sigma-cls / README.md
paramitopia's picture
Upload 14 files
a3afcc0 verified
metadata
base_model: sentence-transformers/all-roberta-large-v1
library_name: setfit
metrics:
  - accuracy
pipeline_tag: text-classification
tags:
  - setfit
  - sentence-transformers
  - text-classification
  - generated_from_setfit_trainer
widget:
  - text: >-
      USER: Please rewrite this poem in paragraph form (One paragraph is fine).
      Keep the tone the same; since it is no longer a poem, make sure it is not
      rhyming or in the rhythm of a poem.


      At the Zoo

      First I saw the white bear, then I saw the black;

      Then I saw the camel with a hump upon his back;

      Then I saw the grey wolf, with mutton in his maw;

      Then I saw the wombat waddle in the straw;

      Then I saw the elephant a-waving of his trunk;

      Then I saw the monkeys—mercy, how unpleasantly they smelt!


      https://poets.org/poem/zoo


      ASSISTANT: Please see the rewritten poem "At the Zoo" below, in a
      paragraph form.


      When we went to the zoo, I saw a bunch of animals. I saw many animals with
      different colors and attributes. 

      In the beginning, I saw both a white and a black bear. Afterward, there
      was a camel that had a hump. The next group of animals was a grey wolf, a
      wombat hanging out in the straw, and an elephant waving his trunk around.
      The last animals that I saw were the stinky smelly monkeys!
  - text: |-
      USER: Please write a short poem about ducks.

      ASSISTANT: Ducks one of the luckiest birds.
      They come in all shapes and love all grapes 
      They are better than pups when it comes to deluxe fluff
      Ducks amazing at flying and diving
      Ducks on of the luckiest birds.
  - text: >-
      USER: Tell me how to get more Instagram followers please, as I want to
      become Instagram famous! Ten tips would be ideal. I deleted everything
      from my feed, so you can write it as if I'm starting from scratch! 


      ASSISTANT: Good luck! Here are ten ways to get more Instagram followers:


      1. Create an Aesthetic Instagram Feed: Instagram is all about aesthetics,
      so having a strong aesthetic on your Instagram feed is key to attracting
      more followers. The first thing people see when they look at your page is
      your collection of photos, so having pictures that blend well together
      will go down nicely with potential new followers.


      2. Engage With Followers: Don't forget to engage with your current
      followers to build your reputation and make sure people will recommend
      you. You can do this by liking and responding to comments, as well as
      following people back who seem to have similar interests or content to
      you. 


      3. Optimize Your Instagram Bio: Make sure your Instagram bio makes it
      clear what your Instagram is about, and focus on a niche if possible. For
      example, your Instagram might revolve around your pets and your make-up,
      or it might be more about travel. You can post a variety of things, but
      focusing primarily on one can be extremely helpful in building your
      follower base and drawing in people with similar interests.


      4. Find the Best Hashtags for Your Followers: Hashtags are key to bringing
      an audience who might never have come across your feed otherwise. Look up
      what hashtags are trending that relate to your niche and make sure you use
      them in your post. Don't be tempted to use irrelevant hashtags just
      because they're trending though, or people may be turned away from your
      page.


      5. Create a Consistent Instagram Posting Schedule: No one wants to be
      spammed with content all day every day, but they also don't want to be
      left without it. Creating a consistent posting schedule, so people know
      when to expect new content and how much they're going to get it when they
      follow you, can be extremely helpful in attracting the right audience who
      will enjoy what you post.


      6. Make Sure Your Instagram Content is Accessible: Accessibility is key
      for online spaces. For example, you could include descriptions of the
      picture in each caption so that blind audiences can also enjoy your
      content. Make sure to avoid any videos with flashing lights, and avoid
      posting triggering content without a warning. Caring about your followers
      will go a long way in building brand loyalty in an audience.


      7. Work With Other Brands, Creators, and Influencers: You can also work
      with other Instagram brands and creators, as this will help draw people
      into your page from theirs and you can offer them the same thing. Brands
      tend to reach out to larger pages to offer sponsorships, but there's no
      reason you can't reach out to them first and see what you can build
      together.


      8. Start Creating Instagram Reels: Although Instagram has always been
      known as a picture platform, it's starting to become popular in the video
      space. Making reels with trending audio can help draw people in that might
      not have seen your page otherwise.


      9. Create Memes and Other Shareable Content: Making content that's
      shareable is another sure way to bring in audiences. For example, posting
      memes that people might relate to means they might share it on their
      story, exposing your Instagram to their followers. Make sure the memes are
      authentic to your page and the message you're trying to send.


      10. Run an Instagram Giveaway: You could also run an Instagram giveaway,
      mandating that people must comment on your post and follow your page to
      enter. Some giveaways also dictate that entrants must share the post,
      exposing the giveaway—and therefore, the account—to even more people.


      Building a following on Instagram is all about being authentic and true to
      yourself while creating content that your followers will love.
  - text: >-
      USER: Write a haiku about a kid who hates doing homework, but doesn't have
      a choice. Make it funny but keep it to the point. Keep it vivid even
      though it's a haiku.


