File size: 9,038 Bytes
e7abd9e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f1e551a
 
 
e7abd9e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
893ecc7
1de2d20
 
e7abd9e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f1e551a
e7abd9e
 
 
 
 
 
f1e551a
e7abd9e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
from app.core.cache import cache_config
from datetime import datetime
from typing import List, Dict, Any
import datasets
from fastapi import HTTPException
import logging
from app.config.base import HF_ORGANIZATION
from app.utils.logging import LogFormatter

logger = logging.getLogger(__name__)

class LeaderboardService:
    def __init__(self):
        pass
        
    async def fetch_raw_data(self) -> List[Dict[str, Any]]:
        """Fetch raw leaderboard data from HuggingFace dataset"""
        try:
            logger.info(LogFormatter.section("FETCHING LEADERBOARD DATA"))
            logger.info(LogFormatter.info(f"Loading dataset from {HF_ORGANIZATION}/contents"))
            
            dataset = datasets.load_dataset(
                f"{HF_ORGANIZATION}/contents",
                cache_dir=cache_config.get_cache_path("datasets")
            )["train"]
            
            df = dataset.to_pandas()
            data = df.to_dict('records')
            
            stats = {
                "Total_Entries": len(data),
                "Dataset_Size": f"{df.memory_usage(deep=True).sum() / 1024 / 1024:.1f}MB"
            }
            for line in LogFormatter.stats(stats, "Dataset Statistics"):
                logger.info(line)
                
            return data
            
        except Exception as e:
            logger.error(LogFormatter.error("Failed to fetch leaderboard data", e))
            raise HTTPException(status_code=500, detail=str(e))

    async def get_formatted_data(self) -> List[Dict[str, Any]]:
        """Get formatted leaderboard data"""
        try:
            logger.info(LogFormatter.section("FORMATTING LEADERBOARD DATA"))
            
            raw_data = await self.fetch_raw_data()
            formatted_data = []
            type_counts = {}
            error_count = 0
            
            # Initialize progress tracking
            total_items = len(raw_data)
            logger.info(LogFormatter.info(f"Processing {total_items:,} entries..."))
            
            for i, item in enumerate(raw_data, 1):
                try:
                    formatted_item = await self.transform_data(item)
                    formatted_data.append(formatted_item)
                    
                    # Count model types
                    model_type = formatted_item["model"]["type"]
                    type_counts[model_type] = type_counts.get(model_type, 0) + 1
                    
                except Exception as e:
                    error_count += 1
                    logger.error(LogFormatter.error(f"Failed to format entry {i}/{total_items}", e))
                    continue
                
                # Log progress every 10%
                if i % max(1, total_items // 10) == 0:
                    progress = (i / total_items) * 100
                    logger.info(LogFormatter.info(f"Progress: {LogFormatter.progress_bar(i, total_items)}"))
            
            # Log final statistics
            stats = {
                "Total_Processed": total_items,
                "Successful": len(formatted_data),
                "Failed": error_count
            }
            logger.info(LogFormatter.section("PROCESSING SUMMARY"))
            for line in LogFormatter.stats(stats, "Processing Statistics"):
                logger.info(line)
            
            # Log model type distribution
            type_stats = {f"Type_{k}": v for k, v in type_counts.items()}
            logger.info(LogFormatter.subsection("MODEL TYPE DISTRIBUTION"))
            for line in LogFormatter.stats(type_stats):
                logger.info(line)
                
            return formatted_data
            
        except Exception as e:
            logger.error(LogFormatter.error("Failed to format leaderboard data", e))
            raise HTTPException(status_code=500, detail=str(e))

    async def transform_data(self, data: Dict[str, Any]) -> Dict[str, Any]:
        """Transform raw data into the format expected by the frontend"""
        try:
            # Extract model name for logging
            model_name = data.get("fullname", "Unknown")
            logger.debug(LogFormatter.info(f"Transforming data for model: {model_name}"))
            
