tasks/text.py

from datetime import datetime
import random

import numpy as np
import torch
from torch import nn
from torch.utils.data import DataLoader, Dataset
from transformers import AutoModel, AutoModelForSequenceClassification, AutoTokenizer
from fastapi import APIRouter
from datasets import load_dataset
from sklearn.metrics import accuracy_score
from skops.io import load

from .utils.evaluation import TextEvaluationRequest
from .utils.emissions import tracker, clean_emissions_data, get_space_info
from .utils.text_preprocessor import preprocess
from accelerate.test_utils.testing import get_backend
from .custom_classifiers import SentenceBERTClassifier, MoEClassifier

router = APIRouter()

DESCRIPTION = "Random Baseline"
ROUTE = "/text"

MODEL_DESCRIPTIONS = {
    "baseline": "random baseline", # Baseline
    "tfidf_xgb": "TF-IDF vectorizer and XGBoost classifier", # Submitted
    "bert_base_pruned": "Pruned BERT base model", # Submitted
    # 'climate_bert_pruned': "Fine-tuned and pruned DistilRoBERTa pre-trained on climate texts", # Not working
    "sbert_distilroberta": "Fine-tuned sentence transformer DistilRoBERTa", # working, not submitted
    "embedding_moe": "Mixture of expert classifier with DistilBERT Embeddings" # working, not submitted
}


def baseline_model(dataset_length: int):
    # Make random predictions (placeholder for actual model inference)
    predictions = [random.randint(0, 7) for _ in range(dataset_length)]

    return predictions

def tree_classifier(test_dataset: dict, model: str):
    print("Starting tree model run")

    texts = test_dataset["quote"]

    texts = preprocess(texts)

    model_path = f"tasks/text_models/{model}.skops"

    model = load(model_path, 
                 trusted=[
                     'scipy.sparse._csr.csr_matrix',
                     'xgboost.core.Booster',
                     'xgboost.sklearn.XGBClassifier'])

    predictions = model.predict(texts)
    print("Finished tree model run")


    return predictions

class TextDataset(Dataset):
    def __init__(self, texts, tokenizer, max_length=512):
        self.texts = texts
        self.tokenized_texts = tokenizer(
            texts,
            truncation=True,
            padding=True,
            max_length=max_length,
            return_tensors="pt",
        )

    def __getitem__(self, idx):
        item = {key: val[idx] for key, val in self.tokenized_texts.items()}
        return item

    def __len__(self) -> int:
        return len(self.texts)


def bert_classifier(test_dataset: dict, model: str):
    print("Starting BERT model run")
    texts = test_dataset["quote"]

    model_repo = f"theterryzhang/frugal_ai_{model}"

    tokenizer = AutoTokenizer.from_pretrained(model_repo)

    if model in ["bert_base_pruned"]:
        model = AutoModelForSequenceClassification.from_pretrained(model_repo)
    elif model in ["sbert_distilroberta"]:
        model = SentenceBERTClassifier.from_pretrained(model_repo)
    else:
        raise(ValueError)

    device, _, _ = get_backend()

    model = model.to(device)

    # Prepare dataset
    dataset = TextDataset(texts, tokenizer=tokenizer)
    dataloader = DataLoader(dataset, batch_size=32, shuffle=False)

    model.eval()
    with torch.no_grad():
        predictions = np.array([])
        for batch in dataloader:
            test_input_ids = batch["input_ids"].to(device)
            test_attention_mask = batch["attention_mask"].to(device)
            outputs = model(test_input_ids, test_attention_mask)
            p = torch.argmax(outputs.logits, dim=1)
            predictions = np.append(predictions, p.cpu().numpy())
    
    print("Finished BERT model run")
    
    return predictions

def moe_classifier(test_dataset: dict, model: str):
    print("Starting MoE run")

    device, _, _ = get_backend()

    texts = test_dataset["quote"]

    model_path = f"tasks/text_models/0131_MoE_final.pt"

    embedding_model = AutoModel.from_pretrained("sentence-transformers/all-distilroberta-v1")
    embedding_model = embedding_model.to(device)
    tokenizer = AutoTokenizer.from_pretrained("sentence-transformers/all-distilroberta-v1")

    dataset = TextDataset(texts, tokenizer=tokenizer, max_length=512)
    dataloader = DataLoader(dataset, batch_size=64, shuffle=False)

    model = MoEClassifier(3, 0.05)
    model.load_state_dict(torch.load(model_path))
    model = model.to(device)
    
    print("Starting MoE Classifier")

    model.eval()
    with torch.no_grad():
        predictions = np.array([])
        for batch in dataloader:
            input_ids = batch['input_ids'].to(device)
            attn_mask = batch['attention_mask'].to(device)
            embedding_outputs = embedding_model(input_ids, attn_mask)
            embeddings = embedding_outputs.last_hidden_state[:, 0, :]

            outputs = model(embeddings)
            p = torch.argmax(outputs, dim=1)
            predictions = np.append(predictions, p.cpu().numpy())

    print("Finished running MoE Classifier")

    return predictions

@router.post(ROUTE, tags=["Text Task"])
async def evaluate_text(request: TextEvaluationRequest,
                        model: str = "embedding_moe"):
    """
    Evaluate text classification for climate disinformation detection.
    
    Current Model: Random Baseline
    - Makes random predictions from the label space (0-7)
    - Used as a baseline for comparison
    """
    # 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)

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

    # Split dataset
    train_test = dataset["train"].train_test_split(test_size=request.test_size, seed=request.test_seed)
    test_dataset = train_test["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 random predictions (placeholder for actual model inference)
    true_labels = test_dataset["label"]

    if model == "baseline":
        predictions = baseline_model(len(true_labels))
    elif model == "tfidf_xgb":
        predictions = tree_classifier(test_dataset, model='xgb_pipeline')
    elif 'bert' in model:
        predictions = bert_classifier(test_dataset, model)
    elif 'moe' in model:
        predictions = moe_classifier(test_dataset, model)

    #--------------------------------------------------------------------------------------------
    # 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": MODEL_DESCRIPTIONS[model],
        "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