      ASSISTANT: My pen-covered hands

      slave away at my homework

      enslaved until done.
  - text: >-
      USER: I would like a quick description of the basic creation of a
      statistical learning model. This should be simple enough that the average
      person could understand it. The description should be a paragraph of 5
      sentences.


      ASSISTANT: When we look to create a model, we are attempting to predict a
      single feature from all the other features of something. This starts by
      looking at a set of instances that all have data points for the same
      features. A key assumption is that these features all have some
      distribution even if we know nothing about it. We then run them through a
      processing component in order to create that prediction component, after
      assuming there is a completely correct version of this function. If we do
      not have complete data or knowledge of the distributions, the prediction
      function can rarely be completely perfect.
inference: true
model-index:
  - name: SetFit with sentence-transformers/all-roberta-large-v1
    results:
      - task:
          type: text-classification
          name: Text Classification
        dataset:
          name: Unknown
          type: unknown
          split: test
        metrics:
          - type: accuracy
            value: 0.9282511210762332
            name: Accuracy

SetFit with sentence-transformers/all-roberta-large-v1

This is a SetFit model that can be used for Text Classification. This SetFit model uses sentence-transformers/all-roberta-large-v1 as the Sentence Transformer embedding model. A LogisticRegression instance is used for classification.

The model has been trained using an efficient few-shot learning technique that involves:

  1. Fine-tuning a Sentence Transformer with contrastive learning.
  2. Training a classification head with features from the fine-tuned Sentence Transformer.

