his hands, and sanitized everything. But later he took a passenger to the hospital who said he had been coughing up blood. that driver’s story The California Uber driver, now under quarantine, told The Rideshare Guy that he received a payout of about $2,100 from Uber after some back and forth with the company, an amount equal to 14 days of his average daily earnings over the last six months. Several tech companies have agreed to provide paid sick leave in the event that workers are diagnosed with COVID-19. , , , and have said they will provide up to 14 days of financial assistance for diagnosed or quarantined workers. Uber Postmates Instacart DoorDash Additionally, DoorDash will provide funds to proven housemates of others who are diagnosed, and Postmates has a fund for further financial help, regardless of diagnosis, provided that a worker lives in an area with reported cases of COVID-19, has completed at least one delivery since Feb. 25, and has a health savings account through Postmates. is also providing funds for workers diagnosed with COVID-19, though for an unspecified period. Lyft What Are Tech Companies Not Doing? At least two companies have struggled to provide protective gear for their workers, facing the same the rest of the nation is facing. Lyft announced plans to distribute , but drivers about not being able to get the supplies after Lyft . dwindling supplies 200,000 bottles of hand sanitizer complained on social media closed its driver hubs A spokesperson for Lyft, Campbell Matthews, said many supplies have been distributed to drivers at no cost to them, and the company is working to distribute the rest. Workers at Instacart, whose grocery shopping service has boomed during the pandemic, are so in need of disinfectant wipes and hand sanitizer that they went on strike for a day earlier this week, sounding an alarm that rang at . news outlets across the country Sarah Polito, an Instacart shopper based in Newark, N.Y., that she spent $50 on disinfectant wipes and hand sanitizer as the coronavirus outbreak intensified. On each job, Polito wiped the inside of her delivery bags and the handlebar of her shopping cart. told MarketWatch Still, she felt anxious. “You get flooded with emotion,” she told MarketWatch. Polito was among the Instacart workers who walked off the job, believing she deserves hazard pay. “Anybody who works at this point deserves hazard pay,” she said. The gig workers’ newly instituted sick leave policies also have loopholes. The policies of Uber, Lyft, and Instacart do not cover contract workers who have not yet been diagnosed with COVID-19 but who, for example, are feeling sick and deciding whether to take time off or continue working and risk infecting others. The Instacart workers who walked off their jobs this week similarly demanded broader access to paid sick leave not dependent on a COVID-19 diagnosis. “What I would love to see them do is take a little more initiative to provide some sort of sick pay to drivers who are feeling sick and not quarantined,” said Harry Campbell, founder of The Rideshare Guy, referring to Uber and Lyft. In California, gig workers theoretically have more protections but may not be getting them. A law went into effect in California on Jan. 1 that , putatively giving them rights to a minimum wage, unemployment insurance, and sick leave, but , among other companies, have refused to comply with the law. reclassifies most gig workers as employees Uber, Lyft, and Postmates Uber, Lyft, and DoorDash pledged to spend $90 million on a California ballot , and Uber and Postmates arguing that the law is unconstitutional. initiative to exempt them from the law filed a lawsuit late last year Industry officials have said that designating drivers as employees costs by 20 to 30 percent, risking their businesses. Lyft commissioned a study that found that a traditional 40-hour work week would . would raise force the company to let go of more than 300,000 drivers Before the law passed, the companies : providing the contractors with a minimum wage, portable benefits, and collective bargaining without classifying them as employees. But labor officials, including the state building trades council, which represents construction workers, did not support the proposal, fearing that exempting Uber and Lyft . argued for a third way would lead to similar exemptions in other industries What Can You Do? With some models of the pandemic forecasting a , and authorities telling citizens that we could be stuck at home for just as long, the problem of how to get essential items from the outside world is not going away soon. monthslong battle with the coronavirus and many more deaths For those who wish to eliminate the risk gig workers are taking on their behalf, there are now numerous on how people who are performing these essential tasks themselves can lessen their risk of contracting the coronavirus. sites with instructions However, many people are quarantined or otherwise physically unable to leave home, and gig workers have taken these jobs (and ) because they need the work. For that reason, some advocates suggest that people who need to rely on gig workers consider . After all, these workers are risking their own well-being for your grocery delivery. continue to sign up for them tipping very, very highly Credits Originally published as " How Are Gig Workers Faring During the COVID-19 Crisis? " with the Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0) license.

