a1c00l's picture
Update src/aibom_generator/api.py
4f95bff verified
raw
history blame
27.9 kB
#!/usr/bin/env python
import os
import json
import logging
import sys
from fastapi import FastAPI, HTTPException, Request, Form
from fastapi.responses import HTMLResponse, JSONResponse, FileResponse
from fastapi.staticfiles import StaticFiles
from fastapi.templating import Jinja2Templates
from pydantic import BaseModel
from datetime import datetime
from datasets import Dataset, load_dataset, concatenate_datasets
import os
import logging
from urllib.parse import urlparse
import re # Import regex module
import html # Import html module for escaping
from huggingface_hub import HfApi
from huggingface_hub.utils import RepositoryNotFoundError # For specific error handling
# Configure logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
# --- Add Counter Configuration (as of May 3, 2025) ---
HF_REPO = "aetheris-ai/aisbom-usage-log" # User needs to create this private repo
HF_TOKEN = os.getenv("HF_TOKEN") # User must set this environment variable
# --- End Counter Configuration ---
# Define directories
templates_dir = "templates"
OUTPUT_DIR = "/tmp/aibom_output"
# Initialize templates
templates = Jinja2Templates(directory=templates_dir)
# Create app
app = FastAPI(title="AI SBOM Generator API")
# Ensure output directory exists
os.makedirs(OUTPUT_DIR, exist_ok=True)
# Mount output directory as static files
app.mount("/output", StaticFiles(directory=OUTPUT_DIR), name="output")
# Status response model
class StatusResponse(BaseModel):
status: str
version: str
generator_version: str
# --- Model ID Validation and Normalization Helpers ---
# Regex for valid Hugging Face ID parts (alphanumeric, -, _, .)
# Allows owner/model format
HF_ID_REGEX = re.compile(r"^[a-zA-Z0-9\.\-\_]+/[a-zA-Z0-9\.\-\_]+$")
def is_valid_hf_input(input_str: str) -> bool:
"""Checks if the input is a valid Hugging Face model ID or URL."""
if not input_str or len(input_str) > 200: # Basic length check
return False
if input_str.startswith(("http://", "https://") ):
try:
parsed = urlparse(input_str)
# Check domain and path structure
if parsed.netloc == "huggingface.co":
path_parts = parsed.path.strip("/").split("/")
# Must have at least owner/model, can have more like /tree/main
if len(path_parts) >= 2 and path_parts[0] and path_parts[1]:
# Check characters in the relevant parts
if re.match(r"^[a-zA-Z0-9\.\-\_]+$", path_parts[0]) and \
re.match(r"^[a-zA-Z0-9\.\-\_]+$", path_parts[1]):
return True
return False # Not a valid HF URL format
except Exception:
return False # URL parsing failed
else:
# Assume owner/model format, check with regex
return bool(HF_ID_REGEX.match(input_str))
def _normalise_model_id(raw_id: str) -> str:
"""
Accept either validated 'owner/model' or a validated full URL like
'https://huggingface.co/owner/model'. Return 'owner/model'.
Assumes input has already been validated by is_valid_hf_input.
"""
if raw_id.startswith(("http://", "https://") ):
path = urlparse(raw_id).path.lstrip("/")
parts = path.split("/")
# We know from validation that parts[0] and parts[1] exist
return f"{parts[0]}/{parts[1]}"
return raw_id # Already in owner/model format
# --- End Model ID Helpers ---
# --- Add Counter Helper Functions ---
def log_sbom_generation(model_id: str):
"""Logs a successful SBOM generation event to the Hugging Face dataset."""
if not HF_TOKEN:
logger.warning("HF_TOKEN not set. Skipping SBOM generation logging.")
