Instruction Tuned (IT) models

This document provides information for SambaStudio’s Instruction Tuned (IT) models. These checkpoints are trained on instruction tuning data, which is a diverse collection of tasks that teach the model to respond to instructions and generalized for a wide variety of applications.

Data preparation for training

The Generative data preparation repo External link describes how to prepare data to be used to train SambaStudio’s Instruction Tuned (IT) models. To access the data preparation package, including its associated documentation, please visit the SambaNova public GitHub using the following link: https://github.com/sambanova/generative_data_prep External link

GitHub generative data preparation
Figure 1. Generative data preparation repo on SambaNova public GitHub

Prompt guidelines

End prompts with either a colon (:), a question mark (?), or another way of letting the model know that it is time for it to start generating. For example, using Please summarize the previous article: (with a colon) is a better prompt than Please summarize the previous article (without a colon). Adding these annotations tends to lead to better generations as it indicates to the model that you’ve finished your question and are expecting an answer.

Example prompts

The few-shot examples below demonstrate prompts for SambaStudio’s Instruction Tuned (IT) models. Each example is identified by a task type. Click the triangle to expand and view each example prompt.

Extractive QA (Natural Questions)

Prompt:
Passage: The Little White Bird is a series of short episodes, including both accounts of the narrator’s day-to-day activities in contemporary London and fanciful tales set in Kensington Gardens and elsewhere.The story is set in several locations; the earlier chapters are set in the town of London, contemporaneous to the time of Barrie’s writing, and involving some time travel of a few years, and other fantasy elements, while remaining within the London setting. The middle chapters that later became Peter Pan in Kensington Gardens are set in London’s famous Kensington Gardens, introduced by the statement that "All perambulators lead to Kensington Gardens". The Kensington Gardens chapters include detailed descriptions of the features of the Gardens, along with fantasy names given to the locations by the story’s characters, especially after "Lock-Out Time", described by Barrie as the time at the end of the day when the park gates are closed to the public, and the fairies and other magical inhabitants of the park can move about more freely than during the daylight, when they must hide from ordinary people. The third section of the book, following the Kensington Gardens chapters, are again set generally in London, though there are some short returns to the Gardens that are not part of the Peter Pan stories. In a two-page diversion in chapter 24, Barrie brings the story to Patagonia, and a journey by ship returning to England at the "white cliffs of Albion".
Question: Where was the ship in Patagonia returned to?
Answer: England

Passage: The film explores several flashbacks and present timelines to show how Dean and Cindy became involved. Dean is a young high school dropout, working for a moving company in Brooklyn. Cindy is a pre-med student living with her constantly fighting parents and also caring for her grandmother in Pennsylvania. Cindy and Dean meet at Cindy’s grandmother’s nursing home while Dean is delivering a new resident’s furniture and they begin dating afterwards. Cindy discovers she is pregnant, and tells Dean that the baby is most likely not his, as her ex-boyfriend Bobby didn’t use protection. Dean asks Cindy whether or not she wants to keep the baby. At an abortion clinic, Cindy decides at the last moment to cancel the procedure, and on a bus ride home, Dean tells her he doesn’t mind if the child is not his, and that he wants to begin a family with her. Before the wedding, Bobby finds out about Dean, and beats him up. Five years later, the couple lives in rural Pennsylvania with their daughter Frankie. Dean works at painting houses while Cindy is a nurse at a clinic. One evening, Dean insists on taking Cindy out for a romantic getaway at a motel so they can have some time off from their preoccupied lives, much to Cindy’s reluctance. While buying wine in a liquor store, Cindy sees Bobby, who asks Cindy if she has ever cheated on her husband. She hesitates, but eventually says no. In the car, Cindy and Dean get into an argument when she mentions seeing Bobby again. At the motel, they continue fighting during sex. Cindy is called away early in the morning to work at the clinic, and she leaves a note for Dean. At the clinic, Cindy’s boss, Dr. Feinberg, talks to her about a position he had offered her, and asks if she would move closer to work, suggesting that they would be able to spend time together on weekends. Visibly upset, Cindy says she previously thought he was offering her the position because she was good at her job. Angered that Cindy left the motel without waking him, Dean shows up drunk at the clinic, leading to a violent altercation with Dr. Feinberg. Cindy says she wants a divorce after Dr. Feinberg fires her. After leaving the clinic, Dean tries to persuade Cindy to give the marriage another chance, asking if she wants their daughter to grow up in a broken home. Cindy says she does not want Frankie to grow up with parents who are so hateful to each other. Dean reminds Cindy of their wedding vows, and the two apologize to each other. Dean is seen walking away from the house, with Frankie running after him. Dean tells Frankie to go back to her mom despite Frankie begging him to stay. Dean tricks Frankie by challenging her to a race in an attempt to send her back to Cindy, and he continues walking away while Cindy picks up an upset Frankie, who cries "I love him".
Question: Who beats up Dean before the wedding?
Answer: Bobby