How Are Gig Workers Coping During the Crisis

Food ordering industry is the one which has a huge potential. According to the research by Morgan Stanley, online food delivery comprises only less than 5 percent of the total restaurant industry. This article give you some idea about how to build a food ordering app like Eat24, JustEat, Deliveroo, Doordash, UberEats, Zomato. Key Stakeholders The key stakeholders for a food delivery app like Zomato are: Platform Owners Restaurant Partners Delivery Partners Customers A more advanced version of the food delivery app need the following modules: Web app for the customers Mobile app for the customers Menu manager (web app) for the restaurants Order receiving mobile app for the restaurants Mobile app for the delivery partners How does the food delivery app work? The food delivery app charges a fixed amount of fee from the restaurant owners to list their restaurant on the platform. The customers can browse the menus and make the order from the mobile app or from the website. The restaurant receives the notification when new orders come and they have the option to either reject or accept it. When a customer places an order, the system sends the notification to the delivery person about the order, and they can accept or skip the order. The customer can track the order in real-time through the mobile app or website. The food delivery app pays a variable amount to the delivery person for each successful delivery. Revenue Model The food delivery app can make revenue through 3 channels: Subscription The platform charges a fixed amount from the restaurant owners to list their restaurant on the platform. Advertisements The restaurants can buy a banner to have a premium listing on the platform. These premium listings are the sponsored ads shown on the website or mobile app to grab the customer's attention. Commission The platform takes a cut out of the order total from the restaurant. The commissions typically range from 5-20% of the order value. Features of the online food delivery app like Zomato Customers (Web and Mobile App) Customers can access the restaurants via the website or the mobile app. Both the mobile app and website share the common features. Let's see what are the features of the web and mobile app for the customers. Sign up Sign in Search Restaurants Browse Menus Add/Remove Cart Items Checkout & Online Payments Track the Order Rate the Order Sign up ​The users should be able to sign up using their email. It would be nice to add social media login - Facebook or Google. Sign in Users can log in using their email and password. It would be nice to add an OTP based login - the user will receive the OTP via email. ​Search Restaurants ​After the user shares the location with the platform, the platform shows restaurants that deliver the food to the user's location. The users can search for restaurants using the name and filter the restaurants by Rating, Cuisines, etc. ​Browse Menus The users can browse the restaurant menu. They can search the menu by name and filter the menus by veg/non-veg. ​Add/Remove Cart Items The users can add items to the cart and remove the items. Some menu items might have customization (or modifiers), which allows the users to customize that item. ​Checkout & Online Payments When a user is ready to order, he/she can do the checkout.  The users should be able to make payments using options such as Card, Wallets, Net Banking, UPI or Cash on delivery ​Track the Order After a successful order, the user can track the order progress in the Google map. They can see current order status, driver location, estimated delivery time, etc. ​Rate the Order The users should be able to rate the order after successful delivery. ​Previous Orders The users can see their previous orders and they should be able to re-order in one click. ​Address Book Users can manage their addresses in the address book. The users can add the different addresses for Home, Work or Other. ​ Notifications The users receive push notifications at different stages of the orders - Approved, Preparing, On the way, Delivered. The users receive notifications also via SMS or E-mail. Restaurant Partners (Web and Mobile App) This application helps the restaurants to manage the menus, orders, etc. They get access to the following: Menu Manager A web app to manage the menus, categories, items, and modifiers. Order Receiver A mobile app to manage the orders, set the menu availability, with minimum features of the Menu Manager. Let's see what are the features of the restaurant partners web and mobile app. Manage Menus Manage Categories Manage Items Manage Modifier Store Settings Update Store Hours Orders History Payments Deposit History Ratings & Review Sign In The restaurant partners sign in using their email and password. The restaurant partner on-boarding process is handled by the Administrator. Manage Menus The restaurant partners can do the followings: Add menus Update menus Remove menus Update the menu's availability Manage Categories The restaurant partners can do the followings: Add categories Update categories Remove categories Manage Modifier The restaurant partners can do the followings: Add modifiers Remove modifiers Manage Items The restaurant partners can do the followings: Add items Update items Remove items Update the items' availability Assign modifier to the items Store Settings The restaurant can update their basic details such as name, address, logo, phone, email, etc. Store Hours The restaurant can update their working hours to show when the restaurant is open for business on the Platform. Orders History The restaurant can browse the orders and filter it by date ranges.  They can see orders total, amounts, commission deducted, etc. Revenue The restaurant can see how their business is doing from the Revenue page. Deposit History The restaurant can view deposits made to their account From the Deposits page. They can view the deposit amount, download the deposit history as CSV. Ratings & Reviews The restaurant can see the reviews about their menus from the diners. They can respond to the reviews but can't remove or update reviews. Delivery Partners (Mobile App) Delivery partners pick the orders from the restaurants and deliver them to the customers. They get a delivery request and as soon as they accept the request, they are provided with the information on the order, restaurant, and address of the customer. Let's see what are the features of the delivery partner mobile app. Sign In Profile Orders Earnings Settings Sign In The delivery partners can sign in using their mobile number and password. The delivery partners have to receive their account information from the Platform Administrator. Profile The delivery person can change their mobile number, password, etc. Orders The delivery partners receive the orders from the platform, they can either accept or reject the orders. If they accept the order, they are provided with the information on the order, restaurant, and address of the customer. They can see the route to the customer's location on the Google map and has the option to contact the customer/restaurant. Earnings The delivery person can see their everyday earnings. They receive Delivery Pay + 100% of tips for every order they deliver. Settings The delivery person can change their availability online or offline. Only partners with status online receive orders for delivery. Platform Administrators (Web App) Platform Administrators are the owners of the food delivery app and have full privileges to the system. They can view the restaurant partners, delivery partners, customers, orders, etc and make changes. Let's see what are the features of the platform administrator web app. Manage Restaurant Partners Manage Delivery Partners Manage Orders Manage Customers Manage Reviews & Ratings Analytics and Reporting Financials Dashboard Manage Restaurant Partners The Administrators can create, update, remove the restaurant partners. They can also create or update the menus, categories, items or modifiers of the restaurants. Manage Delivery Partners The Administrators can create, update, remove the delivery partners. Manage Orders The Administrators can see all the orders received for the restaurants. They can add notes, approve, cancel, refund the orders. Manage Customers The Administrators can view all the customers registered on the platform. They can block or unblock the customers. Manage Reviews & Ratings The Administrators can view the reviews and ratings from the customers for each order. They can hide the reviews from the public or mark it as spam. Analytics and Reporting Analytics and reporting are the most important things for the Administrators to track how the business is doing. They can download different reports from the dashboard and also see analytics about the revenue, orders, customers, etc. Technology Stacks Choosing the right technology stack is very crucial when building a food delivery app. You have to look into future growth and how are you going to maintain the platform. Web App Development You can use either Angular or React for the frontend. Mobile App Development You have 2 options here. Build a hybrid mobile app or a native one. For the hybrid mobile app development, you can use Flutter or React Native. Backend Development You have many options here. We would prefer PHP or NodeJs. You can use a well-known PHP or NodeJs framework for backend development. Databases You can use MySQL, MongoDB. 3rd Party Services You might need the following 3rd party services for the platform operations. SMS: Twilio Email: Sendgrid Location Tracking: Google Map Push Notifications: Firebase Development team to build a food delivery app like Zomato You can hire a that has experienced building an . web and mobile app development team online food ordering app like Zomato or UberEats For building a food delivery app like Zomato, you will need the followings: 1. UI/UX designers 2. Backend developers 3. Frontend developers 4. QA team 5. Project manager You can visit our website to know more about the solutions. Learn more about multi-restaurant food ordering platforms development . Previously published at https://www.cedextech.com/blog/app-like-zomato-ubereats

