src/submission/check_validity.py

import json
import os
import re
from collections import defaultdict
from datetime import datetime, timedelta, timezone
import logging

import huggingface_hub
from huggingface_hub import ModelCard, HfApi, hf_hub_download
from huggingface_hub.hf_api import ModelInfo
from transformers import AutoConfig
from transformers.models.auto.tokenization_auto import tokenizer_class_from_name, get_tokenizer_config

logger = logging.getLogger(__name__)
logger.setLevel(logging.INFO)  # Only show INFO and above, hide DEBUG messages

def check_model_card(repo_id: str) -> tuple[bool, str]:
    """Checks if the model card and license exist and have been filled"""
    try:
        card = ModelCard.load(repo_id)
    except huggingface_hub.utils.EntryNotFoundError:
        return False, "Please add a model card to your model to explain how you trained/fine-tuned it."

    # Enforce license metadata
    if card.data.license is None:
        if not ("license_name" in card.data and "license_link" in card.data):
            return False, (
                "License not found. Please add a license to your model card using the `license` metadata or a"
                " `license_name`/`license_link` pair."
            )

    # Enforce card content
    if len(card.text) < 200:
        return False, "Please add a description to your model card, it is too short."

    return True, ""


def is_model_on_hub(model_name: str, revision: str, token: str = None, trust_remote_code=False, test_tokenizer=False) -> tuple[bool, str]:
    """Makes sure the model is on the hub, and uses a valid configuration (in the latest transformers version)"""
    try:
        config = AutoConfig.from_pretrained(model_name, revision=revision, trust_remote_code=trust_remote_code, token=token)
        if test_tokenizer:
            tokenizer_config = get_tokenizer_config(model_name) 
            if tokenizer_config is not None:
                tokenizer_class_candidate = tokenizer_config.get("tokenizer_class", None)
            else:
                tokenizer_class_candidate = config.tokenizer_class 


            tokenizer_class = tokenizer_class_from_name(tokenizer_class_candidate)
            if tokenizer_class is None:
                return (
                    False,
                    f"uses {tokenizer_class_candidate}, which is not in a transformers release, therefore not supported at the moment.",
                    None
                )
        return True, None, config

    except ValueError:
        return (
            False,
            "needs to be launched with `trust_remote_code=True`. For safety reason, we do not allow these models to be automatically submitted to the leaderboard.",
            None
        )

    except Exception as e:
        return False, "was not found on hub!", None


def get_model_size(model_info: ModelInfo, precision: str):
    """Gets the model size from the configuration, or the model name if the configuration does not contain the information."""
    try:
        model_size = round(model_info.safetensors["total"] / 1e9, 3)
    except (AttributeError, TypeError):
        return 0  # Unknown model sizes are indicated as 0, see NUMERIC_INTERVALS in app.py

    size_factor = 8 if (precision == "GPTQ" or "gptq" in model_info.modelId.lower()) else 1
    model_size = size_factor * model_size
    return model_size

def get_model_arch(model_info: ModelInfo):
    """Gets the model architecture from the configuration"""
    return model_info.config.get("architectures", "Unknown")

def already_submitted_models(requested_models_dir: str) -> set[str]:
    depth = 1
    file_names = []
    users_to_submission_dates = defaultdict(list)

    for root, _, files in os.walk(requested_models_dir):
        current_depth = root.count(os.sep) - requested_models_dir.count(os.sep)
        if current_depth == depth:
            for file in files:
                if not file.endswith(".json"):
                    continue
                with open(os.path.join(root, file), "r") as f:
                    info = json.load(f)
                    file_names.append(f"{info['model']}_{info['revision']}_{info['precision']}")

                    # Select organisation
                    if info["model"].count("/") == 0 or "submitted_time" not in info:
                        continue
                    organisation, _ = info["model"].split("/")
                    users_to_submission_dates[organisation].append(info["submitted_time"])

    return set(file_names), users_to_submission_dates

def check_adapter_config_and_base_model(model_name: str, base_model: str, token: str = None) -> tuple[bool, str]:
    """
    Checks if the model exists on HuggingFace and is accessible.
    
