Spaces:
Running
Running
| import gradio as gr | |
| import random | |
| import torch | |
| from transformers import AutoConfig | |
| from transformers import GPT2Tokenizer, GPT2LMHeadModel | |
| from itertools import chain | |
| config = AutoConfig.from_pretrained('gorkemgoknar/gpt2chatbotenglish') | |
| model = GPT2LMHeadModel.from_pretrained('gorkemgoknar/gpt2chatbotenglish', config=config) | |
| tokenizer = GPT2Tokenizer.from_pretrained('gorkemgoknar/gpt2chatbotenglish') | |
| tokenizer.model_max_length = 1024 | |
| #Dynamic Temperature | |
| base_temperature = 1.3 | |
| dynamic_temperature_range = 0.15 | |
| rand_range = random.uniform(-1 * dynamic_temperature_range , dynamic_temperature_range ) | |
| temperature = base_temperature + rand_range | |
| SPECIAL_TOKENS = ["<bos>", "<eos>", "<speaker1>", "<speaker2>", "<pad>"] | |
| #See document for experiment https://www.linkedin.com/pulse/ai-goes-job-interview-g%C3%B6rkem-g%C3%B6knar/ | |
| def get_chat_response(name, input_txt = "Hello , what is your name?",history=[]): | |
| history.append(input_txt) | |
| history_e = [tokenizer.encode(e) for e in history] | |
| personality = "My name is " + name | |
| bos, eos, speaker1, speaker2 = tokenizer.convert_tokens_to_ids(SPECIAL_TOKENS[:-1]) | |
| sequence = [[bos] + tokenizer.encode(personality)] + history_e | |
| sequence = [sequence[0]] + [[speaker2 if (len(sequence)-i) % 2 else speaker1] + s for i, s in enumerate(sequence[1:])] | |
| sequence = list(chain(*sequence)) | |
| #bot_input_ids = tokenizer.encode(personality + tokenizer.eos_token + input_txt + tokenizer.eos_token , return_tensors='pt') | |
| sequence_len = len(sequence) | |
| #optimum response and speed | |
| chat_history_ids = model.generate( | |
| torch.tensor(sequence).unsqueeze(0), max_length=50, | |
| pad_token_id=tokenizer.eos_token_id, | |
| no_repeat_ngram_size=3, | |
| do_sample=True, | |
| top_k=60, | |
| top_p=0.8, | |
| temperature = 1.3 | |
| ) | |
| out_str = tokenizer.decode(chat_history_ids[0][sequence_len:], skip_special_tokens=True) | |
| #out_str = tokenizer.decode(chat_history_ids[:, sequence.shape[-1]:][0], skip_special_tokens=False) | |
| return out_str | |
| ##you can use anyone from below | |
| ''' | |
| | Macleod | Moran | Brenda | Ramirez | Peter Parker | Quentin Beck | Andy | |
| | Red | Norton | Willard | Chief | Chef | Kilgore | Kurtz | Westley | Buttercup | |
| | Vizzini | Fezzik | Inigo | Man In Black | Taylor | Zira | Zaius | Cornelius | |
| | Bud | Lindsey | Hippy | Erin | Ed | George | Donna | Trinity | Agent Smith | |
| | Morpheus | Neo | Tank | Meryl | Truman | Marlon | Christof | Stromboli | Bumstead | |
| | Schreber | Walker | Korben | Cornelius | Loc Rhod | Anakin | Obi-Wan | Palpatine | |
| | Padme | Superman | Luthor | Dude | Walter | Donny | Maude | General | Starkiller | |
| | Indiana | Willie | Short Round | John | Sarah | Terminator | Miller | Sarge | Reiben | |
| | Jackson | Upham | Chuckie | Will | Lambeau | Sean | Skylar | Saavik | Spock | |
| | Kirk | Bones | Khan | Kirk | Spock | Sybok | Scotty | Bourne | Pamela | Abbott | |
| | Nicky | Marshall | Korshunov | Troy | Vig | Archie Gates | Doc | Interrogator | |
| | Ellie | Ted | Peter | Drumlin | Joss | Macready | Childs | Nicholas | Conrad | |
| | Feingold | Christine | Adam | Barbara | Delia | Lydia | Cathy | Charles | Otho | |
| | Schaefer | Han | Luke | Leia | Threepio | Vader | Yoda | Lando | Elaine | Striker | |
| | Dr. Rumack | Kramer | David | Saavik | Kirk | Kruge | Holden | Deckard | Rachael | |
| | Batty | Sebastian | Sam | Frodo | Pippin | Gandalf | Kay | Edwards | Laurel | |
| | Edgar | Zed | Jay | Malloy | Plissken | Steve Rogers | Tony Stark | Scott Lang | |
| | Bruce Banner | Bruce | Edward | Two-Face | Batman | Chase | Alfred | Dick | |
| | Riddler | Din Djarin | Greef Karga | Kuiil | Ig-11 | Cara Dune | Peli Motto | |
| | Toro Calican | Ripley | Meredith | Dickie | Marge | Peter | Lambert | Kane | |
| | Dallas | Ripley | Ash | Parker | Threepio | Luke | Leia | Ben | Han | Common Bob | |
| | Common Alice | Jack | Tyler | Marla | Dana | Stantz | Venkman | Spengler | Louis | |
| | Fry | Johns | Riddick | Kirk | Decker | Spock | "Ilia | Indy | Belloq | Marion | |
| | Brother | Allnut | Rose | Qui-Gon | Jar Jar | |
| ''' | |
| def greet(character,input_txt): | |
| ##add a bootstrap history (item 1 is speaker, item 2 is character) | |
| history = ["Who are you?","I am " + character] | |
| return get_chat_response(character,history=history,input_txt=input_txt) | |
| #some selected ones are in for demo use | |
| personality_choices = ["Gandalf", "Riddick", "Macleod", "Morpheus", "Neo","Spock","Vader","Indy", "Ig-11","Threepio","Tony Stark","Batman","Vizzini"] | |
| examples= ["Gandalf", "What is your name?"] | |
| title = "Metayazar - Movie Chatbot" | |
| description = "Chat with your favorite movie characters. This space demo has simple interface and no history, test it out in metayazar.com/chatbot for more movie/character options and history memorized." | |
| #History not implemented in this demo, use metayazar.com/chatbot for a movie and character dropdown chat interface | |
| interface = gr.Interface(fn=greet, inputs=[gr.inputs.Dropdown(personality_choices) ,"text"], title=title, description=description, outputs="text") | |
| if __name__ == "__main__": | |
| interface.launch() |