How to Build Food Delivery App Like Zomato

Table of Links Abstract and 1. Introduction Abstract and 1. Introduction 2 Related works 2 Related works 3 Methodology and 3.1 Causal language model as a classification model 3 Methodology and 3.1 Causal language model as a classification model 3.2 Functional token 3.2 Functional token 3.3 Dataset collection 3.3 Dataset collection 3.4 Model development and training 3.4 Model development and training 4 Experiments and 4.1 Android function calls 4 Experiments and 4.1 Android function calls 4.2 Extension to Vehicle, Yelp, and DoorDash function sets 4.2 Extension to Vehicle, Yelp, and DoorDash function sets 4.3 Full and partial training datasets and 4.4 Full training and LoRA training 4.3 Full and partial training datasets and 4.4 Full training and LoRA training 4.5 Parallel and nested function call and 4.6 Weighted loss function for special tokens 4.5 Parallel and nested function call and 4.6 Weighted loss function for special tokens 5 Discussion and future works and References 5 Discussion and future works and References Appendix Appendix A.1 Android function examples A.1 Android function examples A.2 Vehicle function examples A.2 Vehicle function examples 4.2 Extension to Vehicle, Yelp, and DoorDash function sets In addition to Android function calls, we expanded our evaluation to include 20 vehicle function calls, showcasing the algorithm’s adaptability to diverse use cases. For vehicle functions, we focused on essential control methods such as volume adjustment, air conditioning, and seat positioning. We conducted benchmarks for vehicle functions paralleling the Android function evaluation, observing consistent performance patterns. Details on vehicle functions are provided in the Appendix, enabling users to customize a new set of functional APIs for their specific needs. Furthermore, tests conducted with Yelp and DoorDash APIs confirmed a similar performance, underscoring our method’s versatility across various function sets. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. available on arxiv Authors:
(1) Wei Chen, Stanford University, with equal contribution and a corresponding author {weichen6}@stanford.edu;
(2) Zhiyuan Li, Stanford University and a corresponding author {zhiyuan8}@stanford.edu. Authors: Authors: (1) Wei Chen, Stanford University, with equal contribution and a corresponding author {weichen6}@stanford.edu; (2) Zhiyuan Li, Stanford University and a corresponding author {zhiyuan8}@stanford.edu.

Demonstrating Adaptability: Evaluating Function Calling on Vehicle, Yelp, and DoorDash APIs

Unrealistic Valuations and Getting Back to Fundamentals It’s almost as if creating a company with a valuation of $1B+ is the single most important goal of entrepreneurs rather than creating a profitable and sustainable business. In the on-demand food/delivery space, there’s a ton of competition and a race to take the lead. Instead of achieving this goal through pure business growth fundamentals, popular thinking seems to dictate that the easier route is to just gain status and use the value of the company as the metric of market leader. (Granted, should be funding companies that have actual solid growth metrics or at the very least a leadership team that has absolutely zero doubt in their minds that they will win no matter what.) unicorn logical VCs The concept of enormous funding rounds and stretched valuations paving an easy road to success/growth may be changing. This is highlighted by recent failure of a new funding round at a supposed $1B valuation. DoorDash Perhaps racing to goals for investment should not be the primary objective. Perhaps companies should be focusing on core growth fundamentals such as the following strategies: Beating competition in quality vs. price. For example, a similar service like selects only top quality restaurants to feature in a local market. They don’t feature brands like Denny’s. The choose quality vs. quantity. Caviar Walking before running. Nailing operations. Gaining traction in the market, becoming the leader in the locale and moving onto another. Learning from and don’t disclose the capital raise amount or valuation. Wallapop’s leadership Focusing on getting good PR for outstanding service, an exceptional value proposition and building a fun and trustworthy brand, rather than the latest Techcrunch story on closing the Series B. The goal should not be to raise an obscene amount of capital which can lead to an artificial valuation. Note what happened with Gilt Groupe, as they once had a $1B valuation but we’re recently acquired by Hudson Bay (SAKS) for only $250M, less than the total amount of funding they took before being bought. Maybe it’s time for startups need to get back to small business fundamentals. (you know, those principles of “ ” that have been around for thousands of years…) buy for one, sell for two Just a thought…