Passage: Olive Penderghast, a 17-year-old girl living in Ojai, California lies to her best friend Rhiannon Abernathy about going on a date in order to get out of camping with Rhiannon’s hippie parents. Instead, she hangs around the house all weekend listening to Natasha Bedingfield’s "Pocketful of Sunshine", which is played by a greeting card she was sent. The following Monday, pressed by Rhiannon, Olive lies about losing her virginity to a college guy. Marianne Bryant, a prissy and strictly religious Christian at their school, overhears her telling the lie and soon it spreads like wildfire. The school’s conservative church group run by Marianne decides Olive will be their next project. Olive confides the truth to her friend Brandon, and he explains how others bully him because of his homosexuality. He later asks Olive to pretend to sleep with him so that he will be accepted by everyone as a 'straight stud'. Brandon convinces Olive to help him and they pretend to have sex at a party. After having a fight with Rhiannon over Olive’s new identity as a "dirty skank", Olive decides to counteract the harassment by embracing her new image as the school tramp. She begins to wear more provocative clothing and stitches a red "A" to everything she wears. Boys who usually have had no luck with girls in the past beg Olive to say they have had sex with her in order to increase their own popularity, in exchange for gift cards to various stores, in turn increasing her reputation. Things get worse when Micah, Marianne’s 22-year-old boyfriend, contracts chlamydia from sleeping with Mrs. Griffith, the school guidance counsellor, and blames it all on Olive. Olive agrees to lie to cover up the affair so that the marriage of her favorite teacher, Mr. Griffith, would be spared. Marianne’s religious clique, which now includes Rhiannon, begins harassing Olive in order to get her to leave school. After an ill-fated date with Anson, a boy who wants to pay her to actually sleep with him and not just pretend she did, Olive reconnects with Todd, her old crush, who is also the school’s mascot. Todd then tells her that he does not believe the rumors because he remembers when she lied for him when he was not ready for his first kiss years ago. Olive then begins to ask everyone she lied for to help her out by telling the truth, but Brandon and Micah have abruptly left town and everyone else is enjoying their newfound popularity and do not want the truth to get out. Mrs. Griffith also refuses to tell the truth and when Olive threatens to expose her, Mrs. Griffith rebuffs her, saying no one would believe her. Olive, out of spite, then immediately tells Mr. Griffith, who believes her and separates from Mrs. Griffith. After a friendly talk with her eccentric, open-minded mother Rosemary, Olive comes up with a plan to get everything finally out in the open. She then does a song and dance number at a school pep rally to get people’s attention to watch her via web cam, where she confesses what she has done (the web cam is the framing device of the film). The various boys whose reputations Olive helped improve are also shown watching. Later, Olive texts Rhiannon, apologizing for lying to her. When she is finishing up her web cast, Todd comes by riding a lawnmower and tells her to come outside. She signs off by saying she may lose her virginity to Todd, and proudly declares it’s nobody’s business (much to Marianne’s disgrace). She goes outside to meet him, they kiss and the two are shown riding off on the lawnmower.
Question: Who is Todd besides Olive’s old crush?
Answer:

Summarization (CNN Daily Mail)

Prompt:
###
Article: (CNN Student News) — November 9, 2012 . Download PDF maps related to today’s show: . Greece . Guatemala . Japan . Michigan; Utah . Click here to access the transcript of today’s CNN Student News program. Please note that there may be a delay between the time when the video is available and when the transcript is published.

Summarize the above article in 3 sentences.
The daily transcript is a written version of each day’s CNN Student News program . Use this transcript to help students with reading comprehension and vocabulary . Use the weekly Newsquiz to test your knowledge of stories you saw on CNN Student News .

###
Article: KATHMANDU, Nepal (CNN) — Two people were killed and about a dozen others were injured when a bomb exploded in a Catholic church in Kathmandu on Saturday morning, police said. The damage inside the church in Kathmandu following Saturday’s bomb blast. The explosion in the Nepalese capital killed a 15-year-old girl and a 30-year-old woman. "The bomb exploded inside the church when the explosion happened," senior police officer Kedar Man Singh Bhandari told CNN over the phone. About 100 people were in the church when the bomb exploded, police said. Manish Amatya, who was injured, said the blast interrupted their prayers. "There was a loud explosion while we were praying and all of us ran out screaming," he said. Investigations are under way to determine who planted the bomb, which damaged the church. CNN’s Manesh Shrestha contributed to this report.

