from transformers import AutoTokenizer, AutoModelForCausalLM from unidecode import unidecode from collections import Counter import torch import os import gradio as gr import numpy as np import re import string from peft import PeftModel, PeftConfig from transformers import AutoModelForCausalLM from transformers import AutoTokenizer tokenizer = AutoTokenizer.from_pretrained("osiria/primo") model = AutoModelForCausalLM.from_pretrained("osiria/primo") model = PeftModel.from_pretrained(model, "osiria/primo") class Prime: def __init__(self, tokenizer, model): self.tokenizer = tokenizer self.model = model def _check_sublist(self, lst, sub_lst, sep = " "): l_type = type(lst[0]) lst = sep.join(list(map(str, lst))) sub_lst = sep.join(list(map(str, sub_lst))) return sub_lst in lst def _exclude_sublist(self, lst, sub_lst, sep = " "): l_type = type(lst[0]) lst = sep.join(list(map(str, lst))) sub_lst = sep.join(list(map(str, sub_lst))) lst = re.sub("\s+", " ", lst.replace(sub_lst, "")).strip().split(sep) lst = list(map(l_type, lst)) return lst def generate(self, prompt, message = "", sep = " [AI]", max_tokens = 100, excluded = [[40, 19]], lookback = 5, resample_tokens = [27793], replace_tokens = {11302: 23318}, stop_tokens = [239], sample = False, top_k = 5): if message: prompt = message + ". " + prompt prompt = prompt.replace("“", '"').replace("”", '"').replace("’", "'") if not sample: top_k = 2 tokens = tokenizer.encode("[HUMAN] " + prompt + sep) tokens_generated = [] checkpoint = 0 while tokens[-1] not in stop_tokens and len(tokens_generated) < max_tokens: output = model.forward(input_ids=torch.tensor([tokens]).to(device)).logits[0,-1] output = torch.softmax(output, dim = 0) candidates = torch.topk(output, k = top_k) if sample: indices = candidates.indices scores = candidates.values next_token = indices[torch.multinomial(scores, 1)[0].item()] else: next_token = candidates.indices[0] next_token = next_token.item() sub_tokens = tokens_generated[-lookback:] + [next_token] if next_token in resample_tokens: next_token = candidates.indices[1] next_token = next_token.item() if len(tokens_generated) >= (lookback + 1) and next_token in tokens_generated[-2:]: next_token = candidates.indices[1] next_token = next_token.item() elif len(tokens_generated) >= lookback and self._check_sublist(tokens_generated, sub_tokens): if checkpoint: tokens = tokens[:checkpoint] break else: next_token = candidates.indices[1] next_token = next_token.item() sample = True if next_token in replace_tokens: next_token = replace_tokens[next_token] tokens = tokens + [next_token] tokens_generated = tokens_generated + [next_token] if next_token == 5: checkpoint = len(tokens) for ex_lst in excluded: tokens = self._exclude_sublist(tokens, ex_lst) output = tokenizer.decode(tokens, skip_special_tokens=True) output = output.split(sep)[-1].strip() output = output[0].upper() + output[1:] if output[-1] == tokenizer.decode(stop_tokens[0]): output = output[:-1] if len(re.findall("\d\.", output)) > 1: output = re.sub("\d\.", "<br>•", output) return output model.eval() device = torch.device("cuda") prime = Prime(tokenizer = tokenizer, model = model) def process_input(user_input, max_tokens, sample, top_k, message): return prime.generate(prompt = user_input, message = message, max_tokens = max_tokens, sample = sample, top_k = top_k) header = '''-------------------------------------------------------------------------------------------------- <style> .vertical-text { writing-mode: vertical-lr; text-orientation: upright; background-color:red; } </style> <center> <body> <span class="vertical-text" style="background-color:lightgreen;border-radius: 3px;padding: 3px;"> </span> <span class="vertical-text" style="background-color:orange;border-radius: 3px;padding: 3px;"> </span> <span class="vertical-text" style="background-color:lightblue;border-radius: 3px;padding: 3px;"> </span> <span class="vertical-text" style="background-color:tomato;border-radius: 3px;padding: 3px;"> </span> <span class="vertical-text" style="background-color:lightgrey;border-radius: 3px;padding: 3px;"> </span> <span class="vertical-text" style="background-color:#CF9FFF;border-radius: 3px;padding: 3px;"> </span> </body> </center> <br> <center><img src="file/prime.png" width="100"></center> ''' import gradio as gr import random import time with gr.Blocks(title="primo", css="footer {visibility: hidden}", theme=gr.themes.Default(text_size="md", spacing_size="md")) as interface: gr.Markdown(header) with gr.Row(): with gr.Column(scale=1): gr.Markdown("<b>options</b>") max_tokens = gr.Slider(1, 250, value=150, label="max tokens", info="choose a limit between 1 and 250") sample = gr.Checkbox(label="sampling") top_k = gr.Slider(1, 5, value=1, label="creativity", info="choose a level between 1 and 5") message = gr.Textbox(label="system message", value = "") clear = gr.Button("clear chat") with gr.Column(scale=8): chatbot = gr.Chatbot(label = "prime").style(height=600) msg = gr.Textbox(label = "query") def user(user_message, history): return gr.update(value="", interactive=False), history + [[user_message, None]] def bot(history, message, max_tokens, sample, top_k): bot_message = process_input(history[-1][0], message = message, max_tokens = max_tokens, sample = sample, top_k = top_k) history[-1][1] = "" for character in bot_message: history[-1][1] += character time.sleep(0.05) yield history response = msg.submit(user, [msg, chatbot], [msg, chatbot], queue=False).then( bot, [chatbot, message, max_tokens, sample, top_k], chatbot ) response.then(lambda: gr.update(interactive=True), None, [msg], queue=False) clear.click(lambda: None, None, chatbot, queue=False) with gr.Column(scale=1): gr.Markdown("<b>warning</b>") gr.Markdown("the model might behave erratically when presented with prompts which are too far away from its pre-training or fine-tuning and, because of the probabilistic nature of its generation mechanism, it might occasionally produce biased or offensive content with respect to gender, race, ideologies, and political or religious beliefs<br><br>these limitations imply that the model and its outputs should be used with caution, and should not be involved in situations that require the generated text to be fair or true") interface.queue() interface.launch()