            # Create unique ID combining model name, precision, sha and chat template status
            unique_id = f"{data.get('fullname', 'Unknown')}_{data.get('Precision', 'Unknown')}_{data.get('Model sha', 'Unknown')}_{str(data.get('Chat Template', False))}"
            
            evaluations = {
                "ifeval": {
                    "name": "IFEval",
                    "value": data.get("IFEval Raw", 0),
                    "normalized_score": data.get("IFEval", 0)
                },
                "bbh": {
                    "name": "BBH",
                    "value": data.get("BBH Raw", 0),
                    "normalized_score": data.get("BBH", 0)
                },
                "math": {
                    "name": "MATH Level 5",
                    "value": data.get("MATH Lvl 5 Raw", 0),
                    "normalized_score": data.get("MATH Lvl 5", 0)
                },
                "gpqa": {
                    "name": "GPQA",
                    "value": data.get("GPQA Raw", 0),
                    "normalized_score": data.get("GPQA", 0)
                },
                "musr": {
                    "name": "MUSR",
                    "value": data.get("MUSR Raw", 0),
                    "normalized_score": data.get("MUSR", 0)
                },
                "mmlu_pro": {
                    "name": "MMLU-PRO",
                    "value": data.get("MMLU-PRO Raw", 0),
                    "normalized_score": data.get("MMLU-PRO", 0)
                }
            }

            features = {
                "is_not_available_on_hub": data.get("Available on the hub", False),
                "is_merged": data.get("Merged", False),
                "is_moe": data.get("MoE", False),
                "is_flagged": data.get("Flagged", False),
                "is_highlighted_by_maintainer": data.get("Official Providers", False)
            }

            metadata = {
                "upload_date": data.get("Upload To Hub Date"),
                "submission_date": data.get("Submission Date"),
                "generation": data.get("Generation"),
                "base_model": data.get("Base Model"),
                "hub_license": data.get("Hub License"),
                "hub_hearts": data.get("Hub ❤️"),
                "params_billions": data.get("#Params (B)"),
                "co2_cost": data.get("CO₂ cost (kg)", 0)
            }

            # Clean model type by removing emojis if present
            original_type = data.get("Type", "")
            model_type = original_type.lower().strip()
            
            # Remove emojis and parentheses
            if "(" in model_type:
                model_type = model_type.split("(")[0].strip()
            model_type = ''.join(c for c in model_type if not c in '🔶🟢🟩💬🤝🌸 ')
                
            # Map old model types to new ones
            model_type_mapping = {
                "fine-tuned": "fined-tuned-on-domain-specific-dataset",
                "fine tuned": "fined-tuned-on-domain-specific-dataset",
                "finetuned": "fined-tuned-on-domain-specific-dataset",
                "fine_tuned": "fined-tuned-on-domain-specific-dataset",
                "ft": "fined-tuned-on-domain-specific-dataset",
                "finetuning": "fined-tuned-on-domain-specific-dataset",
                "fine tuning": "fined-tuned-on-domain-specific-dataset",
                "fine-tuning": "fined-tuned-on-domain-specific-dataset"
            }

            mapped_type = model_type_mapping.get(model_type.lower().strip(), model_type)
            
            if mapped_type != model_type:
                logger.debug(LogFormatter.info(f"Model type mapped: {original_type} -> {mapped_type}"))
            
            transformed_data = {
                "id": unique_id,
                "model": {
                    "name": data.get("fullname"),
                    "sha": data.get("Model sha"),
                    "precision": data.get("Precision"),
                    "type": mapped_type,
                    "weight_type": data.get("Weight type"),
                    "architecture": data.get("Architecture"),
                    "average_score": data.get("Average ⬆️"),
                    "has_chat_template": data.get("Chat Template", False)
                },
                "evaluations": evaluations,
                "features": features,
                "metadata": metadata
            }
            
            logger.debug(LogFormatter.success(f"Successfully transformed data for {model_name}"))
            return transformed_data
            
        except Exception as e:
            logger.error(LogFormatter.error(f"Failed to transform data for {data.get('fullname', 'Unknown')}", e))
            raise