app.py

# import gradio as gr
# from peft import PeftModel, PeftConfig
# from transformers import AutoModelForSequenceClassification, AutoTokenizer
# import torch
# import json

# # Load id2label
# with open('id2label.json', 'r') as f:
#     id2label = json.load(f)

# # Load label2id
# with open('label2id.json', 'r') as f:
#     label2id = json.load(f)

# print("ID2LABEL", id2label)
# print("LABEL2ID", label2id)

# # Download model from end-point
# MODEL ="xlm-roberta-large"
# peft_model_id = "JAdeojo/xlm-roberta-large-lora-consumer-complaints-cfpb"
# config = PeftConfig.from_pretrained(peft_model_id)
# inference_model = AutoModelForSequenceClassification.from_pretrained(
#     MODEL, 
#     num_labels=len(id2label), 
#     id2label=id2label, label2id=label2id, 
#     # ignore_mismatched_sizes=True
# )
# tokenizer = AutoTokenizer.from_pretrained(MODEL)
# model = PeftModel.from_pretrained(inference_model, "JAdeojo/xlm-roberta-large-lora-consumer-complaints-cfpb")

# # run inference
# def classify_complaint(Complaints, id2label):
#     inputs = tokenizer(Complaints, return_tensors="pt")
#     with torch.no_grad():
#         logits = model(**inputs).logits
#         # tokens = inputs.tokens()
#     predictions = torch.argmax(logits, dim=-1)
#     predicted_label = predictions.item()
#     predicted_class = id2label[predicted_label]
#     return predicted_class

# demo = gr.Interface(fn=classify_complaint, inputs="text", outputs="text")
    
# demo.launch() 


import gradio as gr
from peft import PeftModel, PeftConfig
from transformers import AutoModelForSequenceClassification, AutoTokenizer
import torch
import json

# Load id2label
with open('id2label.json', 'r') as f:
    id2label = json.load(f)

# Load label2id
with open('label2id.json', 'r') as f:
    label2id = json.load(f)

print("ID2LABEL", id2label)
print("LABEL2ID", label2id)

# Download model from end-point
MODEL ="xlm-roberta-large"
peft_model_id = "JAdeojo/xlm-roberta-large-lora-consumer-complaints-cfpb_49k"
config = PeftConfig.from_pretrained(peft_model_id)
inference_model = AutoModelForSequenceClassification.from_pretrained(
    MODEL, 
    num_labels=len(id2label), 
    id2label=id2label, label2id=label2id, 
    # ignore_mismatched_sizes=True
)
tokenizer = AutoTokenizer.from_pretrained(MODEL)
model = PeftModel.from_pretrained(inference_model, peft_model_id)

# run inference
def classify_complaint(Complaints):
    inputs = tokenizer(Complaints, return_tensors="pt")
    with torch.no_grad():
        logits = model(**inputs).logits
    predictions = torch.argmax(logits, dim=-1)
    predicted_label = predictions.item()
    predicted_class = id2label[str(predicted_label)]  # Make sure to access id2label with a string key
    return predicted_class

# demo = gr.Interface(fn=classify_complaint, inputs=("text", "Enter the financial complaint:"), outputs=("text", "Complaint Category"))
# demo.launch() 
input_component = gr.inputs.Textbox(label="Enter the financial complaint:")
output_component = gr.outputs.Textbox(label="Complaint Category")

demo = gr.Interface(fn=classify_complaint, inputs=input_component, outputs=output_component)
demo.launch()