Summarize the above article in 3 sentences.
Explosion in Nepalese capital killed 15-year-old girl, 30-year-old woman . 100 people were in the church when the bomb exploded . Investigations are under way to determine who planted the bomb

###
Article: NEW DELHI, India (CNN) — At least 441 people have died in floods in India from this season’s monsoon rains, federal authorities said in their latest report. An Indian child plays in a flooded street in Mumbai earlier this month. Flooding has affected more than 1.5 million people in parts of India, said the disaster management division of the federal home ministry. The country’s main weather office has warned of more heavy rain in western and central parts of India. Monsoon rains sweep across the subcontinent from June till September. Though they bring much-needed relief to often-parched farmlands, they also leave a trail of landslides, home collapses and floods that can kill. In neighboring Pakistan, torrential monsoon rains left more than three dozen people dead and broke a 32-year record over the weekend. CNN’s Harmeet Shah Singh contributed to this report.

Summarize the above article in 3 sentences.
7 die as bus carrying 40 passengers sinks in overflowing canal in eastern India . 7-year-old girl and her mother among the dead . Bus driver ignored warnings from his passengers about flooding in canal .

###
Article: (CNN)Each day, CNN producers select a user-submitted photo to be our Travel Photo of the Day. Click through the gallery above to see stunning shots from around the world, and be sure to come back every day for a new image. Have a gorgeous travel photo of your own to share? Submit it for the gallery at CNN iReport!

Summarize the above article in 3 sentences.
See more iReport galleries: Glorious Ireland, beautiful beaches . Follow us on Twitter @cnnireport and @CNNTravel .

###
Article: NEW YORK (CNN) — A nude photograph of pop singer Madonna was sold for $37,500 Thursday afternoon at a Christie’s Art House auction. Christie’s auctioned this nude photo of Madonna (partially shown) taken by Lee Friedlander for $37,500. The photo, originally expected to go for between $10,000 and $15,000, was purchased for more than double its original estimated selling price, a Christie’s spokesperson confirmed. The 13-inch by 8 5/8-inch framed photograph was purchased by an anonymous bidder over the phone. The full frontal photograph was one of several taken by American photographer Lee Friedlander in 1979. Madonna, then a cash-strapped student, received $25 for the entire photo shoot. Most of the pictures from the shoot were ultimately featured in Playboy magazine in 1985.

Summarize the above article in 3 sentences.
Nude photograph of Madonna taken when she was student in 1979 . Lee Friedlander pic sold by Christie’s for $37,500 . Anonymous bidder made purchase over the phone .

###
Article: Former PM Tony Blair was given an award for his 'global legacy' by Save the Children last year . Save the Children has apologised over its decision to give Tony Blair an award for his 'global legacy' after a furious public backlash. Justin Forsyth, the charity’s chief executive, said he was 'very sorry' for those upset by the decision, but insisted it was only meant to mark his work in Africa and not his entire time in Number 10. Mr Forsyth, a former Downing Street aide, admitted it would have been 'wrong' to give an award to the former PM for his political legacy rather than for his work on international aid. Mr Blair was handed an award for spending billions of taxpayers' money on foreign aid at a star-studded ceremony in New York last year. The former PM, who set Britain on course towards spending 0.7 per cent of GDP on overseas aid, was given the charity’s Global Legacy Award in recognition of his leadership in international development. But the award sparked widespread fury when it was announced in November. More than 100,000 people signed an online petition calling for the award to be rescinded and around 500 Saved the Children staff members have reportedly backed a similar petition. Questioned about the award on the BBC’s Today programme this morning, Mr Forsyth said: 'I know that many of our supporters and volunteers were very upset and our staff, several of our staff too, and I’m very sorry for that.' He said the decision had been made by the British charity’s 'sister organisation' in the US and insisted the award was only for his Africa work. Mr Forsyth said: 'Yes, it was a global legacy award. It was called that. But actually it was an award very, very specifically for Tony Blair’s efforts on Africa at two G8 summits at Birmingham and Gleneagles, not his wider legacy. 'And if it had been for his wider legacy I think it would be wrong, but it was for something quite specific that helped Africa and children. 'I’m sorry it’s upset so many people. It’s not really what we do at Save the Children. What we’re really doing is on the ground in Syria with the Ebola treatment centre in Sierra Leone – that’s the work we do. 'This has been really an unnecessary distraction. I do apologise to those people that it’s upset.' Justin Forsyth (left), the charity’s chief executive pictured in Lebanon with Samantha Cameron, said he was 'very sorry' for those upset by the decision to honour Tony Blair . Mr Blair was handed the award for spending billions of public money on foreign aid at a star-studded ceremony in New York last year - also attended by Mike Bezos (centre left), Dakota Fanning (centre right), and Nicholas Kristo (right) Mr Forsyth admitted the row had 'in part' damaged the charity in eyes of some staff but expressed hope people would accept organisations 'make mistakes'. Mr Forsyth was recruited to Number 10 by Mr Blair to help with work on poverty and