Model Details

Model Description

Model Sources

Model Labels

Label Examples
Brainstorm
  • "USER: Help me come up with some properties of an alien metal that would mark it as particularly strange in a sci-fi mystery novel. Explain how they might attract unwanted attention as well.\n\nASSISTANT: Here are some properties of your sci-fi metal that might garner unwanted attention for those who possess it. \n\n1. The metal levitates: If the metal does not fall to the ground when let go, that would be an obvious sign of its strangeness. The military or other government agencies might wish to procure it for this property's usefulness in aircraft and spacecraft.\n2. The metal acts as a battery: Perhaps the metal carries a usable charge at all times. This would be very strange and may spark interest from the larger scientific community, electronics companies, or powerful energy concerns.\n3. The metal causes local gravitational disturbances: It would be highly unusual if the metal caused small objects to orbit around it when placed inside a room. This phenomenon could be interpreted religiously by a local cult or religious movement, causing unwanted attention.\n4. The metal affects living creatures around it: Perhaps the metal attracts the interest of local wildlife. Wherever the metal goes deer follow as if in a trance. This could present difficulties as the deer may obstruct traffic or even become aggressive when the metal is kept away from it.\n5. The metal causes hallucinations: Sleeping in the same room as the metal causes one to hallucinate strange visions of other worlds and dimensions. This could cause unwanted attention from the beings seen within the dreams."
  • 'USER: What are some fun outdoor activities for two kids, ages 10 and 11 to do at home in a small backyard? They like to be physically active, play in dirt, and with bugs. The ideas should be creative and engaging but do not require too much adult guidance.\n\nASSISTANT: Here are a few ideas for fun outdoor activities in a small backyard:\n\n1. Plant a small garden: If the kids already enjoy playing in the dirt, set up a small garden bed and teach them how to plant flowers, herbs, or even vegetables in a small garden bed. They can also learn about the different plants and how to care for them.\n\n2. Scavenger hunt: Create a list of treasures, then hide them along with clues in the backyard. Let the kids search for hidden treasure while exploring the outdoors.\n\n3. Bug exploration: Encourage the kids to look for bugs in the backyard and collect them in jars to examine them up close. Ask questions about what they find to help identify the different bugs.\n\n4. Nature-themed art: Use natural materials in the backyard, like leaves, flowers, and rocks, to create artwork on posterboard or cardboard. Paint larger rocks with acrylic paint to display in the garden.\n\n5. Water play: Set up an area with a kiddie pool, bowl, cups, or other utensils to let the kids play with water. Play Sink or Float in the water by guessing whether an object will sink or float before putting it in the pool.'
  • 'USER: I am looking for a Mario game to play on the Nintendo 3DS. What are my options for different games? Please include titles and descriptions.\n\nASSISTANT: Sure! Here is a list of Mario games for the 3DS.\n\n• Mario Kart 7 is a racing game where you can race against other Mario characters in a variety of courses.\n\n• Super Mario 3D Land is a platformer where you can explore a variety of 3D worlds like Mario.\n\n• Mario & Luigi: Dream Team is an RPG game where you control Mario and Luigi as they travel through a dream world.\n\n• Mario & Luigi: Paper Jam is an RPG where you control Mario, Luigi, and Paper Mario as they travel through the Mushroom Kingdom filled with paper.\n\n• Mario Golf: World Tour is a golf game where you can play against other Mario characters in a variety of courses.'
Rewrite
  • 'USER: Rewrite the poem into questions and answers about the poem so that a second grader could answer them.\n\n“At the Zoo” by William Makepeace Thackeray\n\nFirst I saw the white bear, then I saw the black;\nThen I saw the camel with a hump upon his back;\nThen I saw the grey wolf, with mutton in his maw;\nThen I saw the wombat waddle in the straw;\nThen I saw the elephant a-waving of his trunk;\nThen I saw the monkeys—mercy, how unpleasantly they smelt!\n\nASSISTANT: 1. What did the narrator see first?\nAnswer: First the narrator saw the white bear.\n2. After the white bear, what did the narrator see next?\nAnswer: The narrator saw the black bear next.\n3. What animal from the poem had a hump on it’s back?\nAnswer: The camel with a hump on it’s back\n4. What animal had mutton in his maw? \nAnswer: The grey wolf.\n5. What did the wombat do?\nAnswer: The wombat waddled in the straw.\n6. What was the elephant waving?\nAnswer: The elephant a-waving of his trunk.\n7. What was wrong with the monkeys?\nAnswer: How unpleasantly they smelt!'
  • "USER: Rewrite this advertisement from the perspective of a cynical Gen-Xer who loathes capitalism.\n\nNeed some new furniture? Now's the time to shop.\n\nAhead of Memorial Day weekend, we're searching for all of the best furniture sales to help you fill your home with all the items you've been eyeing this year.\n\nFor example, shop up to 70% off at Wayfair during their Memorial Day clearance sale, going on now. And Crate & Barrel has deals on bestselling sofas and outdoor furniture, among other offerings.\n\nContinue below to start shopping the deals and be sure to check back to shop more sales from your favorite retailers.\n\nASSISTANT: Need some new furniture? Good, because that's what they want you to need.\n\nAhead of Memorial Day weekend, they're searching for the best way to get you to blow all your hard earned money and fill your home with all the items you don't need this year.\n\nFor example, check out the 70% off at sale Wayfair during their Memorial Day clearance sale, it's suspicious. And Crate & Barrel has deals on bestselling sofas and outdoor furniture, among other ways to brainwash you into living the corporate American dream.\n\nDon't continue below to start shopping the deals and be sure not to check back to shop more sales from your favorite retailers (a.k.a. The Man)."