Lessons From DoorDash’s Unicorn Quest

Table of Links Abstract and 1. Introduction Abstract and 1. Introduction 2 Related works 2 Related works 3 Methodology and 3.1 Causal language model as a classification model 3 Methodology and 3.1 Causal language model as a classification model 3.2 Functional token 3.2 Functional token 3.3 Dataset collection 3.3 Dataset collection 3.4 Model development and training 3.4 Model development and training 4 Experiments and 4.1 Android function calls 4 Experiments and 4.1 Android function calls 4.2 Extension to Vehicle, Yelp, and DoorDash function sets 4.2 Extension to Vehicle, Yelp, and DoorDash function sets 4.3 Full and partial training datasets and 4.4 Full training and LoRA training 4.3 Full and partial training datasets and 4.4 Full training and LoRA training 4.5 Parallel and nested function call and 4.6 Weighted loss function for special tokens 4.5 Parallel and nested function call and 4.6 Weighted loss function for special tokens 5 Discussion and future works and References 5 Discussion and future works and References Appendix Appendix A.1 Android function examples A.1 Android function examples A.2 Vehicle function examples A.2 Vehicle function examples A.1 Android function examples This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. available on arxiv Authors:
(1) Wei Chen, Stanford University, with equal contribution and a corresponding author {weichen6}@stanford.edu;
(2) Zhiyuan Li, Stanford University and a corresponding author {zhiyuan8}@stanford.edu. Authors: Authors: (1) Wei Chen, Stanford University, with equal contribution and a corresponding author {weichen6}@stanford.edu; (2) Zhiyuan Li, Stanford University and a corresponding author {zhiyuan8}@stanford.edu.