    Args:
        model_name: Name of the model to check
        base_model: Expected base model name (not used anymore)
        token: HuggingFace API token (optional)
        
    Returns:
        Tuple[bool, str]: A tuple containing:
            - is_valid: Whether the model exists and is accessible
            - error_message: Error message if the model is invalid
    """
    try:
        # Check if model exists on HuggingFace
        # First try without token (for public models)
        try:
            # Try to access public model without token
            api_public = HfApi()
            model_info = api_public.model_info(repo_id=model_name)
            logger.debug(f"Successfully accessed model {model_name}")
            return True, None
        except Exception as e:
            logger.debug(f"Could not access model without token")
            # If that fails, try with token (for private models)
            if token:
                try:
                    api_with_token = HfApi(token=token)
                    model_info = api_with_token.model_info(repo_id=model_name)
                    logger.debug(f"Successfully accessed model {model_name} with authentication")
                    return True, None
                except Exception as e:
                    return False, f"Model {model_name} not found or not accessible: {str(e)}"
            else:
                return False, f"Model {model_name} not found or not accessible: {str(e)}"
    except Exception as e:
        return False, f"Error validating model: {str(e)}"

def has_adapter_config(model_name: str, token: str = None) -> tuple[bool, str]:
    """
    Checks if the model repository contains adapter configuration files.
    
    Args:
        model_name: Name of the model to check
        token: HuggingFace API token (optional)
        
    Returns:
        Tuple[bool, str]: A tuple containing:
            - has_adapter: Whether the model contains adapter configuration
            - message: Additional information or error message
    """
    try:
        # Initialize API with or without token
        api = HfApi(token=token) if token else HfApi()
        
        # Get the list of files in the repository
        repo_files = api.list_repo_files(repo_id=model_name)
        
        # Check for specific adapter configuration files
        adapter_files = [
            "adapter_config.json",
            "adapter_model.bin",
            "adapter_model.safetensors",
            "adapter.json",
            "adapter.safetensors",
            "adapter.bin"
        ]
        
        # Look for specific adapter files
        found_adapter_files = []
        for file in repo_files:
            file_lower = file.lower()
            if any(adapter_file.lower() in file_lower for adapter_file in adapter_files):
                found_adapter_files.append(file)
        
        # Check if we found adapter configuration
        has_adapter = len(found_adapter_files) > 0
        
        if has_adapter:
            adapter_files_str = ", ".join(found_adapter_files)
            return True, f"Found adapter configuration: {adapter_files_str}"
        else:
            return False, "No adapter configuration found"
        
    except Exception as e:
        return False, f"Error checking for adapter configuration: {str(e)}"

def has_safetensor_model(model_name: str, token: str = None) -> tuple[bool, str]:
    """
    Checks if the model repository contains safetensor model files.
    
    Args:
        model_name: Name of the model to check
        token: HuggingFace API token (optional)
        
    Returns:
        Tuple[bool, str]: A tuple containing:
            - has_safetensor: Whether the model contains safetensor model files
            - message: Additional information or error message
    """
    try:
        # Initialize API with or without token
        api = HfApi(token=token) if token else HfApi()
        
        # Get the list of files in the repository
        repo_files = api.list_repo_files(repo_id=model_name)
        
        # Look for safetensor model files (model_*.safetensors)
        safetensor_files = []
        model_pattern = "model_"
        safetensor_extension = ".safetensors"
        
        for file in repo_files:
            file_lower = file.lower()
            if model_pattern in file_lower and file_lower.endswith(safetensor_extension):
                safetensor_files.append(file)
        
        # Check if we found any safetensor model files
        has_safetensor = len(safetensor_files) > 0
        
        if has_safetensor:
            safetensor_files_str = ", ".join(safetensor_files)
            return True, f"Found safetensor model files: {safetensor_files_str}"
        else:
            # If no model_*.safetensors files, check for any .safetensors files
            any_safetensor_files = [file for file in repo_files if file.lower().endswith(safetensor_extension)]
            
            if any_safetensor_files:
                safetensor_files_str = ", ".join(any_safetensor_files)
                return True, f"Found safetensor files: {safetensor_files_str}"
            else:
                return False, "No safetensor model files found"
        
    except Exception as e:
        return False, f"Error checking for safetensor model files: {str(e)}"

def determine_model_type(model_name: str, token: str = None) -> tuple[str, str]:
    """
    Determines the type of model based on the files in the repository.
    
    Args:
        model_name: Name of the model to check
        token: HuggingFace API token (optional)
        
    Returns:
        Tuple[str, str]: A tuple containing:
            - model_type: Type of model (adapter, merged_model, unknown)
            - message: Additional information or details
    """
    try:
        # Check for adapter configuration
        has_adapter, adapter_message = has_adapter_config(model_name, token)
        
        # Check for safetensor model files
        has_safetensor, safetensor_message = has_safetensor_model(model_name, token)
        
        # Determine model type based on checks
        if has_adapter:
            return "adapter", adapter_message
        elif has_safetensor:
            return "merged_model", safetensor_message
        else:
            return "unknown", "Could not determine model type: no adapter config or safetensor model files found"
        
    except Exception as e:
        return "unknown", f"Error determining model type: {str(e)}"