  • "USER: Rewrite this job application email to make it more persuasive and compelling.\n\nDear (Recipient's name),\n\nPlease find my application for the (job name) post. I’ve attached (a CV, resume, cover letter, etc.)\n\nI’m excited to apply for the position. I’ve reviewed the job specification and researched your company, and I believe I have the skills, experience, and approach to perform in the role.\n\nYou can find details of my qualifications and achievements in my CV. If you have any questions, please get in touch with me (insert details).\n\nPlease provide me with details on the next steps in the process.\n\nBest,\n(Your name)\n\nASSISTANT: Dear (Recipient's name),\nI am writing to express my strong interest in the (Job Name) position at your company. I have attached my CV, which pinpoints my qualifications and makes me a great fit for this role.\nMy skills, experience, and innovative approach align well with the job requirements. After researching your company, I am impressed by your dedication to [specific aspect or achievement of the company]. Joining your team and contributing to your success would be a great opportunity.\nI am available to discuss my qualifications and answer any questions. Please feel free to contact me at (insert details).\nI am excited to move forward in the application process. Please provide me with details on the next steps. \nThank you for considering my application. I look forward to bringing my skills and enthusiasm to your organization.\nBest regards,\n(Your Name)"
Coding
  • "USER: Write me a function that will import the fourclass dataset and calculate the logistic regression using Python.\n\nASSISTANT: You can use sklean to calculate logistic regression with Python. Here's how you do it:\n\n# Import the necessary libraries\nfrom sklearn import datasets\nfrom sklearn import linear_model\nfrom sklearn.datasets import load_fourclass\nfrom sklearn.linear_model import LogisticRegression\n\n# Load the dataset\nX, y = load_fourclass(return_X_y = True)\n# Fit the data to the regression curve\nreg = LogisticRegression(random_state = 0).fit(X, y)\n# Predict the regression with a classification\nlog = clf.predict(X[:2, :])\nprint(log)\n\nThis code imports datasets, linear_model, load_fourclass, and LogisticRegression from the the sklearn library to allow you to calculate the logistic regression of the data. It first uses load_fourclass with the return_X_y parameter set to true in order to define the X and y variables. These variables are then fit to the logistic regression, reg, using LogisticRegression().fit(). The random_state is set to 10, which will allow you to reproduce the data if need be. Without setting the random state, the shuffle of the data will be different each time. Finally, clf.predict() is used to predict the classification of the data as log, which is then printed using print()."
  • 'USER: I've been learning interprocess communication. My last project used tempfiles to send data and commands back and forth. Now I want to try using sockets. Can you show me how to make a basic client/server architecture in python using sockets?\n\nASSISTANT: Certainly! We can use Python's socket module to communicate between two processes. Here's an example implementation. \n\nimport socket\n# server.py\n\n# create server socket\nserver_socket = socket.socket(socket.AF_INET, socket.SOCK_STREAM)\n\n# bind the socket to localhost:8000\nserver_address = ('localhost', 8000)\nserver_socket.bind(server_address)\n\n# listen for connections\nserver_socket.listen(1)\nprint(f"Listening on {server_address}")\nclient_socket, client_address = server_socket.accept()\nprint(f"{client_address} connected")\n\n# accept data from client\ndata = client_socket.recv(1024)\nprint(f" {data}")\n\n# echo message back to client\nclient_socket.sendall(f"Message received: {data}".encode('utf-8'))\n\n# close connection\nclient_socket.close()\n\n...\n\nimport socket\n# client.py\n\n# create client socket\nclient_socket = socket.socket(socket.AF_INET, socket.SOCK_STREAM)\n\n# connect the socket to the server\nserver_address = ('localhost', 8000)\nclient_socket.connect(server_address)\nprint(f"Connected to {server_address}")\n\n# say hello to the server\nmessage = "Hello, server!"\nclient_socket.sendall(message.encode())\n\n# accept response from server\ndata = client_socket.recv(1024)\nprint(f" {data.decode()}")\n\n# close connection\nclient_socket.close()'
  • "USER: Write a Python function to calculate the sample standard deviation of an array of numbers. (Don't use any prewritten standard deviation functions).\n\nASSISTANT: Here's a function to calculate the sample standard deviation, written from scratch:\n\ndef stdev(nums):\n mean = sum(nums) / len(nums)\n diffs = [(num - mean)**2 for num in nums]\n return math.sqrt(sum(diffs) / (len(nums) - 1))\n\n"
Summarize
  • "USER: Tell me in a sentence or two what this text says about Nirvana.\n\nNirvana was an American rock band formed in Aberdeen, Washington, in 1987. Founded by lead singer and guitarist Kurt Cobain and bassist Krist Novoselic, the band went through a succession of drummers, most notably Chad Channing, and then recruited Dave Grohl in 1990. Nirvana's success popularized alternative rock, and they were often referenced as the figurehead band of Generation X. Their music maintains a popular following and continues to influence modern rock culture.\n\nASSISTANT: Nirvana was a Washington-based rock band formed in 1987 by Kurt Cobain and Krist Novoselic, with Chad Channing and Dave Grohl being the most notable drummers. They are often cited as an influencer of modern rock culture."
  • "USER: Could you give me a summary of this?\n\nThis is what I want summarized:\n\nCompleting his resemblance to a man who was sitting for his portrait, Mr. Lorry dropped off to sleep. The arrival of his breakfast roused him, and he said to the drawer, as he moved his chair to it:\n\n“I wish accommodation prepared for a young lady who may come here at any time to-day. She may ask for Mr. Jarvis Lorry, or she may only ask for a gentleman from Tellson’s Bank. Please to let me know.”\n\n“Yes, sir. Tellson’s Bank in London, sir?”\n\n“Yes.”\n\n“Yes, sir. We have oftentimes the honour to entertain your gentlemen in their travelling backwards and forwards betwixt London and Paris, sir. A vast deal of travelling, sir, in Tellson and Company’s House.”\n\n“Yes. We are quite a French House, as well as an English one.”\n\n“Yes, sir. Not much in the habit of such travelling yourself, I think, sir?”\n\n“Not of late years. It is fifteen years since we—since I—came last from France.”\n\n“Indeed, sir? That was before my time here, sir. Before our people’s time here, sir. The George was in other hands at that time, sir.”\n\n“I believe so.”\n\n“But I would hold a pretty wager, sir, that a House like Tellson and Company was flourishing, a matter of fifty, not to speak of fifteen years ago?”\n\n“You might treble that, and say a hundred and fifty, yet not be far from the truth.”\n\n“Indeed, sir!”\n\nRounding his mouth and both his eyes, as he stepped backward from the table, the waiter shifted his napkin from his right arm to his left, dropped into a comfortable attitude, and stood surveying the guest while he ate and drank, as from an observatory or watchtower. According to the immemorial usage of waiters in all ages.\n\nASSISTANT: Mr. Lorry's breakfast arrived, and he asked that the hotel make accommodations for a young woman who would arrive today and that he be told when she arrived. He and the waiter then talked briefly about the success and history of Tellson and Company."