Summarize the above article in 3 sentences.

Dialogue summarization (SamSum)

Prompt:
In this task, you are given a conversation and your task is to generate a summary from the information present in the given conversation. Generate a summary in such a way that the context should be present in the conversation. It should cover the complete context of the conversation.

Conversation: Lucas: Hey! How was your day?, Demi: Hey there! , Demi: It was pretty fine, actually, thank you!, Demi: I just got promoted! :D, Lucas: Whoa! Great news!, Lucas: Congratulations!, Lucas: Such a success has to be celebrated., Demi: I agree! :D, Demi: Tonight at Death & Co.?, Lucas: Sure!, Lucas: See you there at 10pm?, Demi: Yeah! See you there! :D
Summary: Demi got promoted. She will celebrate that with Lucas at Death & Co at 10 pm.

Conversation: Mary: Hi Mike!, Mike: Hello :), Mary: do u have any plans for tonight?, Mike: I’m going to visit my grandma., Mike: You can go with me., Mike: She likes u very much., Mary: Good idea, i’ll buy some chocolate for her.
Summary: Mike and Mary are going to visit Mike’s grandma tonight. Mary will buy her some chocolate.

Conversation: Claudia: Where am I supposed to look for a previous watcher? I am there and I cant see anyone who could hand over the lists or keys etc., Richard: Yes, we keep it short and precise. Keys are always in the White Room. Water under the sill is for the instructors., Natalie: (Y), Max: (Y), Alex: white room is the one on the left?, Richard: Exactly :)
Summary: The keys are always in the White Room.

Conversation: Nancy: How do I rent a car?, Frank: You have to have a driver’s license that is valid., Nancy: No problem., Frank: You have to be 25 and I happen to know you are!, Nancy: Next!, Frank: A credit card. That’s it., Nancy: Sweet. I think I’ll do that this weekend., Frank: Would save time.
Summary:

Text classification (RAFT, Onestopenglish)

Prompt:
The following is an article sourced from The Guardian newspaper, and rewritten by teachers to suit three levels of adult English as Second Language (ESL) learners: elementary, intermediate, and advanced. Predict the level of the article. Possible labels:

  1. Advanced

  2. Elementary

  3. Intermediate

Article: As soon as the children at one primary school in Stirling hear the words “daily mile”, they down their pencils and head out of the classroom to start running laps around the school field. For three-and-a-half years, all pupils at St Ninian’s Primary have walked or run a mile each day. They do so at random times during the day, apparently happily, and, despite the rise in childhood obesity across the UK, none of the children at the school are overweight. The daily mile has done so much to improve these children’s fitness, behaviour and concentration in lessons that scores of nursery and primary schools across Britain are following suit and getting pupils to get up from their desks and take 15 minutes to walk or run round the school or local park.
Elaine Wyllie, headteacher of St Ninian’s, said: “I get at least two emails a day from other schools and local authorities asking how we do it. The thought of children across the country running every day because of something we’ve done is phenomenal.”
One in ten children are obese when they start school at the age of four or five, according to figures from the Health & Social Care Information Centre, and, in the summer of 2015, a study found that schoolchildren in England are the least fit they have ever been. Primary schools have therefore been quick to note the benefits of the daily mile. It has been introduced in schools in London, Gateshead, Wales and other parts of Scotland, while others are planning to launch the initiative during the 2015-6 academic year. In Stirling alone, 30 schools have already started or are to start the daily mile.
The extent of the benefits have yet to be determined but researchers from Stirling University have launched a comparative study to look for quantitative evidence of the physical, cognitive and emotional benefits of the daily mile. Dr Colin Moran, who is leading the study, said: “The children at St Ninian’s don’t seem to have problems with obesity; they seem happier and staff say they settle into lessons faster so we designed a study that would test all of these things. There is a lot of anecdotal evidence about the benefits but there aren’t any scientific facts yet.” St Ninian’s pupils will be compared with children from another school in Stirling that has yet to start the scheme.
Kevin Clelland, a primary school teacher from Leeds, visited St Ninian’s before convincing his colleagues it was a great idea. He said: “It’s such a simple thing to do but seems to have such an amazing impact. We’re really committed to improving the fitness of our pupils beyond the two-hour statutory PE that we are expected to deliver.” His school is now constructing a track. Active Cheshire, a strategic body for sports and fitness in Cheshire and Warrington, is taking a group of senior figures from the local authority up to Scotland to assess the results of the daily mile. The hope is to introduce it across the 450 schools in the region if a pilot is successful. Paralympian, Tanni Grey-Thompson, chair of ukactive, the UK’s leading not-for-profit health body for physical activity, said: “All children need to achieve 60 active minutes every day, whether in a lesson, on the walk to school or in the playground. It’s fantastic to see initiatives like the daily mile be established, showing real leadership from the education sector to improve children’s fitness levels and their cognitive behaviour, and make a real difference to schools, teachers, parents and young people’s lives. We know sitting still kills; not sitting still helps children build skills that will stay with them for life.”
The Scottish government is also supportive. A spokesperson said: “Learning in PE is enhanced by initiatives like the daily mile, which can encourage and support parents in fostering healthy habits with their children from a young age. We are pleased to see so many Scottish schools are taking part or planning to do so.”
Label:

Article: You can be Aagot, Arney or Ásfríður; Baldey, Bebba or Brá. Dögg, Dimmblá, Etna and Eybjört are fine; likewise Frigg, Glódís, Hörn and Ingunn. Jórlaug works OK, as do Obba, Sigurfljóð, Úranía and – should you choose – Vagna. But you cannot, as a girl in Iceland, be called Harriet.
“The whole situation,” said Tristan Cardew, with very British understatement, “is really rather silly.” With his Icelandic wife, Kristin, Cardew is appealing against a decision by the National Registry in the capital Reykjavik not to renew their ten-year-old daughter Harriet’s passport on the grounds that it does not recognize her first name.
The impasse meant the family, from Kópavogur, risked missing their holiday in France until they applied to the British embassy for an emergency UK passport, which should now allow them to leave.
Names matter in Iceland, a country of barely 320,000 people, whose phone book lists subscribers by their first name for the very sensible reason that the vast majority of Icelandic surnames simply record the fact that you are your father’s (or mother’s) son or daughter. Jón Einarsson’s offspring, for example, might be Ólafur Jónsson and Sigríður Jónsdóttir.
The law dictates that the names of children born in Iceland must – unless both parents are foreign – be submitted to the National Registry within six months of birth. If they are not on a recognized list of 1,853 female and 1,712 male names, the parents must seek the approval of a body called the Icelandic Naming Committee.
For the 5,000 or so children born in Iceland each year, the committee reportedly receives about 100 applications and rejects about half under a 1996 act aimed mainly at preserving the language of the sagas. Among its requirements are that given names must be “capable of having Icelandic grammatical endings”, may not “conflict with the linguistic structure of Iceland” and should be “written in accordance with the ordinary rules of Icelandic orthography”.
What this means in practice is that names containing letters that do not officially exist in Iceland’s 32-letter alphabet, such as “c ”, are out. Similarly, names unable to accommodate the endings required by the nominative, accusative, genitive and dative cases used in Icelandic are also routinely turned down. “That was the problem with Harriet,” said Cardew.
The country’s naming laws have come under increasing fire in recent years: in 2013, Blær – “Light Breeze” – Bjarkardóttir Rúnarsdottir won the right to be officially known by her given name, as opposed to “Girl”, when a court ruled that denying her was a violation of the Icelandic constitution. The former mayor of Reykjavik, Jón Gnarr, has also called Iceland’s naming law “unfair, stupid and against creativity”.
The Cardews could get round Harriet’s problem by giving her an Icelandic middle name. “But it’s a bit late for that and way too silly,” said Cardew. “Are they saying they don’t want us here?”
Label:

Sentiment analysis (IMDB)