return
try:
# Normalize model_id before logging
normalized_model_id_for_log = _normalise_model_id(model_id) # added to normalize id
log_data = {
"timestamp": [datetime.utcnow().isoformat()],
"event": ["generated"],
"model_id": [normalized_model_id_for_log] # use normalized_model_id_for_log
}
ds_new_log = Dataset.from_dict(log_data)
# Try to load existing dataset to append
try:
# Use trust_remote_code=True if required by the dataset/model on HF
# Corrected: Removed unnecessary backslashes around 'train'
existing_ds = load_dataset(HF_REPO, token=HF_TOKEN, split='train', trust_remote_code=True)
# Check if dataset is empty or has different columns (handle initial creation)
if len(existing_ds) > 0 and set(existing_ds.column_names) == set(log_data.keys()):
ds_to_push = concatenate_datasets([existing_ds, ds_new_log])
elif len(existing_ds) == 0:
logger.info(f"Dataset {HF_REPO} is empty. Pushing initial data.")
ds_to_push = ds_new_log
else:
logger.warning(f"Dataset {HF_REPO} has unexpected columns {existing_ds.column_names} vs {list(log_data.keys())}. Appending new log anyway, structure might differ.")
# Attempt concatenation even if columns differ slightly, HF might handle it
# Or consider more robust schema migration/handling if needed
ds_to_push = concatenate_datasets([existing_ds, ds_new_log])
except Exception as load_err:
# Handle case where dataset doesn't exist yet or other loading errors
# Corrected: Removed unnecessary backslash in doesn't
logger.info(f"Could not load existing dataset {HF_REPO} (may not exist yet): {load_err}. Pushing new dataset.")
ds_to_push = ds_new_log # ds is already prepared with the new log entry
# Push the updated or new dataset
# Corrected: Removed unnecessary backslash in it's
ds_to_push.push_to_hub(HF_REPO, token=HF_TOKEN, private=True) # Ensure it's private
logger.info(f"Successfully logged SBOM generation for {normalized_model_id_for_log} to {HF_REPO}") # use normalized model id
except Exception as e:
logger.error(f"Failed to log SBOM generation to {HF_REPO}: {e}")
def get_sbom_count() -> str:
"""Retrieves the total count of generated SBOMs from the Hugging Face dataset."""
if not HF_TOKEN:
logger.warning("HF_TOKEN not set. Cannot retrieve SBOM count.")
return "N/A"
try:
# Load the dataset - assumes 'train' split exists after first push
# Use trust_remote_code=True if required by the dataset/model on HF
# Corrected: Removed unnecessary backslashes around 'train'
ds = load_dataset(HF_REPO, token=HF_TOKEN, split='train', trust_remote_code=True)
count = len(ds)
logger.info(f"Retrieved SBOM count: {count} from {HF_REPO}")
# Format count for display (e.g., add commas for large numbers)
return f"{count:,}"
except Exception as e:
logger.error(f"Failed to retrieve SBOM count from {HF_REPO}: {e}")
# Return "N/A" or similar indicator on error
return "N/A"
# --- End Counter Helper Functions ---
@app.on_event("startup")
async def startup_event():
os.makedirs(OUTPUT_DIR, exist_ok=True)
logger.info(f"Output directory ready at {OUTPUT_DIR}")
logger.info(f"Registered routes: {[route.path for route in app.routes]}")
@app.get("/", response_class=HTMLResponse)
async def root(request: Request):
sbom_count = get_sbom_count() # Get count
try:
return templates.TemplateResponse("index.html", {"request": request, "sbom_count": sbom_count}) # Pass to template
except Exception as e:
logger.error(f"Error rendering template: {str(e)}")
# Attempt to render error page even if main page fails
try:
return templates.TemplateResponse("error.html", {"request": request, "error": f"Template rendering error: {str(e)}", "sbom_count": sbom_count})
except Exception as template_err:
# Fallback if error template also fails
logger.error(f"Error rendering error template: {template_err}")
raise HTTPException(status_code=500, detail=f"Template rendering error: {str(e)}")
@app.get("/status", response_model=StatusResponse)
async def get_status():
return StatusResponse(status="operational", version="1.0.0", generator_version="1.0.0")
# Import utils module for completeness score calculation
def import_utils():
"""Import utils module with fallback paths."""