  • 'USER: Summarize what this article says about playing games with Chromecast. \n\nUse this information to respond: \n\nChromecast can be used to play games\nMany Android users probably know that they can mirror the screen of their Android phone or tablet to a Chromecast device to share what they’re doing on the big screen. While screen mirroring performance might not be good enough to play a lot of regular games, there are actually a number that are specifically designed with Chromecast in mind.\n\nThere are a number of multiplayer games that work with Chromecast designed specifically to take advantage of the big screen for party gaming. DrawParty and PictoParty are digital takes on the classic party game Pictionary. Puzzle Cast allows you to put together a jigsaw puzzle without all the mess and missing pieces.\n\nIf you’re partial to word games, Wordcast is a big-screen multiplayer equivalent of the classic Scrabble. Or for something different, you can try Scrabble Blitz, which challenges you to make up words as quickly as possible. Whatever your taste is, you should be able to find something to play during your next friendly get together.\n\nASSISTANT: This article states that while you can play games on your television from your Android phone via Chromecast, the screen mirroring performance is not good enough to play all games. \n\nThe games you can play via Chromecast include DrawParty, PictoParty, PuzzleCast, Wordcast, and Scrabble Blitz.'
Closed QA
  • 'USER: When does Starbucks say Spring begins? \n\nStarbucks Coffee\n@Starbucks\nEmotionally, spring starts the first time you can drink iced coffee outside without gloves on.\n8:07 AM · Mar 14, 2023\n·\n389.5K\n Views\n\nASSISTANT: According to Starbucks, spring begins the first time you drink iced coffee outside without wearing gloves. '
  • 'USER: Extract the answer to these questions: What year was Philip Thicknesse born, what year did he die and who did he have his severed hand sent to? \n\nDespite having a name that somehow sounds obscene, Philip Thicknesse (1719-1792) was not a happy man. He spent his younger years traveling the world, getting up to various racist colonial goings-on, before settling down in England, where he published a sketchy book on healthy living that recommended as much inhalation of “the breath of young women” as possible. \n\nBut he found normal life on land didn’t work for him, and he ended up living in the grounds of his house as a hermit, digging up Saxon coffins and making up stories about the bodies he found. When Thicknesse died, he arranged for his severed hand to be delivered to his estranged son George. This, he said in an accompanying note, was to “remind him of his duty to God after having so long abandoned the duty he owed to a father, who once so affectionately loved him.” \n\nA combination of a vile guilt-trip and a subtle “I did not love you anymore” message, all in one. Petty!\n\nASSISTANT: Philip Thicknesse was born in 1719. \nHe died in 1792. \nHe had his severed hand sent to his son, George. '
  • 'USER: Who is Kevin Plank? Where did he go to college?\n\nUnder Armour was founded on September 25, 1996, by Kevin Plank, a then-24 year old former special teams captain of the University of Maryland football team. Plank initially began the business from his grandmother's basement in Washington, D.C.[4] He spent his time traveling along the East Coast with nothing but apparel in the trunk of his car. His first team sale came at the end of 1996 with a $17,000 sale. From his grandmother's Washington, D.C.-area rowhouse, Plank moved to Baltimore.[5]\n\nAs a fullback at the University of Maryland, Plank got tired of having to change out of the sweat-soaked T-shirts worn under his jersey; however, he noticed that his compression shorts worn during practice stayed dry. This inspired him to make a T-shirt using moisture-wicking synthetic fabric.[5] After graduating from the University of Maryland, Plank developed his first prototype of the shirt, which he gave to his Maryland teammates and friends who had gone on to play in the NFL. Plank soon improved the design, creating a new T-shirt built from microfibers that wicked moisture away.[4] Major competing brands including Nike, Adidas and Reebok would soon follow with their own moisture-wicking apparel.[6] Plank opted to use the British spelling "armour" in the company name because the toll-free vanity number was still available for that version.[7]\n\nASSISTANT: Kevin Plank is the founder of Under Armour. He went to college at the University of Maryland.'
Open QA
  • 'USER: What does it mean to braise meat?\n\nASSISTANT: Braising is a cooking method used to make a tough cut of meat more tender. The process involves searing the meat in hot oil and then simmering it in liquid until tender. Here are some steps to successfully braising meat:\n1. Heat a heavy pot or dutch oven with oil and sear the meat until golden brown on all sides.\n2. Remove the meat and deglaze the pot by pouring in liquid (such as broth or water) and stirring around to loosen the flavorful stuck-on bits.\n3. Return the meat to the pot and add enough liquid to cover the meat at least halfway and cover it with a lid. You may wish to add seasonings as well. \n4. Cook the meat slowly until fork-tender over low heat. '
  • 'USER: My friend and I are having a debate about who the 23rd president was.\n\nASSISTANT: The 23rd President of the United States was Benjamin Harrison. He served as president from 1889-1893 and was one of the first nominees to deliver speeches from his home in Indianapolis. This type of campaigning became known as "front-porch" campaigning. '
  • 'USER: Can you tell me when the Library of Congress was founded?\n\nASSISTANT: The Library of Congress was founded on April 24, 1800, in Washington D.C. It was established thanks to an act of Congress and was originally housed in the U.S. Capitol. '
Math
  • 'USER: Peter is buying a set of glasses. They cost $3 for small glasses and $5 for large ones. He has $50. If he buys 8 small ones and leaves with $1 in change, how many large ones did he buy?\n\nASSISTANT: He spent $49 on glasses because 50-49=<<50-49=1>>1\nHe spent $24 on small glasses because 8x3=<<8*3=24>>24\nHe spent $25 on large glasses because 49-24=<<49-24=25>>25\nHe bought 5 large glasses because 25/5=<<25/5=5>>5\n#### 5'
  • "USER: Leo's assignment was divided into three parts. He finished the first part of his assignment in 25 minutes. It took him twice as long to finish the second part. If he was able to finish his assignment in 2 hours, how many minutes did Leo finish the third part of the assignment?\n\nASSISTANT: It took Leo 25 x 2 = <<25*2=50>>50 minutes to finish the second part of the assignment.\nLeo finished the first and second parts of the assignment in 25 + 50 = <<25+50=75>>75 minutes.\nHe finished the entire assignment in 60 x 2 = <<60*2=120>>120 minutes.\nTherefore, it took Leo 120 - 75 = <<120-75=45>>45 minutes to finish the third part of the assignment.\n#### 45"