Android Function Examples That You Should Know

Table of Links Abstract and 1. Introduction Abstract and 1. Introduction 2 Related works 2 Related works 3 Methodology and 3.1 Causal language model as a classification model 3 Methodology and 3.1 Causal language model as a classification model 3.2 Functional token 3.2 Functional token 3.3 Dataset collection 3.3 Dataset collection 3.4 Model development and training 3.4 Model development and training 4 Experiments and 4.1 Android function calls 4 Experiments and 4.1 Android function calls 4.2 Extension to Vehicle, Yelp, and DoorDash function sets 4.2 Extension to Vehicle, Yelp, and DoorDash function sets 4.3 Full and partial training datasets and 4.4 Full training and LoRA training 4.3 Full and partial training datasets and 4.4 Full training and LoRA training 4.5 Parallel and nested function call and 4.6 Weighted loss function for special tokens 4.5 Parallel and nested function call and 4.6 Weighted loss function for special tokens 5 Discussion and future works and References 5 Discussion and future works and References Appendix Appendix A.1 Android function examples A.1 Android function examples A.2 Vehicle function examples A.2 Vehicle function examples 5 Discussion and future works Our current training initiative proves that any specific function can be encapsulated into a newly coined term, functional token, a novel token type seamlessly integrated into both the tokenizer and the model. This model, through a cost-effective training process amounting to merely two cents, facilitates the deployment of AI agents characterized by their remarkably low latency and high accuracy. The potential impacts of our research are extensive. For application developers, including those at DoorDash and Yelp, our model paves the way for training on application-specific scenarios. Developers can pinpoint the APIs most utilized by their audience, transform these into functional tokens for the Octopus model, and proceed with deployment. This strategy has the capacity to fully automate app workflows, emulating functionalities akin to Apple’s Siri, albeit with significantly enhanced response speeds and accuracy. Furthermore, the model’s application within operating systems of PCs, smartphones, and wearable technology presents another exciting avenue. Software developers could train minor LoRAs specific to the operating system. By accumulating multiple LoRAs, the model facilitates efficient function calling across diverse system components. For instance, incorporating this model into the Android ecosystem would enable Yelp and DoorDash developers to train distinct LoRAs, thus rendering the model operational on mobile platforms as well. Looking ahead, we aim to develop a model dedicated to on-device reasoning. Our ambitions are dual-pronged: firstly, to achieve notable speed enhancements for cloud deployments, vastly outpacing GPT-4 in speed metrics. Secondly, to support local deployment, offering a valuable solution for users mindful of privacy or operational costs. This dual deployment strategy not only extends the model’s utility across cloud and local environments, but also caters to user preferences for either speed and efficiency or privacy and cost savings. References [1] Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al. Gpt-4 technical report. arXiv preprint arXiv:2303.08774, 2023. [2] Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. Language models are few-shot learners. Advances in neural information processing systems, 33:1877–1901, 2020. [3] Erik Brynjolfsson, Danielle Li, and Lindsey R Raymond. Generative ai at work. Technical report, National Bureau of Economic Research, 2023. [4] Rich Caruana. Multitask learning. Machine learning, 28:41–75, 1997. [5] Imran Chaudhri. Humane ai, 2024. URL https://humane.com/. Accessed on March 31, 2024. [6] Tim Dettmers, Artidoro Pagnoni, Ari Holtzman, and Luke Zettlemoyer. Qlora: Efficient finetuning of quantized llms. Advances in Neural Information Processing Systems, 36, 2024. [7] Xin Luna Dong, Seungwhan Moon, Yifan Ethan Xu, Kshitiz Malik, and Zhou Yu. Towards next-generation intelligent assistants leveraging llm techniques. In Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pages 5792–5793, 2023. [8] Yu Du, Fangyun Wei, and Hongyang Zhang. Anytool: Self-reflective, hierarchical agents for large-scale api calls. arXiv preprint arXiv:2402.04253, 2024. [9] Div Garg. Multion ai, 2024. URL https://www.multion.ai/. Accessed on March 31, 2024. [10] Gemma Team, Google DeepMind. Gemma: Open models based on gemini research and technology, 2023. URL https://goo.gle/GemmaReport. [11] Allyson I Hauptman, Beau G Schelble, Nathan J McNeese, and Kapil Chalil Madathil. Adapt and overcome: Perceptions of adaptive autonomous agents for human-ai teaming. Computers in Human Behavior, 138: 107451, 2023. [12] Sirui Hong, Xiawu Zheng, Jonathan Chen, Yuheng Cheng, Jinlin Wang, Ceyao Zhang, Zili Wang, Steven Ka Shing Yau, Zijuan Lin, Liyang Zhou, et al. Metagpt: Meta programming for multi-agent collaborative framework. arXiv preprint arXiv:2308.00352, 2023. [13] Yining Hong, Haoyu Zhen, Peihao Chen, Shuhong Zheng, Yilun Du, Zhenfang Chen, and Chuang Gan. 3d-llm: Injecting the 3d world into large language models. Advances in Neural Information Processing Systems, 36:20482–20494, 2023. [14] Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. Lora: Low-rank adaptation of large language models. arXiv preprint arXiv:2106.09685, 2021. [15] Zhengbao Jiang, Frank F Xu, Luyu Gao, Zhiqing Sun, Qian Liu, Jane Dwivedi-Yu, Yiming Yang, Jamie Callan, and Graham Neubig. Active retrieval augmented generation. arXiv preprint arXiv:2305.06983, 2023. [16] Zhanming Jie, Trung Quoc Luong, Xinbo Zhang, Xiaoran Jin, and Hang Li. Design of chain-of-thought in math problem solving, 2023. [17] Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, et al. Retrieval-augmented generation for knowledgeintensive nlp tasks. Advances in Neural Information Processing Systems, 33:9459–9474, 2020. [18] Bo Li, Yuanhan Zhang, Liangyu Chen, Jinghao Wang, Fanyi Pu, Jingkang Yang, Chunyuan Li, and Ziwei Liu. Mimic-it: Multi-modal in-context instruction tuning. arXiv preprint arXiv:2306.05425, 2023. [19] Huayang Li, Yixuan Su, Deng Cai, Yan Wang, and Lemao Liu. A survey on retrieval-augmented text generation. arXiv preprint arXiv:2202.01110, 2022. [20] Yaobo Liang, Chenfei Wu, Ting Song, Wenshan Wu, Yan Xia, Yu Liu, Yang Ou, Shuai Lu, Lei Ji, Shaoguang Mao, et al. Taskmatrix. ai: Completing tasks by connecting foundation models with millions of apis. arXiv preprint arXiv:2303.16434, 2023. [21] Ji Lin, Jiaming Tang, Haotian Tang, Shang Yang, Xingyu Dang, and Song Han. Awq: Activation-aware weight quantization for llm compression and acceleration. arXiv preprint arXiv:2306.00978, 2023. [22] Yi Liu, Gelei Deng, Yuekang Li, Kailong Wang, Tianwei Zhang, Yepang Liu, Haoyu Wang, Yan Zheng, and Yang Liu. Prompt injection attack against llm-integrated applications. arXiv preprint arXiv:2306.05499, 2023. [23] Zechun Liu, Changsheng Zhao, Forrest Iandola, Chen Lai, Yuandong Tian, Igor Fedorov, Yunyang Xiong, Ernie Chang, Yangyang Shi, Raghuraman Krishnamoorthi, et al. Mobilellm: Optimizing sub-billion parameter language models for on-device use cases. arXiv preprint arXiv:2402.14905, 