SITCC-T5-Classifier Model Card
Model Description
The SITCC-T5-Classifier model is a fine-tuned version of the google/flan-t5-base model. It has been specifically trained to process IT ticket descriptions and extract the request/issue and the software/system that the ticket is about. The model was fine-tuned using 5716 synthetically generated input/output pairs generated with OpenAI GPT-4 Turbo.
Model Details
- Base Model: google/flan-t5-base
- Fine-tuning Data: 5716 synthetic IT ticket description pairs generated by OpenAI GPT-4 Turbo
Intended Use
The SITCC-T5-Classifier model is designed to be used for IT ticket classification and information extraction tasks. It can be used to automatically identify the request/issue and the software/system mentioned in an IT ticket description.
Limitations and Known Issues
- The model's performance may vary depending on the quality and diversity of the input IT ticket descriptions.
- The model may struggle with understanding complex or ambiguous ticket descriptions.
- The model may not perform well on ticket descriptions that are significantly different from the training data.
Example Usage
This example is running on cpu
import re
import pandas as pd
from transformers import T5Tokenizer, T5ForConditionalGeneration
from time import perf_counter
class SITCC_T5_Classifier:
"""
A class for classifying text using the SITCC T5 model.
Attributes:
tokenizer (T5Tokenizer): The tokenizer for the T5 model.
model (T5ForConditionalGeneration): The T5 model for classification.
"""
def __init__(self):
# Load the tokenizer and model from the fine-tuned model directory
self.tokenizer = T5Tokenizer.from_pretrained("KameronB/sitcc-t5-classifier")
self.model = T5ForConditionalGeneration.from_pretrained("KameronB/sitcc-t5-classifier", device_map="cpu")
def process_response(self, response:str) -> dict:
"""
Process the response and extract the software/system and issue/request.
Args:
response (str): The response text.
Returns:
dict: A dictionary containing the software/system and issue/request.
"""
# handle null input
if not response:
print("Error: Response is empty")
return {
"Software/System": "Unknown",
"Issue/Request": "Unknown"
}
matches = re.search(r'Software/System: (.*) Issue/Request: (.*)</s>', response, re.DOTALL)
# handle improperly formatted responses
# TODO: Provide more robust improper format handing
if not matches or len(matches.groups()) < 2:
print(f"Error: Response format does not match expected format \"Software/System: (.*) Issue/Request: (.*)</s>\"\nResponse: \"{response}\" ")
return {
"Software/System": "Unknown",
"Issue/Request": "Unknown"
}
return {
"Software/System": matches.group(1),
"Issue/Request": matches.group(2)
}
def classify_entry(self, entry:str, max_new_tokens=60) -> dict:
"""
Classify the input text and return the classification results.
Args:
entry (str): The input text to be classified.
max_new_tokens (int): The maximum number of tokens to generate.
Returns:
dict: The classification results.
"""
# Tokenize the input text
input_ids = self.tokenizer(entry, return_tensors="pt").input_ids.to("cpu")
# Generate the output text
outputs = self.model.generate(input_ids, max_new_tokens=max_new_tokens)
# Decode and return the output text
return self.process_response(self.tokenizer.decode(outputs[0]))
# Create the SITCC T5 Classifier wrapper class for the fine-tuned T5 model
sitcc_t5 = SITCC_T5_Classifier()
# Define the input text
input_text = [
"The customer is getting the following error when using rSATS:\nERROR: 'Failed to connect'. \nI have tried restarting the application and the computer, but the issue persists. \nEscalating to Team",
"The customer is experiencing issues with their network connectivity, which is causing slow internet speeds and frequent disconnections.",
"The customer is unable to access the shared drive on the network. They receive an error message stating 'Network path not found'. \nEscalating to Network Team",
"The customer is unable to print from their computer. They have checked the printer connections and restarted the printer, but the issue persists. \nEscalating to Printer Support Team",
]
# measure the time performance of the model
start = perf_counter()
for i in range(len(input_text)):
# Classify the input text
print(sitcc_t5.classify_entry(input_text[i]))
# measure the time performance of the model
end = perf_counter()
print(f"Time taken: {end - start} seconds")
- Downloads last month
- 4
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social
visibility and check back later, or deploy to Inference Endpoints (dedicated)
instead.