  • 'USER: Mara has 40 crayons and 10 percent of her crayons are pink. Luna has 50 crayons and 20 percent of them are pink. In total how many pink crayons do Mara and Luna have?\n\nASSISTANT: Mara has 40 x (10 / 100) = <<40*(10/100)=4>>4 pink crayons.\nLuna has 50 x (20 / 100) = <<50*(20/100)=10>>10 pink crayons.\nTogether they have 4 + 10 = <<4+10=14>>14 pink crayons\n#### 14'
Generation
  • "USER: Write me 4 haikus about new year's day.\n\nASSISTANT: 12 months have ended\nAnother year has passed by\nCelebration time\n\nResolutions made\nThings I need to do next year\nPlans and fitness goals\n\nPops fizzes and bangs\nExplosions in the sky make\nColorful displays\n\nEnd-of-year party\nAud Lang Syne sung very loud\nShout 'happy new year!'"
  • 'USER: Write 3 sentences for an about me section on a website. It should include how I am a great leader and that I graduated summa cum laude.\n\nASSISTANT: I am an industry veteran with proven leadership skills and a track record of success. Following my college graduation where I earned summa cum laude accolades, I entered into the workforce and have not looked back. Let me put my expertise to work for you.'
  • 'USER: Can you please write me a tutorial of how to make a paper airplane that will definitely fly? Step by step and numbered instructions would be great. Thanks so much.\n\nASSISTANT: 1. Take a piece of printer paper and fold it in half the long way. \n2. Unfold it and then take both corners at the top and fold them into the center line. \n3. Fold the top point down almost an inch from the bottom edge of the paper. \n4. Then, take both of the top corners and, again, fold them into the center line. \n5. Now, take the top point and fold it over the two bits in the center and lock everything in its place. \n6. Fold the entire plane in half towards the top. \n7. Take the wings and fold them down about 1 inch from the bottom edge. \n8. Take the sides and fold up the bits on the end so they stand about half an inch tall. \n9. Grab scissors and cut two small slits on the back of each wing. \n10. Fold up the tabs you created with the slits. \n11. If you want, you can add double-stick tape to the inside for added stability. '
Classify
  • "USER: Read this Facebook post, classify the tone and explain why:\n\nI know I'm going to fail this test because I always fail at everything I do! (Crying emoji inserted)\n\n\nASSISTANT: The tone is pessimistic because the author believes the worst outcome is going to happen. The author says they always fail at everything they do. This is an example of a pessimistic tone."
  • 'USER: From this description, what is the tone of this podcast episode? Humorous, whimsical, nasty, or analytical.\n\n"Museums are purging their collections of looted treasures. Can they also get something in return? And what does it mean to be a museum in the 21st century?"\n\nASSISTANT: Analytical'
  • 'USER: Which category would this poem best fall into? Romantic, Sad, Humorous, Reflective\n\nDaddy Fell into the Pond\n\nBy Alfred Noyes\n\nEveryone grumbled. The sky was grey.\nWe had nothing to do and nothing to say.\nWe were nearing the end of a dismal day,\nAnd then there seemed to be nothing beyond,\nThen Daddy fell into the pond!\n\nAnd everyone’s face grew merry and bright,\nAnd Timothy danced for sheer delight.\n“Give me the camera, quick, oh quick!\nHe’s crawling out of the duckweed!” Click!\n\nThen the gardener suddenly slapped his knee,\nAnd doubled up, shaking silently,\nAnd the ducks all quacked as if they were daft,\nAnd it sounded as if the old drake laughed.\nOh, there wasn’t a thing that didn’t respond\nWhen Daddy Fell into the pond!\n\nASSISTANT: Humorous.'
Extract
  • 'USER: Extract all sentences that have "copyright" as their first or second word. Put in a numbered list. \n\nCopyright is a law that gives the owner of a work (for example, a book, movie, picture, song or website) the right to say how other people can use it. Copyright laws make it easier for authors to make money by selling their works. It is one part of a group of laws about intellectual property (the others being trademark and patent law). It helps protect authors from other people copying their works without permission and/or for commercial purposes.\n\nWith copyright, a work can only be copied if the owner gives permission. If someone copies a work without permission, the owner can say they infringed their copyright. When this happens, the owner may sue for the amount that should have been paid. Most cases are handled by civil law. In more serious cases, a person who copies a work that is protected under copyright could be arrested, fined or even go to prison. Commonly, the copyright law will protect the authors and their heirs for 50 to 100 years since the author's death, or the first day of multiple authors' deaths.\n\nMany countries have modified their copyright law to meet international standards. However, there are still differences in national laws. In some countries, someone violating copyright law will be sued only to the civil law courts but other countries they can also be charged by criminal courts.\n\nASSISTANT: 1. Copyright is a law that gives the owner of a work (for example, a book, movie, picture, song or website) the right to say how other people can use it.\n2. Copyright laws make it easier for authors to make money by selling their works.\n3. With copyright, a work can only be copied if the owner gives permission. '
  • 'USER: Make a comma-separated list of all jobs or occupations mentioned in the following article that are affected by products containing toxic chemicals.\nHere is the article to reference:\n\nConsumer products released more than 5,000 tons of chemicals in 2020 inside California homes and workplaces that are known to cause cancer, adversely affect sexual function and fertility in adults or harm developing fetuses, according to our newly published study.\n\nWe found that many household products like shampoos, body lotions, cleaners and mothballs release toxic volatile organic compounds, or VOCs, into indoor air. In addition, we identified toxic VOCs that are prevalent in products heavily used by workers on the job, such as cleaning fluids, adhesives, paint removers and nail polish. However, gaps in laws that govern ingredient disclosure mean that neither consumers nor workers generally know what is in the products they use.