2024. [24] llama.cpp team. llama-cpp. Software available at https://github.com/ggerganov/llama.cpp, 2023. Accessed on March 31, 2024. [25] David Luan. Adept ai, 2024. URL https://www.adept.ai/. Accessed on March 31, 2024. [26] Jesse Lyu. Rabbit r1, 2024. URL https://www.rabbit.tech/. Accessed on March 31, 2024. [27] Yuning Mao, Pengcheng He, Xiaodong Liu, Yelong Shen, Jianfeng Gao, Jiawei Han, and Weizhu Chen. Generation-augmented retrieval for open-domain question answering. arXiv preprint arXiv:2009.08553, 2020. [28] Tomas Mikolov, Kai Chen, Greg Corrado, and Jeffrey Dean. Efficient estimation of word representations in vector space. arXiv preprint arXiv:1301.3781, 2013. [29] Bhargavi Paranjape, Scott Lundberg, Sameer Singh, Hannaneh Hajishirzi, Luke Zettlemoyer, and Marco Tulio Ribeiro. Art: Automatic multi-step reasoning and tool-use for large language models. arXiv preprint arXiv:2303.09014, 2023. [30] Shishir G Patil, Tianjun Zhang, Xin Wang, and Joseph E Gonzalez. Gorilla: Large language model connected with massive apis. arXiv preprint arXiv:2305.15334, 2023. [31] Nikhil Pinnaparaju, Reshinth Adithyan, Duy Phung, Jonathan Tow, James Baicoianu, and Nathan Cooper. Stable code 3b, 2023. URL https://huggingface.co/stabilityai/stable-code-3b. [32] Yujia Qin, Shihao Liang, Yining Ye, Kunlun Zhu, Lan Yan, Yaxi Lu, Yankai Lin, Xin Cong, Xiangru Tang, Bill Qian, et al. Toolllm: Facilitating large language models to master 16000+ real-world apis. arXiv preprint arXiv:2307.16789, 2023. [33] Alec Radford, Karthik Narasimhan, Tim Salimans, Ilya Sutskever, et al. Improving language understanding by generative pre-training. OpenAI blog, 2018. [34] Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, Ilya Sutskever, et al. Language models are unsupervised multitask learners. OpenAI blog, 1(8):9, 2019. [35] Ori Ram, Yoav Levine, Itay Dalmedigos, Dor Muhlgay, Amnon Shashua, Kevin Leyton-Brown, and Yoav Shoham. In-context retrieval-augmented language models. Transactions of the Association for Computational Linguistics, 11:1316–1331, 2023. [36] Jingqing Ruan, Yihong Chen, Bin Zhang, Zhiwei Xu, Tianpeng Bao, Guoqing Du, Shiwei Shi, Hangyu Mao, Xingyu Zeng, and Rui Zhao. Tptu: Task planning and tool usage of large language model-based ai agents. arXiv preprint arXiv:2308.03427, 2023. [37] Timo Schick, Jane Dwivedi-Yu, Roberto Dessì, Roberta Raileanu, Maria Lomeli, Eric Hambro, Luke Zettlemoyer, Nicola Cancedda, and Thomas Scialom. Toolformer: Language models can teach themselves to use tools. Advances in Neural Information Processing Systems, 36, 2024. [38] Weizhou Shen, Chenliang Li, Hongzhan Chen, Ming Yan, Xiaojun Quan, Hehong Chen, Ji Zhang, and Fei Huang. Small llms are weak tool learners: A multi-llm agent. arXiv preprint arXiv:2401.07324, 2024. [39] Yongliang Shen, Kaitao Song, Xu Tan, Dongsheng Li, Weiming Lu, and Yueting Zhuang. Hugginggpt: Solving ai tasks with chatgpt and its friends in hugging face. Advances in Neural Information Processing Systems, 36, 2024. [40] Noah Shinn, Beck Labash, and Ashwin Gopinath. Reflexion: an autonomous agent with dynamic memory and self-reflection. arXiv preprint arXiv:2303.11366, 2023. [41] Benjamin Spector and Chris Re. Accelerating llm inference with staged speculative decoding. arXiv preprint arXiv:2308.04623, 2023. [42] Venkat Krishna Srinivasan, Zhen Dong, Banghua Zhu, Brian Yu, Damon Mosk-Aoyama, Kurt Keutzer, Jiantao Jiao, and Jian Zhang. Nexusraven: a commercially-permissive language model for function calling. In NeurIPS 2023 Foundation Models for Decision Making Workshop, 2023. [43] Yashar Talebirad and Amirhossein Nadiri. Multi-agent collaboration: Harnessing the power of intelligent llm agents. arXiv preprint arXiv:2306.03314, 2023. [44] Qiaoyu Tang, Ziliang Deng, Hongyu Lin, Xianpei Han, Qiao Liang, and Le Sun. Toolalpaca: Generalized tool learning for language models with 3000 simulated cases. arXiv preprint arXiv:2306.05301, 2023. [45] Gemini Team, Rohan Anil, Sebastian Borgeaud, Yonghui Wu, Jean-Baptiste Alayrac, Jiahui Yu, Radu Soricut, Johan Schalkwyk, Andrew M Dai, Anja Hauth, et al. Gemini: a family of highly capable multimodal models. arXiv preprint arXiv:2312.11805, 2023. [46] MLC team. Mlc-llm. Software available at https://github.com/mlc-ai/mlc-llm, 2023. Accessed on March 31, 2024. [47] Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, et al. Llama 2: Open foundation and fine-tuned chat models. arXiv preprint arXiv:2307.09288, 2023. [48] Trelis. Llama-2-7b-chat-hf-function-calling-v3: A model fine-tuned for function calling. Hugging Face Model Repository, 2023. URL https://huggingface.co/Trelis/ Llama-2-7b-chat-hf-function-calling-v3. Accessed on March 31, 2024. [49] Junyang Wang, Haiyang Xu, Jiabo Ye, Ming Yan, Weizhou Shen, Ji Zhang, Fei Huang, and Jitao Sang. Mobile-agent: Autonomous multi-modal mobile device agent with visual perception. arXiv preprint arXiv:2401.16158, 2024. [50] Lei Wang, Chen Ma, Xueyang Feng, Zeyu Zhang, Hao Yang, Jingsen Zhang, Zhiyuan Chen, Jiakai Tang, Xu Chen, Yankai Lin, et al. A survey on large language model based autonomous agents. arXiv preprint arXiv:2308.11432, 2023. [51] Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, et al. Chain-of-thought prompting elicits reasoning in large language models. Advances in Neural Information Processing Systems, 35:24824–24837, 2022. [52] Hao Wen, Yuanchun Li, Guohong Liu, Shanhui Zhao, Tao Yu, Toby Jia-Jun Li, Shiqi Jiang, Yunhao Liu, Yaqin Zhang, and Yunxin Liu. Empowering llm to use smartphone for intelligent task automation. arXiv preprint arXiv:2308.15272, 2023. [53] Qingyun Wu, Gagan Bansal, Jieyu Zhang, Yiran Wu, Shaokun Zhang, Erkang Zhu, Beibin Li, Li Jiang, Xiaoyun Zhang, and Chi Wang. Autogen: Enabling next-gen llm applications via multi-agent conversation framework. arXiv preprint arXiv:2308.08155, 2023. [54] Zhiheng Xi, Wenxiang Chen, Xin Guo, Wei He, Yiwen Ding, Boyang Hong, Ming Zhang, Junzhe Wang, Senjie Jin, Enyu Zhou, et al. The rise and potential of large language model based agents: A survey. arXiv preprint arXiv:2309.07864, 2023. [55] Daliang Xu, Wangsong Yin, Xin Jin, Ying Zhang, Shiyun Wei, Mengwei Xu, and Xuanzhe Liu. Llmcad: Fast and scalable on-device large language model inference. arXiv preprint arXiv:2309.04255, 2023. [56] Zhao Yang, Jiaxuan Liu, Yucheng Han, Xin Chen, Zebiao Huang, Bin Fu, and Gang Yu. Appagent: Multimodal agents as smartphone users. arXiv preprint arXiv:2312.13771, 2023. [57] Yifan Yao, Jinhao Duan, Kaidi Xu, Yuanfang Cai, Zhibo Sun, and Yue Zhang. A survey on large language model (llm) security and privacy: The good, the bad, and the ugly. High-Confidence Computing, page 100211, 2024. [58] Zhuosheng Zhang, Aston Zhang, Mu Li, and Alex Smola. Automatic chain of thought prompting in large language models. arXiv preprint arXiv:2210.03493, 2022. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. available on arxiv Authors:
(1) Wei Chen, Stanford University, with equal contribution and a corresponding author {weichen6}@stanford.edu;
(2) Zhiyuan Li, Stanford University and a corresponding author {zhiyuan8}@stanford.edu. Authors: Authors: (1) Wei Chen, Stanford University, with equal contribution and a corresponding author {weichen6}@stanford.edu; (2) Zhiyuan Li, Stanford University and a corresponding author {zhiyuan8}@stanford.edu.

