run_llm.py

import os
import sys
import json
import time
import openai
import pickle
import argparse
import requests
from tqdm import tqdm
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM, LlamaForCausalLM, LlamaTokenizer

from fastchat.model import load_model, get_conversation_template, add_model_args

from nltk.tag.mapping import _UNIVERSAL_TAGS

import gradio as gr

uni_tags = list(_UNIVERSAL_TAGS)
uni_tags[-1] = 'PUNC'

bio_tags = ['B', 'I', 'O']
chunk_tags = ['ADJP', 'ADVP', 'CONJP', 'INTJ', 'LST', 'NP', 'O', 'PP', 'PRT', 'SBAR', 'UCP', 'VP']

syntags = ['NP', 'S', 'VP', 'ADJP', 'ADVP', 'SBAR', 'TOP', 'PP', 'POS', 'NAC', "''", 'SINV', 'PRN', 'QP', 'WHNP', 'RB', 'FRAG',
 'WHADVP', 'NX', 'PRT', 'VBZ', 'VBP', 'MD', 'NN', 'WHPP', 'SQ', 'SBARQ', 'LST', 'INTJ', 'X', 'UCP', 'CONJP', 'NNP', 'CD', 'JJ',
 'VBD', 'WHADJP', 'PRP', 'RRC', 'NNS', 'SYM', 'CC']

openai.api_key = "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"


# determinant vs. determiner
# https://wikidiff.com/determiner/determinant
ents_prompt = ['Noun','Verb','Adjective','Adverb','Preposition/Subord','Coordinating Conjunction',# 'Cardinal Number',
    'Determiner',
    'Noun Phrase','Verb Phrase','Adjective Phrase','Adverb Phrase','Preposition Phrase','Conjunction Phrase','Coordinate Phrase','Quantitave Phrase','Complex Nominal',
    'Clause','Dependent Clause','Fragment Clause','T-unit','Complex T-unit',# 'Fragment T-unit',
][7:]
ents = ['NN', 'VB', 'JJ', 'RB', 'IN', 'CC', 'DT', 'NP', 'VP', 'ADJP', 'ADVP', 'PP', 'CONJP', 'CP', 'QP', 'CN', 'C', 'DC', 'FC', 'T', 'CT'][7:]


ents_prompt_uni_tags = ['Verb', 'Noun', 'Pronoun', 'Adjective', 'Adverb', 'Preposition and Postposition', 'Coordinating Conjunction', 
                        'Determiner', 'Cardinal Number', 'Particles or other function words', 
                        'Words that cannot be assigned a POS tag', 'Punctuation']

ents = uni_tags + ents
ents_prompt = ents_prompt_uni_tags + ents_prompt

for i, j in zip(ents, ents_prompt):
    print(i, j)
# raise


model_mapping = {
    # 'gpt3': 'gpt-3',
    'gpt3.5': 'gpt-3.5-turbo-0613',
    'vicuna-7b': 'lmsys/vicuna-7b-v1.3',
    'vicuna-13b': 'lmsys/vicuna-13b-v1.3',
    'vicuna-33b': 'lmsys/vicuna-33b-v1.3',
    'fastchat-t5': 'lmsys/fastchat-t5-3b-v1.0',
    # 'llama2-7b': 'meta-llama/Llama-2-7b-hf',
    # 'llama2-13b': 'meta-llama/Llama-2-13b-hf',
    # 'llama2-70b': 'meta-llama/Llama-2-70b-hf',
    'llama-7b': './llama/hf/7B',
    'llama-13b': './llama/hf/13B',
    'llama-30b': './llama/hf/30B',
    # 'llama-65b': './llama/hf/65B',
    'alpaca': './alpaca-7B',
    # 'koala-7b': 'koala-7b',
    # 'koala-13b': 'koala-13b',
}

with open('sample_uniform_1k_2.txt', 'r') as f:
    selected_idx = f.readlines()
selected_idx = [int(i.strip()) for i in selected_idx]#[s:e]


ptb = []
with open('ptb.jsonl', 'r') as f:
    for l in f:
        ptb.append(json.loads(l))


## Prompt 1
template_all = '''Please output the <Noun, Verb, Adjective, Adverb, Preposition/Subord, Coordinating Conjunction, Cardinal Number, Determiner, Noun Phrase, Verb Phrase, Adjective Phrase, Adverb Phrase, Preposition Phrase, Conjunction Phrase, Coordinate Phrase, Quantitave Phrase, Complex Nominal, Clause, Dependent Clause, Fragment Clause, T-unit, Complex T-unit, Fragment T-unit> in the following sentence without any additional text in json format: "{}"'''
template_single = '''Please output any <{}> in the following sentence one per line without any additional text: "{}"'''

## Prompt 2
prompt2_pos = '''Please pos tag the following sentence using Universal POS tag set without generating any additional text: {}'''
prompt2_chunk = '''Please do sentence chunking for the following sentence as in CoNLL 2000 shared task without generating any addtional text: {}'''
prompt2_parse = '''Generate textual representation of the constituency parse tree of the following sentence using Penn TreeBank tag set without outputing any additional text: {}'''

prompt2_chunk = '''Please chunk the following sentence in CoNLL 2000 format with BIO tags without outputing any additional text: {}'''

## Prompt 3
with open('demonstration_3_42_pos.txt', 'r') as f:
    demon_pos = f.read()
with open('demonstration_3_42_chunk.txt', 'r') as f:
    demon_chunk = f.read()
with open('demonstration_3_42_parse.txt', 'r') as f:
    demon_parse = f.read()