llama-demo.py

#%%
import scraperwiki
import re
import requests
import json
import pandas as pd
import time


#%%

queryString = "* from comments"
queryResult = scraperwiki.sqlite.select(queryString)

results = []

llama_url = 'http://localhost:11434/api/generate'


for row in queryResult:
    start = time.time()
    comment = row['comment']
    query = f'In the following text someone will specify their favourite word. Please return the favourite word from their text in the following format - "Favourite word: Word" - with no other words in your response. Here is the text: {comment}'

    print(query)

    data = {
    "model": "llama2:13b",
    "prompt": query,
    "stream": False
    }
    r = requests.post(llama_url, data=json.dumps(data))
    word = r.json()['response']
    print(word)
    row = {"word":word, "count":1, "comment":row['comment']}
    results.append(row)
    end = time.time()
    print(end - start)
#%%
df = pd.DataFrame(results)
#%%
df.to_csv("llama-words.csv")

#%%
df['word'] = df['word'].str.strip()
#%%
#%%
df['word2'] = df['word'].str.split("Favourite word: ").str[-1]
#%%
df['word2'] = df['word2'].str.replace("#", "")
#%%
df['word2'] = df['word2'].str.lower()

#%%
sum_df = df[df['word_check'] != False]
blah = df[df['word_check'] == False]
summary = sum_df.groupby('word2').count()
#%%
summary.to_csv('summary.csv')

#%%

def checkWord(row):
    if row['word2'] in row['comment'].lower():
        return True
    else:
        return False

df['word_check'] = df.apply(checkWord, axis=1)    



#%%
df.to_csv("llama-words2.csv")
#%%
#%%
dg = df.groupby(['word']).count()
#%%
dg.to_csv('words.csv')

# %%