The Future of Octopus v2: What Does it Entail?

Table of Links Abstract and 1. Introduction Abstract and 1. Introduction 2 Related works 2 Related works 3 Methodology and 3.1 Causal language model as a classification model 3 Methodology and 3.1 Causal language model as a classification model 3.2 Functional token 3.2 Functional token 3.3 Dataset collection 3.3 Dataset collection 3.4 Model development and training 3.4 Model development and training 4 Experiments and 4.1 Android function calls 4 Experiments and 4.1 Android function calls 4.2 Extension to Vehicle, Yelp, and DoorDash function sets 4.2 Extension to Vehicle, Yelp, and DoorDash function sets 4.3 Full and partial training datasets and 4.4 Full training and LoRA training 4.3 Full and partial training datasets and 4.4 Full training and LoRA training 4.5 Parallel and nested function call and 4.6 Weighted loss function for special tokens 4.5 Parallel and nested function call and 4.6 Weighted loss function for special tokens 5 Discussion and future works and References 5 Discussion and future works and References Appendix Appendix A.1 Android function examples A.1 Android function examples A.2 Vehicle function examples A.2 Vehicle function examples 4.3 Full and partial training datasets The Octopus model demonstrates exceptional performance with 1,000 data points sampled for each API during its training phase. However, for training a new set of functions, cost efficiency becomes a consideration, given the need to generate a training dataset. In our analysis, generating 1,000 data points for a single API incurs a cost of 0.0224 USD, representing the investment required to train an Octopus-0 model for one specific function. By evaluating the Octopus-0, Octopus-2, and Octopus-3 models, we discern that sampling only 100 data points for one API can still achieve an accuracy of 98.095%, as depicted in Figure (4). Therefore, for individuals seeking to train their own Octopus model using our framework, we recommend a dataset size ranging from 100 to 1,000 data points. 4.4 Full training and LoRA training LoRA plays a crucial role in our framework, particularly when integrating the Octopus model across multiple applications to ensure smooth computation. Instead of employing full models for each API set, we opt for diverse LoRA trainings tailored to the specific function setups of different apps. As Figure (4) illustrates, switching to Lora training results in a minor accuracy decrease. Nonetheless, the maintained high accuracy levels are sufficiently robust for production deployment. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. available on arxiv Authors:
(1) Wei Chen, Stanford University, with equal contribution and a corresponding author {weichen6}@stanford.edu;
(2) Zhiyuan Li, Stanford University and a corresponding author {zhiyuan8}@stanford.edu. Authors: Authors: (1) Wei Chen, Stanford University, with equal contribution and a corresponding author {weichen6}@stanford.edu; (2) Zhiyuan Li, Stanford University and a corresponding author {zhiyuan8}@stanford.edu.

Optimizing Function Calling Models: The Role of Dataset Size and LoRA Fine-tuning