\n\nFor this study we analyzed data from the California Air Resources Board (CARB), which tracks VOCs released from consumer products in an effort to reduce smog. The agency periodically surveys companies that sell products in California, collecting information on concentrations of VOCs used in everything from hair spray to windshield wiper fluid.\n\nAnalysis of the world, from experts\nWe cross-referenced the most recent data with a list of chemicals identified as carcinogens or reproductive/developmental toxicants under California’s right-to-know law, Proposition 65. This measure, enacted in 1986, requires businesses to notify Californians about significant exposure to chemicals that are known to cause cancer, birth defects or other reproductive harms.\n\nWe found 33 toxic VOCs present in consumer products. Over 100 consumer products covered by the CARB contain VOCs listed under Prop 65.\n\nOf these, we identified 30 product types and 11 chemicals that we see as high priorities for either reformulation with safer alternatives or regulatory action because of the chemicals’ high toxicity and widespread use.\n\n\nWhy it matters\nOur study identifies consumer products containing carcinogens and reproductive and developmental toxicants that are widely used at home and in the workplace. Consumers have limited information about these products’ ingredients.\n\nWe also found that people are likely co-exposed to many hazardous chemicals together as mixtures through use of many different products, which often contain many chemicals of health concern. For example, janitors might use a combination of general cleaners, degreasers, detergents and other maintenance products. This could expose them to more than 20 different Prop 65-listed VOCs.\n\nSimilarly, people experience aggregate exposures to the same chemical from multiple sources. Methanol, which is listed under Prop 65 for developmental toxicity, was found in 58 product categories. Diethanolamine, a chemical frequently used in products like shampoos that are creamy or foamy, appeared in 40 different product categories. Canada and the European Union prohibit its use in cosmetics because it can react with other ingredients to form chemicals that may cause cancer.\n\nSome chemicals, such as N-methyl-2-pyrrolidone and ethylene gylcol, are listed under Prop 65 because they are reproductive or developmental toxicants. Yet they appeared widely in goods such as personal care products, cleansers and art supplies that are routinely used by children or people who are pregnant.\n\nOur findings could help state and federal agencies strengthen chemical regulations. We identified five chemicals – cumene, 1,3-dichloropropene, diethanolamine, ethylene oxide and styrene – as high-priority targets for risk evaluation and management under the Toxic Substances Control Act by the U.S. Environmental Protection Agency.\n\nA hotel housekeeper stands next to her cart, piled with towels and bottled cleaning supplies.\nMany jobs, including custodian and hotel housekeeper, involve close-range exposure to multiple chemicals daily. Jeff Greenberg/Universal Images Group via Getty Images\nWhat still isn’t known\nOur analysis of the CARB data on volatile toxicants does not paint a complete picture. Many toxic chemicals, such as lead, PFAS and bisphenol A (BPA), don’t have to be reported to the Air Resources Board because they are not volatile, meaning that they don’t readily turn from liquid to gas at room temperature.\n\nIn addition, we were not able to identify specific products of concern because the agency aggregates data over whole categories of products.\n\nWhat other research is being done\nStudies have shown that women generally use more cosmetic, personal care and cleaning products than men, so they are likely to be more highly exposed to harmful chemicals in these categories. Further, women working in settings like nail salons may be exposed from products used both personally and professionally.\n\nResearch by members of our team has also shown that product use varies by race and ethnicity, partly due to racialized beauty standards. Policy interventions could be tailored to prioritize these potentially more-highly exposed groups.\n\nUltimately, a right-to-know law like Prop 65 can only go so far in addressing toxics in products. We’ve found in other research that some manufacturers do choose to reformulate their products to avoid Prop 65 chemicals, rather than having to warn customers about toxic ingredients.\n\nBut Prop 65 does not ban or restrict any chemicals, and there is no requirement for manufacturers to choose safer substitutes. We believe our new analysis points to the need for national action that ensures consumers and workers alike have safer products.\n\nASSISTANT: Janitors, custodians, hotel housekeepers, women working in nail salons.'
  • "USER: Please make a bullet point list of all the movies listed in this paragraph.\n\nLoren became an international film star following her five-picture contract with Paramount Pictures in 1958. Among her films at this time were Desire Under the Elms with Anthony Perkins, based upon the Eugene O'Neill play; Houseboat, a romantic comedy co-starring Cary Grant; and George Cukor's Heller in Pink Tights, in which she appeared as a blonde for the first time.\n\n\nDrawing of Loren by Nicholas Volpe after she won an Oscar for Two Women (1961)\nIn 1960, Loren starred in Vittorio De Sica's Two Women, a stark, gritty story of a mother who is trying to protect her 12-year-old daughter in war-torn Italy. The two end up gang-raped inside a church as they travel back to their home city following cessation of bombings there. Originally cast as the daughter, Loren fought against type and was eventually cast as the mother (actress Eleonora Brown would portray the daughter). Loren's performance earned her many awards, including the Cannes Film Festival's best performance prize, and an Academy Award for Best Actress, the first major Academy Award for a non-English-language performance or to an Italian actress. She won 22 international awards for Two Women. The film was extremely well received by critics and a huge commercial success. Though proud of this accomplishment, Loren did not show up to this award, citing fear of fainting at the award ceremony. Nevertheless, Cary Grant telephoned her in Rome the next day to inform her of the Oscar award.[18]\n\nDuring the 1960s, Loren was one of the most popular actresses in the world, and continued to make films in the United States and Europe, starring with prominent leading men. In 1961 and 1964, her career reached its pinnacle when she received $1 million to appear in El Cid and The Fall of the Roman Empire. In 1965, she received a second Academy Award nomination for her performance in Marriage Italian-Style opposite Marcello Mastroianni.[19]\n\nASSISTANT: • Desire Under the Elms\n• Houseboat\n• Heller in Pink Tights\n• Two Women\n• El Cid\n• The Fall of the Roman Empire\n• Marriage Italian-Style"

