text-ypesk / tasks /text.py
ypesk's picture
Update tasks/text.py
3a019fe verified
raw
history blame
7.37 kB
from fastapi import APIRouter
from datetime import datetime
from datasets import load_dataset
from sklearn.metrics import accuracy_score
import random
import numpy as np
from huggingface_hub import PyTorchModelHubMixin
from tqdm import tqdm, trange
from sentence_transformers import SentenceTransformer
import torch
import torch.nn as nn
from torch.utils.data import TensorDataset, DataLoader, RandomSampler, SequentialSampler
from transformers import BertForPreTraining, BertModel, AutoTokenizer, BertForSequenceClassification, RobertaForSequenceClassification
from .utils.evaluation import TextEvaluationRequest
from .utils.emissions import tracker, clean_emissions_data, get_space_info
router = APIRouter()
DESCRIPTION = "First Baseline"
ROUTE = "/text"
MODEL = "mlp" #mlp, ct, modern
class ConspiracyClassification(
nn.Module,
PyTorchModelHubMixin,
# optionally, you can add metadata which gets pushed to the model card
):
def __init__(self, num_classes):
super().__init__()
self.h1 = nn.Linear(384, 100)
self.h2 = nn.Linear(100, 100)
self.h3 = nn.Linear(100, 100)
self.h4 = nn.Linear(100, 50)
self.h5 = nn.Linear(50, num_classes)
self.dropout = nn.Dropout(0.2)
self.activation = nn.ReLU()
def forward(self, input_texts):
outputs = self.h1(input_texts)
outputs = self.activation(outputs)
outputs = self.dropout(outputs)
outputs = self.h2(outputs)
outputs = self.activation(outputs)
outputs = self.dropout(outputs)
outputs = self.h3(outputs)
outputs = self.activation(outputs)
outputs = self.dropout(outputs)
outputs = self.h4(outputs)
outputs = self.activation(outputs)
outputs = self.dropout(outputs)
outputs = self.h5(outputs)
return outputs
class CovidTwitterBertClassifier(
nn.Module,
PyTorchModelHubMixin,
# optionally, you can add metadata which gets pushed to the model card
):
def __init__(self, num_classes):
super().__init__()
self.n_classes = num_classes
self.bert = BertForPreTraining.from_pretrained('digitalepidemiologylab/covid-twitter-bert-v2')
self.bert.cls.seq_relationship = nn.Linear(1024, num_classes)
self.sigmoid = nn.Sigmoid()
def forward(self, input_ids, token_type_ids, input_mask):
outputs = self.bert(input_ids = input_ids, token_type_ids = token_type_ids, attention_mask = input_mask)
logits = outputs[1]
return logits
@router.post(ROUTE, tags=["Text Task"],
description=DESCRIPTION)
async def evaluate_text(request: TextEvaluationRequest):
"""
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"]
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.
#--------------------------------------------------------------------------------------------
if MODEL =="mlp":
model = ConspiracyClassification.from_pretrained("ypesk/frugal-ai-mlp-baseline")
emb_model = SentenceTransformer("paraphrase-MiniLM-L3-v2")
batch_size = 6
test_texts = torch.Tensor(emb_model.encode([t['quote'] for t in test_dataset]))
test_data = TensorDataset(test_texts)
test_sampler = SequentialSampler(test_data)
test_dataloader = DataLoader(test_data, sampler=test_sampler, batch_size=batch_size)
elif MODEL == "ct":
model = CovidTwitterBertClassifier.from_pretrained("ypesk/ct-baseline")
tokenizer = AutoTokenizer.from_pretrained('digitalepidemiologylab/covid-twitter-bert')
test_texts = [t['quote'] for t in test_dataset]
MAX_LEN = 256 #1024 # < m some tweets will be truncated
tokenized_test = tokenizer(test_texts, max_length=MAX_LEN, padding='max_length', truncation=True)
test_input_ids, test_token_type_ids, test_attention_mask = tokenized_test['input_ids'], tokenized_test['token_type_ids'], tokenized_test['attention_mask']
test_token_type_ids = torch.tensor(test_token_type_ids)
test_input_ids = torch.tensor(test_input_ids)
test_attention_mask = torch.tensor(test_attention_mask)
batch_size = 12 #
test_data = TensorDataset(test_input_ids, test_attention_mask, test_token_type_ids)
test_sampler = SequentialSampler(test_data)
test_dataloader = DataLoader(test_data, sampler=test_sampler, batch_size=batch_size)
model.eval()
predictions = []
for batch in tqdm(test_dataloader):
b_input_ids, b_input_mask, b_token_type_ids = batch
with torch.no_grad():
logits = model(b_input_ids, b_token_type_ids, b_input_mask)
logits = logits.detach().cpu().numpy()
predictions.extend(logits.argmax(1))
true_labels = test_dataset["label"]
# Make random predictions (placeholder for actual model inference)
#true_labels = test_dataset["label"]
#predictions = [random.randint(0, 7) for _ in range(len(true_labels))]
#--------------------------------------------------------------------------------------------
# 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