Table of Links Abstract and 1. Introduction Abstract and 1. Introduction 2 Related works 2 Related works 3 Methodology and 3.1 Causal language model as a classification model 3 Methodology and 3.1 Causal language model as a classification model 3.2 Functional token 3.2 Functional token 3.3 Dataset collection 3.3 Dataset collection 3.4 Model development and training 3.4 Model development and training 4 Experiments and 4.1 Android function calls 4 Experiments and 4.1 Android function calls 4.2 Extension to Vehicle, Yelp, and DoorDash function sets 4.2 Extension to Vehicle, Yelp, and DoorDash function sets 4.3 Full and partial training datasets and 4.4 Full training and LoRA training 4.3 Full and partial training datasets and 4.4 Full training and LoRA training 4.5 Parallel and nested function call and 4.6 Weighted loss function for special tokens 4.5 Parallel and nested function call and 4.6 Weighted loss function for special tokens 5 Discussion and future works and References 5 Discussion and future works and References Appendix Appendix A.1 Android function examples A.1 Android function examples A.2 Vehicle function examples A.2 Vehicle function examples 4 Experiments Our study conducts a comprehensive evaluation of language model capabilities via an extensive benchmarking approach, aimed at assessing their effectiveness in generating accurate function calls. Initially, we compare our model’s accuracy and response time against premier models in the field, namely GPT-4 (checkpoint: gpt-4-0125-preview) and GPT-3.5 (checkpoint: gpt-3.5-turbo-0125). In the next phase, we explore the efficacy of the RAG technique, renowned for its ability to reduce incorrect outputs (hallucinations) and latency by equipping language models with a concise selection of potential functions. Through the integration of Meta’s FAISS for semantic search, we enhance the function call description retrieval process, opting for the top 5 descriptions to navigate context length constraints seen in models like Meta’s Llama-7B and OpenAI’s GPT-3.5. Subsequently, we analyze the impact of training dataset size and model training methods on performance metrics. 4.1 Android function calls To illustrate our model’s application, we select Android system function calls as a case study, focusing on accuracy and latency in function call generation. Initially, we chose 20 Android APIs as our dataset foundation. We adopted two distinct methods for generating function call commands. For details on API design, see Appendix. The first method involves a RAG approach to identify top similar function descriptions based on user queries, which the language model then uses, along with the user query, to generate the expected function call commands. We detail the various models employed during this evaluation. Utilizing Google Gemini, we sample relevant queries for the selected Android function calls and manually label the ground truth as the evaluation dataset. We document our benchmark results, focusing on two critical metrics: accuracy and latency, as illustrated in Figure (4) and Figure (5) respectively. Llama-7B RAG Evaluation Initially, the pretrained Llama-7B model showed limited ability in generating the expected outcomes, leading us to employ a Llama-7B variant fine-tuned for function call generation [48]. For the Llama-7B assessment, we applied the RAG method without strict output format requirements, considering responses with missing parentheses as correct. This evaluation was conducted on a single NVIDIA A100 machine, with all results from Llama-7B compared against the ground truth. The primary errors were incorrect function name selection and erroneous parameter generation. Despite employing few-shot learning to guide the model towards accurate function generation, the performance was modest, with an accuracy rate of 68.095% when overlooking format requirements and a latency of 13.46 seconds, excluding model loading time. To improve latency, we implemented optimizations such as flash attention and a fast tokenizer. Llama-7B RAG Evaluation GPT-3.5 RAG Evaluation Similar to the approach with Llama-7B, we utilized GPT-3.5 for response generation, employing the same semantic search strategy for context acquisition. To enhance GPT-3.5’s performance, we designed a specific prompt style, incorporating one-shot learning to improve accuracy further. GPT-3.5 RAG Evaluation In this benchmark test, an impressive accuracy of 98.095% was achieved, leveraging the gpt-3.5-turbo0125 checkpoint known for its optimization in function calling tasks. The latency was significantly improved to 1.97 seconds for generating a single function call, a notable enhancement over the Llama-7B model’s performance. This improvement in speed is primarily attributed to the efficiency of the language model inference, as the RAG component remained consistent. GPT-3.5’s quicker response may be due to OpenAI’s use of multiple GPUs or a more advanced inference infrastructure. Further analysis revealed that a significant portion of the time was spent on content retrieval, despite only needing to fetch 5 function descriptions from a pool of 20. To optimize latency, all function descriptions’ embeddings were precomputed using OpenAI’s text-embedding-3-large endpoint, with IndexFlatL2 employed for search indexing and parallel computation on multicore CPUs to enhance speed. GPT-3.5 and GPT-4 Evaluations In efforts to further reduce latency for GPT-3.5 and GPT-4, we included all 20 function descriptions directly in the context, bypassing the RAG method to avoid microservices interactions and their associated IO-bound overheads. This adjustment yielded a latency reduction to 1.18 seconds for GPT-3.5. The prompt template mirrored that of the GPT-3.5 RAG, with the addition of more candidate functions. However, accuracy slightly declined to 97.143%, possibly due to diminished language model effectiveness with longer text inputs. Conversely, GPT-4 exhibited superior accuracy at 98.571% and even lower latency than GPT-3.5, despite expectations of GPT-4 being a larger model. This performance, evaluated on March 18 at 2 PM PDT, might reflect variances in API traffic or hardware configurations between the two models. GPT-4’s enhanced performance suggests OpenAI could be allocating more GPU resources to it or that it experiences less demand compared to GPT-3.5. GPT-3.5 and GPT-4 Evaluations Octopus model Now, we present the octopus model with 1000 data points sampled for each API. And we observe a 99.524% accuracy in our evaluation dataset. Moreover, the prompt used for this method is as simple as: Octopus model In our approach, incorporating function information directly into the context is unnecessary, as the Octopus model has already learned to mapping functional tokens to corresponding function descriptions, thereby conserving a significant number of tokens for processing. Given its compact size and the brevity of the context required, the Octopus model demonstrates a reduced latency of 0.38 seconds. To maintain an equitable comparison, we adhered to the same benchmark settings used for the Llama7B evaluation, such as incorporating flash attention and not using quantization. Furthermore, we explored the deployment of our Octopus 2B model on mobile devices through quantization. By precomputing the state for the fixed prefix—"Below is the query from the users, please call the correct function and generate the parameters to call the function. Query:"—our on-device model achieves remarkable performance, completing a function call within 1.1 to 1.7 seconds for typical queries of 20 to 30 tokens using a standard Android phone. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. available on arxiv Authors:
(1) Wei Chen, Stanford University, with equal contribution and a corresponding author {weichen6}@stanford.edu;
(2) Zhiyuan Li, Stanford University and a corresponding author {zhiyuan8}@stanford.edu. Authors: Authors: (1) Wei Chen, Stanford University, with equal contribution and a corresponding author {weichen6}@stanford.edu; (2) Zhiyuan Li, Stanford University and a corresponding author {zhiyuan8}@stanford.edu.

Stories, News & Mentions about DOORDASH

Categories

Trending Topics

Stories, News & Mentions about DOORDASH

Doordash’s Stories on HackerNoon

Doordash's News & Mentions

Categories

Trending Topics