tasks/text.py

from fastapi import APIRouter
from datetime import datetime
from datasets import load_dataset
from sklearn.metrics import accuracy_score
import random
import os

from .utils.evaluation import TextEvaluationRequest
from .utils.emissions import tracker, clean_emissions_data, get_space_info

import tensorflow as tf
from huggingface_hub import hf_hub_download

from transformers import TFElectraForSequenceClassification, ElectraTokenizer, ElectraConfig
import numpy as np

router = APIRouter()

DESCRIPTION = "Electra_Base"
ROUTE = "/text"

# # Load model and tokenizer
# model_weights_path = hf_hub_download(repo_id="jennasparks/electra-tf", filename="tf_model.h5")
# model_config_path = hf_hub_download(repo_id="jennasparks/electra-tf", filename="config.json")

model_repo = "jennasparks/electra_tf"
config = ElectraConfig.from_pretrained(model_repo)
model = TFElectraForSequenceClassification.from_pretrained(model_repo)
tokenizer = ElectraTokenizer.from_pretrained("google/electra-base-discriminator")


@router.post(ROUTE, tags=["Text Task"], 
             description=DESCRIPTION)
async def evaluate_text(request: TextEvaluationRequest):
    """
    Evaluate text classification for climate disinformation detection.
    
    Current Model: Electra
    """
    # Get space info
    username, space_url = get_space_info()

    # Define the label mapping
    LABEL_MAPPING = {
        "0_not_relevant": 0,
        "1_not_happening": 1,
        "2_not_human": 2,
        "3_not_bad": 3,
        "4_solutions_harmful_unnecessary": 4,
        "5_science_unreliable": 5,
        "6_proponents_biased": 6,
        "7_fossil_fuels_needed": 7
    }

    # Load and prepare the dataset
    dataset = load_dataset(request.dataset_name, token=os.getenv("HF_TOKEN"))

    # Convert string labels to integers
    dataset = dataset.map(lambda x: {"label": LABEL_MAPPING[x["label"]]})

    # Split dataset
    test_dataset = dataset["test"]
    
    # Start tracking emissions
    tracker.start()
    tracker.start_task("inference")

    #--------------------------------------------------------------------------------------------
    # YOUR MODEL INFERENCE CODE HERE
    # Update the code below to replace the random baseline by your model inference within the inference pass where the energy consumption and emissions are tracked.
    #--------------------------------------------------------------------------------------------   

    #make predictions
    predictions = []
    
    for i in range(len(test_dataset["quote"])):
      encoded_input = tokenizer(test_dataset["quote"][i], truncation=True, padding=True, return_tensors="tf")
      outputs = model(encoded_input["input_ids"], attention_mask=encoded_input["attention_mask"], training=False)
      predictions.append(tf.argmax(outputs.logits, axis=1))
        
    # Get true labels
    true_labels = test_dataset["label"]

    #--------------------------------------------------------------------------------------------
    # YOUR MODEL INFERENCE STOPS HERE
    #--------------------------------------------------------------------------------------------   

    
    # Stop tracking emissions
    emissions_data = tracker.stop_task()
    
    # Calculate accuracy
    accuracy = accuracy_score(true_labels, predictions)
    
    # Prepare results dictionary
    results = {
        "username": username,
        "space_url": space_url,
        "submission_timestamp": datetime.now().isoformat(),
        "model_description": DESCRIPTION,
        "accuracy": float(accuracy),
        "energy_consumed_wh": emissions_data.energy_consumed * 1000,
        "emissions_gco2eq": emissions_data.emissions * 1000,
        "emissions_data": clean_emissions_data(emissions_data),
        "api_route": ROUTE,
        "dataset_config": {
            "dataset_name": request.dataset_name,
            "test_size": request.test_size,
            "test_seed": request.test_seed
        }
    }
    
    return results