LLM Useful Functions Collection

Recommended posts: 【Python】 Python Useful Functions Collection, 【Algorithms】 Lecture 21. NLP and LLM

1. NLP useful functions

⑴ A function that automatically translates a given sentence into English

! pip install --upgrade googletrans httpx httpcore deep_translator

def to_english (sentence):
    from deep_translator import GoogleTranslator
    translated = GoogleTranslator(source='auto', target='en').translate(sentence)
    return translated

print( to_english("나는 소년입니다.") )
# I am a boy.

print( to_english("단핵구") )
# monocytes

⑵ A function that automatically translates a given sentence into Korean

def to_korean (sentence):
    from deep_translator import GoogleTranslator
    translated = GoogleTranslator(source='auto', target='ko').translate(sentence)
    return translated

print( to_korean("I am a boy.") )
# 저는 남자입니다.

⑶ A function that automatically translates a given sentence into Japanese

def to_japanese (sentence):
    from deep_translator import GoogleTranslator
    translated = GoogleTranslator(source='auto', target='ja').translate(sentence)
    return translated

print( to_japanese("I am a boy.") )
# 私は男の子です。

⑷ Translation of a whole Markdown document

⑸ A function that turns an arbitrary variable-length natural language sentence into a 384-dimensional vector, considering its meaning (cf. CELLama)

from langchain.embeddings.sentence_transformer import SentenceTransformerEmbeddings
import numpy as np
from scipy.sparse import csr_matrix
import pandas as pd
from sklearn.neighbors import NearestNeighbors
import torch
from torch.utils.data import DataLoader, TensorDataset
from xgboost import XGBClassifier

def sentences_to_embedding(sentences):
    embedding_function = SentenceTransformerEmbeddings(model_name="all-MiniLM-L6-v2")
    db = embedding_function.embed_documents(sentences)
    emb_res = np.asarray(db)
    return emb_res


sentences = []
sentences.append("What is the meaning of: obsolete")
sentences.append("What is the meaning of: old-fashioned")
sentences.append("What is the meaning of: demagogue")
emb_res = sentences_to_embedding(sentences)

2. LLM applications

⑴ ollama.ai (free): Llama2/3/4, Phi-3, Mistral, Gemma, etc

⑵ GroqChat (free): Mixtral, Llama3, Gemma

⑶ OpenRouter (charged): Can utilize ChatGPT API, etc.

⑷ English–Korean translation using Llama2 (free)

import ollama

def english_to_korean(sentence):
    content = 'Translate "' + sentence + '" to Korean. Output only the translated sentence.'
    response = ollama.chat(model='llama2', messages=[
      {
        'role': 'user',
        'content': content,
      },
    ])
    return response['message']['content']
    
sentence = "I am a boy."
english_to_korean(sentence)

⑸ Determine whether a given sentence contains a chemical formula

import ollama

def is_chemical_formula(sentence):
    content = 'Please determine if "' + sentence + '" contains a chemical formula or not. If it is correct, answer "sure"; otherwise, "no".'
    response = ollama.chat(model='llama2', messages=[
      {
        'role': 'user',
        'content': content,
      },
    ])
    return response['message']['content']
    
sentence = "NH2OH is an amine."
result = is_chemical_formula(sentence)
print(result)
print('sure' in result.lower())

sentence = "I am a boy."
result = is_chemical_formula(sentence)
print(result)
print('sure' in result.lower())

### Output ###
'''
The term "NH2OH" does contain a chemical formula, so the answer is "yes" or "sure".
True
The statement "I am a boy" does not contain any chemical formulas, so the answer is "no".
False
'''

⑹ Determine whether a given noun is a proper noun or a common noun

import ollama

def is_proper_noun(noun):
    content = 'Please determine if "' + noun + '" is a proper noun or common noun. If it is a proper noun, answer "proper"; otherwise, "common".'
    response = ollama.chat(model='llama2', messages=[
      {
        'role': 'user',
        'content': content,
      },
    ])
    return response['message']['content']
    
sentence = "Pencil"
result = is_proper_noun(sentence)
print(result)
print('proper' in result.lower())

sentence = "Feynman"
result = is_proper_noun(sentence)
print(result)
print('proper' in result.lower())

### Output ###
'''
"Pencil" is a common noun. Therefore, the answer is "common".
False

"Feynman" is a proper noun. Therefore, the answer is "proper".
True
'''

⑺ OpenRouter

def openrouter_chat(messages, max_retries=6, timeout=120):
    url = "https://openrouter.ai/api/v1/chat/completions"
    headers = {
        "Authorization": f"Bearer {OPENROUTER_API_KEY}",
        "Content-Type": "application/json",
    }
    payload = {
        "model": MODEL,
        "temperature": 0,
        "messages": messages,
    }

    last_err = None
    for attempt in range(max_retries):
        try:
            r = requests.post(url, headers=headers, json=payload, timeout=timeout)
            if r.status_code == 200:
                data = r.json()
                return data["choices"][0]["message"]["content"]
            if r.status_code in (429, 500, 502, 503, 504):
                time.sleep(2 ** attempt)
                continue
            r.raise_for_status()
        except Exception as e:
            last_err = e
            time.sleep(2 ** attempt)

    raise RuntimeError(f"OpenRouter request failed after retries: {last_err}")

openrouter_chat([
    {"role": "user", "content": "What is the largest state in US?"}
])