Prompt:
Passage: No,<br /><br />Basically your watching something that doesn’t make sense. To not spoil the film for people who actually want to this take a look at the flick I will explain the story.<br /><br />A normal everyday to day women, is walking down a street then find’s herself driving by in her own car. She follows her and many events take place during that time that include her and her family.<br /><br />I specifically made an account to comment on this film, of how horribly written this was. The acting was great, the events were great, but the story just brought it nowhere - it could of been added to tremendously and be made into a worldwide epidemic. I’m not sure what the writer was trying to accomplish by making this, usually at the end of films most of your questions get answers but this film has you asking, What just happened and 1 hour 20 minutes just passed for nothing.<br /><br />Spoiler Starts__<br /><br />They had this area between 2 dimensions (ours and behind the glass) that would come into our world and kill us. It was not elaborated on all during the film, and you never know how it was happening or why it was or when it happened. Nothing gets explained during the film. The main character shouldn’t of even been the main character. At the end of the film the guy who finally figures it all out and runs away (her sisters boyfriend) should of been the main character but sadly the movie ends 20 seconds after. <br /><br />I bought this movie for $10, threw it out right after.. don’t waste your time. I really hope nothing like this is made again.
Sentiment: Negative

Passage: A big surprise, probably because I was expecting it to suck. The reviews were pretty dismissive of it, even though they all seemed to agree that the concept was golden: a man finds out his new girlfriend is a super hero, and finds, when he wants to break up with her, that she’s kind of a psycho. I kept expecting it to fall apart, but it never really did. Sure, it doesn’t make as much of its awesome premise as it could, and chooses to be short when it might have been better to expand the film’s universe. But I can’t blame it for that. Uma Thurman is great as the bipolar superhero, G-Girl. And I’ve discovered, after several years of disliking him, that Luke Wilson can be absolutely perfect when cast as a schlub. He’s given two of the best comic performances of 2006 (the other in the pretty much unreleased Idiocracy). I absolutely cracked up at the expressions on his face when he and Thurman first have sex. It’s one of the funniest sex scenes ever. My only real complaint is that they make G-Girl a bit too much of a psycho, like almost unbelievably so. Maybe with some background I could have accepted it better. I can forgive its flaws, though, because I had a really good time watching it. Underrated, for sure.
Sentiment: Positive

Passage: A young boy sees his mother getting killed and his father hanging himself. 20 years later he gets a bunch of friends together to perform an exorcism on himself so he won’t turn out like his father. All the stock characters are in place: the nice couple; the "funny" guy; the tough (but sensitive) hood; the smart girl (she wears glasses—​that’s how we know); the nerd and two no-personality blondes. It all involves some stupid wooden statue that comes to life (don’t ask) and kills people. I knew I was in trouble when, after a great opening scene, we jump to 20 years later—​ALL bad horror movies do that!<br /><br />The dialogue is atrocious, the acting is bad (except for Betsy Palmer—​why Betsy?) and the killings are stupid and/or unimaginative. My favorite scene is when two people are supposedly having sex and the statue knocks the guy off the bed to show he’s fully dressed! A real bad, stupid incoherent horror film. Avoid at all costs.
Sentiment: Negative

Passage: Years ago I was lucky enough to have seen this gem at a >Gypsy film festival in Santa Monica. You know the ending >is not going to be rosie and tragedy will strike but it’s >really about the journey and characters and their dynamics and how they all fit into what was "Yugoslavia". >While I am not Yugonostalgic and tend to shy away from >the current crop of "Yugoslavian" films (give me Ademir >Kenovic over late 90s Kustarica) I’d be happy to have the >chance to stumble on this film again, as it shines in my >celluloid memories. Ever since seeing Who’s Singing Over >There" 15 years ago I still hear the theme tune, sung by >the Gypsies, ruminating through my head… "I am miserable, >I was born that way…" with the accompanying jew’s harp and accordian making the tune both funny and sad. The late, great actor Pavle Vujisic (Muzamer from When Father >was Away on Business) was memorable as the bus driver of >the ill-fated trip in his typical gruff yet loveable manner. Hi
Sentiment: Positive

Passage: This movie got off to an interesting start. Down the road however, the story gets convoluted with a poor illustration of ancient black magic rituals. The male lead was very good , even though he gets the worst end of the stick in the climax. In comparison, this is "Boomerang" meets "Extremities".<br /><br />
Sentiment: Negative