Evaluation

Metrics

Label Accuracy
all 0.9283

Uses

Direct Use for Inference

First install the SetFit library:

pip install setfit

Then you can load this model and run inference.

from setfit import SetFitModel

# Download from the 🤗 Hub
model = SetFitModel.from_pretrained("setfit_model_id")
# Run inference
preds = model("USER: Write a haiku about a kid who hates doing homework, but doesn't have a choice. Make it funny but keep it to the point. Keep it vivid even though it's a haiku.

ASSISTANT: My pen-covered hands
slave away at my homework
enslaved until done.")

Training Details

Training Set Metrics

Training set Min Median Max
Word count 20 216.0099 2635
Label Training Sample Count
Coding 50
Rewrite 50
Open QA 54
Generation 50
Classify 50
Extract 50
Summarize 50
Closed QA 51
Math 50
Brainstorm 50

Training Hyperparameters

  • batch_size: (24, 24)
  • num_epochs: (1, 1)
  • max_steps: 2000
  • sampling_strategy: oversampling
  • body_learning_rate: (2e-05, 1e-05)
  • head_learning_rate: 0.01
  • loss: CosineSimilarityLoss
  • distance_metric: cosine_distance
  • margin: 0.25
  • end_to_end: False
  • use_amp: False
  • warmup_proportion: 0.1
  • seed: 42
  • eval_max_steps: -1
  • load_best_model_at_end: True

Training Results

Epoch Step Training Loss Validation Loss
0.0001 1 0.3244 -
0.0052 50 0.129 -
0.0105 100 0.0488 -
0.0157 150 0.0162 -
0.0209 200 0.0013 -
0.0261 250 0.001 -
0.0314 300 0.0004 -
0.0366 350 0.0002 -
0.0418 400 0.0002 -
0.0471 450 0.0002 -
0.0523 500 0.0002 -
0.0575 550 0.0002 -
0.0627 600 0.0001 -
0.0680 650 0.0002 -
0.0732 700 0.0001 -
0.0784 750 0.0001 -
0.0837 800 0.0 -
0.0889 850 0.0001 -
0.0941 900 0.0001 -
0.0993 950 0.0001 -
0.1046 1000 0.0001 -
0.1098 1050 0.0 -
0.1150 1100 0.0001 -
0.1203 1150 0.0001 -
0.1255 1200 0.0 -
0.1307 1250 0.0001 -
0.1359 1300 0.0 -
0.1412 1350 0.0 -
0.1464 1400 0.0 -
0.1516 1450 0.0 -
0.1569 1500 0.0 -
0.1621 1550 0.0 -
0.1673 1600 0.0 -
0.1725 1650 0.0 -
0.1778 1700 0.0 -
0.1830 1750 0.0001 -
0.1882 1800 0.0 -
0.1935 1850 0.0 -
0.1987 1900 0.0 -
0.2039 1950 0.0 -
0.2091 2000 0.0 0.063
  • The bold row denotes the saved checkpoint.

Framework Versions

  • Python: 3.10.14
  • SetFit: 1.0.3
  • Sentence Transformers: 3.0.1
  • Transformers: 4.40.0
  • PyTorch: 2.4.1+cu121
  • Datasets: 2.21.0
  • Tokenizers: 0.19.1

Citation

BibTeX

@article{https://doi.org/10.48550/arxiv.2209.11055,
    doi = {10.48550/ARXIV.2209.11055},
    url = {https://arxiv.org/abs/2209.11055},
    author = {Tunstall, Lewis and Reimers, Nils and Jo, Unso Eun Seo and Bates, Luke and Korat, Daniel and Wasserblat, Moshe and Pereg, Oren},
    keywords = {Computation and Language (cs.CL), FOS: Computer and information sciences, FOS: Computer and information sciences},
    title = {Efficient Few-Shot Learning Without Prompts},
    publisher = {arXiv},
    year = {2022},
    copyright = {Creative Commons Attribution 4.0 International}
}