src/leaderboard/read_evals.py

import json
import os
from dataclasses import dataclass


from src.display.formatting import make_clickable_model
from src.display.utils import AutoEvalColumn, ModelType, Tasks, Precision, WeightType
from src.submission.check_validity import is_model_on_hub


@dataclass
class EvalResult:
    """Represents one full evaluation. Built from a single result file for a given run."""
    eval_name: str # org_model_precision (uid)
    full_model: str # org/model (path on hub)
    org: str 
    model: str
    revision: str # commit hash, "" if main
    results: dict
    precision: Precision = Precision.Unknown
    model_type: ModelType = ModelType.Unknown # Pretrained, fine tuned, ...
    weight_type: WeightType = WeightType.Original # Original or Adapter
    architecture: str = "Unknown" 
    license: str = "?"
    likes: int = 0
    num_params: int = 0
    date: str = "" # submission date of request file
    still_on_hub: bool = False

    @classmethod
    def init_from_json_file(self, json_filepath):
        """Inits the result from the specific model result file"""
        try:
            with open(json_filepath) as fp:
                data = json.load(fp)

            # Extract model information from the JSON data
            full_model_name = data.get('model')
            org_and_model = full_model_name.split("/", 1)
            org = org_and_model[0]
            model = org_and_model[1]
            
            # Extract other metadata
            precision_str = data.get('precision', 'Unknown')
            precision = Precision.from_str(precision_str)
            model_type = ModelType.from_str(data.get('model_type', 'Unknown'))
            weight_type = WeightType.from_str(data.get('weight_type', 'Original'))
            revision = data.get('revision', '')
            date = data.get('submitted_at', '')
            
            # Extract results and metadata
            results = data.get('results', {})
            license = data.get('license', '?')
            likes = data.get('likes', 0)
            num_params = data.get('params', 0)
            architecture = data.get('architecture', 'Unknown')
            
            # Check if the model is still on the hub
            still_on_hub, _, _ = is_model_on_hub(full_model_name, revision=revision)

            return EvalResult(
                eval_name=f"{org}_{model}_{precision.value}",
                full_model=full_model_name,
                org=org,
                model=model,
                revision=revision,
                results=results,
                precision=precision,
                model_type=model_type,
                weight_type=weight_type,
                architecture=architecture,
                license=license,
                likes=likes,
                num_params=num_params,
                date=date,
                still_on_hub=still_on_hub
            )
        except Exception as e:
            print(f"Error reading evaluation file {json_filepath}: {str(e)}")
            return None

    def to_dict(self):
        """Converts the Eval Result to a dict compatible with our dataframe display"""
        # Calculate the average score for the leaderboard
        scores = [v for k, v in self.results.items() if v is not None and k in [task.value.metric for task in Tasks]]
        average = sum(scores) / len(scores) if scores else 0

        AutoEvalColumnInstance = AutoEvalColumn()
        data_dict = {
            "eval_name": self.eval_name,
            AutoEvalColumnInstance.precision.name: self.precision.value.name,
            AutoEvalColumnInstance.model_type.name: self.model_type.value.name,
            AutoEvalColumnInstance.model_type_symbol.name: self.model_type.value.symbol,
            AutoEvalColumnInstance.weight_type.name: self.weight_type.value.name,
            AutoEvalColumnInstance.architecture.name: self.architecture,
            AutoEvalColumnInstance.model.name: make_clickable_model(self.full_model),
            AutoEvalColumnInstance.revision.name: self.revision,
            AutoEvalColumnInstance.average.name: average,
            AutoEvalColumnInstance.license.name: self.license,
            AutoEvalColumnInstance.likes.name: self.likes,
            AutoEvalColumnInstance.params.name: self.num_params,
            AutoEvalColumnInstance.still_on_hub.name: self.still_on_hub,
        }

        # Dynamically map metric values to their corresponding column names
        for task in Tasks:
            task_metric = task.value.metric
            task_col_name = task.value.col_name
            data_dict[task_col_name] = self.results.get(task_metric)
            
        return data_dict




def get_raw_eval_results(results_path: str, requests_path: str) -> list[EvalResult]:
    """From the path of the results folder root, extract all needed info for results"""
    model_result_filepaths = []
    # Recursively find all result files
    for root, _, files in os.walk(results_path):
        json_files = [f for f in files if f.endswith(".json")]
        for file in json_files:
            model_result_filepaths.append(os.path.join(root, file))

    eval_results = []
    for model_result_filepath in model_result_filepaths:
        try:
            eval_result = EvalResult.init_from_json_file(model_result_filepath)
            if eval_result is not None:
                eval_results.append(eval_result)
            else:
                print(f"Skipping invalid evaluation file: {model_result_filepath}")
        except Exception as e:
            print(f"Error processing evaluation file {model_result_filepath}: {str(e)}")
            continue
            
    return eval_results