File size: 5,074 Bytes
e6d07cd
4d6e8c2
 
 
ece5856
2c8310a
1f08781
 
4d6e8c2
 
1a885c6
4d6e8c2
 
 
c339ecd
1c33274
7eb6153
2c8310a
 
4d6e8c2
2c8310a
 
 
 
e6d07cd
 
2c8310a
e6d07cd
 
 
 
 
 
85c5204
e6d07cd
 
 
 
2c8310a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6f0e9af
2c8310a
 
6f0e9af
c339ecd
 
 
 
2c8310a
c339ecd
6f0e9af
c339ecd
 
2c8310a
 
 
6f0e9af
c339ecd
c3f000b
2c8310a
c3f000b
 
2c8310a
c3f000b
 
 
b206095
c339ecd
b206095
 
 
 
 
 
6f0e9af
1f08781
2c8310a
1f08781
c3f000b
 
b206095
 
1f08781
 
 
2c8310a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e6d07cd
2c8310a
 
 
6f0e9af
2c8310a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
from fastapi import APIRouter
from datetime import datetime
from datasets import load_dataset
from sklearn.metrics import accuracy_score
import torch
from transformers import AutoTokenizer, AutoModelForSequenceClassification
from torch.utils.data import DataLoader
from transformers import DataCollatorWithPadding

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

router = APIRouter()

DESCRIPTION = "Climate Guard Toxic Agent is a ModernBERT for Climate Disinformation Detection"
ROUTE = "/text"

@router.post(ROUTE, tags=["Text Task"], 
             description=DESCRIPTION)
async def evaluate_text(request: TextEvaluationRequest):
    """
    Evaluate text classification for climate disinformation detection using ModernBERT.
    """
    # 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"]]})

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

    #--------------------------------------------------------------------------------------------
    # MODEL INFERENCE CODE
    #--------------------------------------------------------------------------------------------   
    
    try:
        # Set device
        device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
        
        # Model and tokenizer paths
        path_model = 'Tonic/climate-guard-toxic-agent'
        path_tokenizer = "answerdotai/ModernBERT-base"
        
        # Initialize tokenizer
        tokenizer = AutoTokenizer.from_pretrained(path_tokenizer)
        
        # Initialize model
        model = AutoModelForSequenceClassification.from_pretrained(path_model).half().to(device)
        
        # Set model to evaluation mode
        model.eval()
        
        # Preprocess function
        def preprocess_function(examples):
            return tokenizer(
                examples["quote"],
                truncation=True,
                return_tensors=None
            )
        
        # Tokenize dataset
        tokenized_test = test_dataset.map(
            preprocess_function,
            batched=True,
            remove_columns=test_dataset.column_names
        )
        
        # Set format for pytorch
        tokenized_test.set_format("torch")
        
        # Create DataLoader
        data_collator = DataCollatorWithPadding(tokenizer=tokenizer)
        test_loader = DataLoader(
            tokenized_test,
            batch_size=16,
            collate_fn=data_collator,
            shuffle=False
        )
        
        # Get predictions
        predictions = []
        with torch.no_grad():
            for batch in test_loader:
                # Move batch to device
                input_ids = batch['input_ids'].to(device)
                attention_mask = batch['attention_mask'].to(device)
                
                # Get model outputs
                outputs = model(input_ids=input_ids, attention_mask=attention_mask)
                preds = torch.argmax(outputs.logits, dim=-1)
                
                # Add batch predictions to list
                predictions.extend(preds.cpu().numpy().tolist())
                
        # Clean up GPU memory
        if torch.cuda.is_available():
            torch.cuda.empty_cache()
            
    except Exception as e:
        print(f"Error during model inference: {str(e)}")
        raise

    #--------------------------------------------------------------------------------------------
    # MODEL INFERENCE ENDS HERE
    #--------------------------------------------------------------------------------------------   

    # Stop tracking emissions
    emissions_data = tracker.stop_task()
    
    # Calculate accuracy
    accuracy = accuracy_score(test_dataset["label"], 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