Passage: What can I say about Cruel intentions 2? Well, I can say in all honesty, I will only watch this film again if I am fastened to a chair and have my eyes opened clockwork-orange-style.<br /><br />The film 'stars' Robin Dunne (No, I never heard of him either), whose awful impression of Ryan Phillipe made me cringe throughout. In a case of terrible casting, Dunne attempts (and fails) to carry off playing a handsome charismatic, charmer. Since the actor is not handsome, nor charismatic nor charming, the character is left wholly unbelievable. Amy Adams, (she was in an episode of buffy one time), tries to pick up where Sarah Michelle Gellar left off and bring scheming Katherine to life…​ However, Adams is not that good a an actress and her performance was flat and lacking in any real emotion, often she looked like she was reading cue cards just off camera. There were two good actors in the film however, Barry Flatman (Saw 2 & Saw 3) and Mimi Rogers (Mrs Kensington in Austion Powers), made very good and entertaining performances as the parents of Sebastian and Katherine and are the only reason why I rated the film as a 2, not a 1.<br /><br />The film itself is a poor version of the original, with such lows as carbon copy’s of dialogue and mimicked scenes which lacked the originality of the previous film.<br /><br />I think that as a TV show, it might have worked, but if it had been recasted with people who could actually act in the main parts.
Sentiment:

Named Entity Recognition (Conll)

Prompt:
In this task, you are given a text from a post. Your task is to find all of the proper nouns and label them. The labels are <B-PER>, <I-PER> for persons; <B-ORG>, <I-ORG> for organizations; <B-LOC>, <I-LOC> for locations; and <B-MISC>, <I-MISC> for other nouns. The capital 'B' denotes the first word of a proper noun phrase. The capital 'I' denotes all following words of the same noun phrase. If a word is not a part of a proper noun phrase, do not label it.

Post: Fischler proposed EU-wide measures after reports from Britain and France that under laboratory conditions sheep could contract Bovine Spongiform Encephalopathy ( BSE ) — mad cow disease .
Labels: Fischler <B-PER> proposed EU-wide <B-MISC> measures after reports from Britain <B-LOC> and France <B-LOC> that under laboratory conditions sheep could contract Bovine <B-MISC> Spongiform <I-MISC> Encephalopathy <I-MISC> ( BSE <B-MISC> ) — mad cow disease .

Post: Rare Hendrix song draft sells for almost $ 17,000 .
Labels:

Usage

Instruction Tuned checkpoints are trained to follow instructions to complete any task, without being specifically optimized for a particular task.

  • Use this checkpoint for fine-tuning.

  • Use this checkpoint with few-shot prompting in industrial use cases, such as automated pipelines when verbosity is unnecessary and not desirable.

Few-shot prompting

Few-shot describes a prompting technique that provides a model with examples to process before attempting a task. A few-shot prompt typically includes several examples of problem/solution pairs, known as shots, which provide demonstrations to the model. The demonstrations prime the model for subsequent generative responses.

Playground tuning parameter settings

The Playground tuning parameters provide additional flexibility and options for generative tuning. We recommend the following settings for Instruction Tuned (IT) models used in the Playground.

  • Setting Do sampling to Off is recommended when using Instruction Tuned (IT) models.

    • When Do sampling is set to Off, Temperature, Top k, and Top p are ignored and will have no affect.

Hyperparameters and settings

The hyperparameters and settings for the Instruction Tuned (IT) models when creating a training job are described below.

Parameter Definition Allowed values

do_eval

Specifies if final evaluation is performed.

true, false

eval_steps

Period of evaluating the model in number of training steps: evaluation_strategy must be set to steps for eval_steps to have an affect.

Integer > 0

evaluation_strategy

Strategy to validate the model during training.

no, steps, epoch

learning_rate

The learning rate to use in optimizer.

0.0 < float < 1.0

logging_steps

Period of logging training loss in number of training steps.

Integer > 0

lr_schedule

Type of learning rate scheduler to use.

polynomial_decay_schedule_with_warmup, cosine_schedule_with_warmup, fixed_lr

max_seq_length

Sequence length to pad or truncate the dataset. Should be set to align your dataset size with your chosen model.

Defined by selected model.

num_iterations

The number of iterations to run.

Integer > 0

prompt_loss_weight

Loss scale for the prompt tokens.

0.0 < float < 1.0

save_optimizer_state

Determines whether to save the optimizer state when saving a checkpoint.

true, false

save_steps

Period of saving the model checkpoints in number of training steps.

Integer > 0

skip_checkpoint

Determines whether or not to save output checkpoints.

true, false

subsample_eval

Subsample for the evaluation dataset.

0.0 < float < 1.0

subsample_eval_seed

Random seed to use for the subsample evaluation.

Integer > 0

use_token_type_ids

Determines whether to use token_type_ids to compute loss.

true, false

Setting to true is recommended if Generative data preparation External link was used.

vocab_size

Maximum size of the vocabulary.

