Update utils.py

utils.py

@@ -6,40 +6,15 @@ import io
 import pandas as pd
 import os
 import torch
-from transformers import AutoTokenizer, AutoModelForSequenceClassification
-from transformers import pipeline
-from openai import OpenAI
-from groq import Groq
-import time 
+import time
 import json
 from openai import OpenAI
 
 openai_key = "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"
-print(openai_key)
 client = OpenAI(api_key = openai_key)
 desc = pd.read_excel('Descriptor.xlsx',header = None)
 desc_list  = desc.iloc[:,0].to_list()
 
-def callAzure(prompt,text):
-    
-    url = "https://Meta-Llama-3-70B-Instruct-fkqip-serverless.eastus2.inference.ai.azure.com"
-    api_key = "o5yaLhTIvg0s5zuYVInBpyneEZO8oonY"
-    client = OpenAI(base_url=url, api_key=api_key)
-    msg = "{} {}".format(prompt, text)
-    
-    response = client.chat.completions.create(
-        messages=[
-            {
-                "role": "user",
-                "content": msg,
-            }
-        ],
-        model="azureai",
-        max_tokens = 1000
-        ) 
-    
-    return response.choices[0].message.content
-    
 def filter(input_json):
 
   sym = pd.read_excel('symbol.xlsx',header = None)
@@ -67,16 +42,16 @@ def filter(input_json):
   return [1, document]
 
 def summary(input_json):
-    
+
     prompt = pd.read_excel('DescriptorPrompt.xlsx')
     promptShort  = prompt.iloc[:,1].to_list()
-    promptLong  = prompt.iloc[:,2].to_list() 
-    
+    promptLong  = prompt.iloc[:,2].to_list()
+
     output = {}
     filtering_results = filter(input_json)
     if filtering_results[0] == 0:
-        #return 0
-        return filtering_results[1]
+        return 0
+        #return filtering_results[1]
 
     id = desc_list.index(input_json['Descriptor'])
     long_text = filtering_results[1]
@@ -84,58 +59,48 @@ def summary(input_json):
     long_text = long_text.rstrip()
 
     long_text = long_text[:6000]
-    
+
     url = 'https://www.bseindia.com/xml-data/corpfiling/AttachLive/'+ input_json['FileURL'].split('Pname=')[-1]
 
     output["Link to BSE website"] = url
 
     output["Date of time of receiving data from BSE"] = input_json["newsdate"] + "Z"
-    
-    output["Stock Ticker"] = input_json['symbol']
-
-    answer = callAzure("You are an financial expert. Wherever possible, mention the name of the company " + promptShort[id] + " Do not exceed over 400 characters", long_text) 
-    try:
-        idx = answer.index("\n")
-    except:
-        idx = -2
-    output['Short Summary'] = answer[idx+2:]
-
-    answer = callAzure("Make sure the following summary of a news article is not more than 80 words. Rewrite it and make it below 80 words ", output['Short Summary'])
-    try:
-        idx = answer.index("\n")
-    except:
-        idx = -2
-    output['Short Summary'] = answer[idx+2:]
-
-    prompt = "Provide the main topic of the news article strictly as a tag, using only one or two words, with only the first word capitalized and the rest in lowercase. No additional text or explanation."
-    answer = callAzure(prompt, output['Short Summary'])
-    output['Tag'] = answer
 
-    prompt = "Generate a precise headline for the news article that includes the name of the company. Be very careful about correctly representing any financial figures mentioned in lakhs and crores. Provide only the headline, with no additional text or explanation."
-    answer = callAzure(prompt, output['Short Summary'])
-    output['Headline'] = answer
+    output["Stock Ticker"] = input_json['symbol']
 
     utc_now = datetime.datetime.utcnow()
     ist_now = utc_now.astimezone(datetime.timezone(datetime.timedelta(hours=5, minutes=30)))
-    
+
     Date = ist_now.strftime("%Y-%m-%d")
     time = ist_now.strftime("%X")
     output['Date and time of data delivery from Skylark'] = Date+"T"+time+"Z"
-    
-    prompt = "Answer in one word the sentiment of this News out of Positive, Negative or Neutral {}"
-    output['Sentiment'] = callAzure(prompt, output['Short Summary'])
+
+    prompt = """
+    Return the output in json format. This is the financial article {}
+    Following are the keys of the json.
+    1. Short Summary - {}  Do not exceed over 400 characters. Make sure it is no more than 80 words
+    2. Tag - Provide the main topic of the news article strictly as a tag, using only one or two words, with only the first word capitalized and the rest in lowercase. No additional text or explanation.
+    3. Headline - Generate a precise headline for the news article that includes the name of the company. Be very careful about correctly representing any financial figures mentioned in lakhs and crores. Provide only the headline, with no additional text or explanation.
+    4. Sentiment - Answer in one word the sentiment of this News out of Positive, Negative or Neutral.
+    5. Long summary - {} Write the summary in max 500 words.
+    """.format(long_text,promptShort[id],promptLong[id])
 
     completion = client.chat.completions.create(
     model="gpt-4o",
     messages=[
-    {"role": "system", "content": "You are a financial expert. Help the client with summarizing the financial newsletter. Write the summary in max 500 words. Do not truncate."},
-    {"role": "user", "content": "{} {}".format(promptLong[id], long_text)}
+    {"role": "system", "content": "You are a financial expert. Help the client with summarizing the financial newsletter. Do not truncate."},
+    {"role": "user", "content": "{}".format(prompt)}
     ],
     temperature=0,
     max_tokens=4000,
     )
 
-    output['Long summary'] = completion.choices[0].message.content
+    answer = json.loads(completion.choices[0].message.content[8:-3])
+    output['Short Summary'] = answer['Short Summary']
+    output['Tag'] = answer['Tag']
+    output['Headline'] = answer['Headline']
+    output['Sentiment'] = answer['Sentiment']
+    output['Long summary'] = answer['Long summary']
 
     # response = client.images.generate(
     # model="dall-e-3",