Defined by selected model.

warmup_steps

Number of warmup steps to use in learning rate scheduler in optimizer.

Integer > 0

weight_decay

Weight decay rate to use in optimizer.

0.0 < float < 1.0

Inference settings

The inference settings for Instruction Tuned (IT) models when creating a batch inference job are described below.

Parameter Definition Allowed values

do_sample

Toggles whether to use sampling. If not enabled, greedy decoding is used. When enabled, the platform randomly picks the next word according to its conditional probability distribution. Language generation using sampling does not remain deterministic. If you need to have deterministic results, set this to off, as the model is less likely to generate unexpected or unusual words. Setting it to on allows the model a better chance of generating a high quality response, even with inherent deficiencies. However, this is not desirable in an industrial pipeline as it can lead to more hallucinations and non-determinism.

true, false

Setting to false is recommended. If set to false, temperature, top_k, and top_p are ignored and will have no affect.

max_seq_length

Sequence length to pad or truncate the dataset. Should be set to align your dataset size with your chosen model.

Defined by selected model.

max_tokens_to_generate

The maximum numbers of tokens to generate, ignoring the number of tokens in the prompt. When using max tokens to generate, make sure your total tokens for the prompt plus the requested max tokens to generate are not more than the supported sequence length of the model. You can use this parameter to limit the response to a certain number of tokens. The generation will stop under the following conditions:

  1. When the model stops generating due to <|endoftext|>.

  2. The generation encounters a stop sequence set up in the parameters.

  3. The generation reaches the limit for max tokens to generate.

This should not exceed max_seq_length.

1 → max_sequence_length of model

repetition_penalty

The repetition penalty, also known as frequency penalty, controls the model’s tendency to repeat predictions. The repetition penalty reduces the probability of words that have previously been generated. The penalty depends on how many times a word has previously occurred in the prediction. This parameter can be used to penalize words that were previously generated or belong to the context. It decreases the model’s likelihood to repeat the same line verbatim.

Between 1 and 2. ~1.2-1.5

A value setting of 1 means no penalty.

stop_sequences

Stop sequences are used to make the model stop generating text at a desired point, such as the end of a sentence or a list. It is an optional setting that tells the API when to stop generating tokens. The completion will not contain the stop sequence. If nothing is passed, it defaults to the token <|endoftext|>. This token represents a probable stopping point in the text.

Any comma separated strings, each stop word must be enclosed in double quotes.

Example: "Stop phrase 1", "stop phrase 2 with sp3ciAl token$"

temperature

The value used to modulate the next token probabilities. As the value decreases, the model becomes more deterministic and repetitive. With a temperature between 0 and 1, the randomness and creativity of the model’s predictions can be controlled. A temperature parameter close to 1 would mean that the logits are passed through the softmax function without modification. If the temperature is close to 0, the highest probable tokens will become very likely compared to the other tokens: the model becomes more deterministic and will always output the same set of tokens after a given sequence of words.

0 < x ⇐ 1

Has no affect when do_sample is set to false.

top_k

The number of highest probability vocabulary tokens to keep for top k filtering. Top k means allowing the model to choose randomly among the top k tokens by their respective probabilities. For example, choosing the top three tokens means setting the top k parameter to a value of 3. Changing the top k parameter sets the size of the shortlist the model samples from as it outputs each token. Setting top k to 1 gives us greedy decoding.

1 ⇐ x ⇐ vocab_size

Has no affect when do_sample is set to false.

top_logprobs

Shows the top <number> (the numerical value entered) of tokens by its probability to be generated. This indicates how likely a token was to be generated next. This helps debug a given generation and see alternative options to the generated token. The highlighted token is the one that the model predicted with the list sorted by probabilities from high to low, until the top <number> is reached. On the basis of tuning other parameters, you can use the feature to analyze how the predicted tokens by the model might change.

0 ⇐ x ⇐ 20

top_p

Top p sampling, sometimes called nucleus sampling, is a technique used to sample possible outcomes of the model. It controls diversity via nucleus sampling as well as the randomness and originality of the model. The top p parameter specifies a sampling threshold during inference time. Top p shortlists the top tokens whose sum of likelihoods does not exceed a certain value. If set to less than 1, only the smallest set of most probable tokens with probabilities that add up to top p or higher are kept for generation.

0 < x ⇐ 1

Has no affect when do_sample is set to false.

vocab_size

Maximum size of the vocabulary.

Defined by selected model.