try:
# Try different import paths
sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
# Try direct import first
try:
from utils import calculate_completeness_score
logger.info("Imported utils.calculate_completeness_score directly")
return calculate_completeness_score
except ImportError:
pass
# Try from src
try:
from src.aibom_generator.utils import calculate_completeness_score
logger.info("Imported src.aibom_generator.utils.calculate_completeness_score")
return calculate_completeness_score
except ImportError:
pass
# Try from aibom_generator
try:
from aibom_generator.utils import calculate_completeness_score
logger.info("Imported aibom_generator.utils.calculate_completeness_score")
return calculate_completeness_score
except ImportError:
pass
# If all imports fail, use the default implementation
logger.warning("Could not import calculate_completeness_score, using default implementation")
return None
except Exception as e:
logger.error(f"Error importing utils: {str(e)}")
return None
# Try to import the calculate_completeness_score function
calculate_completeness_score = import_utils()
# Helper function to create a comprehensive completeness_score with field_checklist
def create_comprehensive_completeness_score(aibom=None):
"""
Create a comprehensive completeness_score object with all required attributes.
If aibom is provided and calculate_completeness_score is available, use it to calculate the score.
Otherwise, return a default score structure.
"""
# If we have the calculate_completeness_score function and an AIBOM, use it
if calculate_completeness_score and aibom:
try:
return calculate_completeness_score(aibom, validate=True, use_best_practices=True)
except Exception as e:
logger.error(f"Error calculating completeness score: {str(e)}")
# Otherwise, return a default comprehensive structure
return {
"total_score": 75.5, # Default score for better UI display
"section_scores": {
"required_fields": 20,
"metadata": 15,
"component_basic": 18,
"component_model_card": 15,
"external_references": 7.5
},
"max_scores": {
"required_fields": 20,
"metadata": 20,
"component_basic": 20,
"component_model_card": 30,
"external_references": 10
},
"field_checklist": {
# Required fields
"bomFormat": "✔ ★★★",
"specVersion": "✔ ★★★",
"serialNumber": "✔ ★★★",
"version": "✔ ★★★",
"metadata.timestamp": "✔ ★★",
"metadata.tools": "✔ ★★",
"metadata.authors": "✔ ★★",
"metadata.component": "✔ ★★",
# Component basic info
"component.type": "✔ ★★",
"component.name": "✔ ★★★",
"component.bom-ref": "✔ ★★",
"component.purl": "✔ ★★",
"component.description": "✔ ★★",
"component.licenses": "✔ ★★",
# Model card
"modelCard.modelParameters": "✔ ★★",
"modelCard.quantitativeAnalysis": "✘ ★★",
"modelCard.considerations": "✔ ★★",
# External references
"externalReferences": "✔ ★",
# Additional fields from FIELD_CLASSIFICATION
"name": "✔ ★★★",
"downloadLocation": "✔ ★★★",
"primaryPurpose": "✔ ★★★",
"suppliedBy": "✔ ★★★",
"energyConsumption": "✘ ★★",
"hyperparameter": "✔ ★★",
"limitation": "✔ ★★",
"safetyRiskAssessment": "✘ ★★",
"typeOfModel": "✔ ★★",
"modelExplainability": "✘ ★",
"standardCompliance": "✘ ★",
"domain": "✔ ★",
"energyQuantity": "✘ ★",
"energyUnit": "✘ ★",
"informationAboutTraining": "✔ ★",
"informationAboutApplication": "✔ ★",
"metric": "✘ ★",
"metricDecisionThreshold": "✘ ★",
"modelDataPreprocessing": "✘ ★",
"autonomyType": "✘ ★",
"useSensitivePersonalInformation": "✘ ★"
},
"field_tiers": {
# Required fields
"bomFormat": "critical",
"specVersion": "critical",
"serialNumber": "critical",
"version": "critical",
"metadata.timestamp": "important",
"metadata.tools": "important",
"metadata.authors": "important",
"metadata.component": "important",
# Component basic info
"component.type": "important",
"component.name": "critical",
"component.bom-ref": "important",
"component.purl": "important",
"component.description": "important",
"component.licenses": "important",
# Model card
"modelCard.modelParameters": "important",
"modelCard.quantitativeAnalysis": "important",
"modelCard.considerations": "important",
# External references
"externalReferences": "supplementary",
# Additional fields from FIELD_CLASSIFICATION
"name": "critical",
"downloadLocation": "critical",
"primaryPurpose": "critical",
"suppliedBy": "critical",
"energyConsumption": "important",
"hyperparameter": "important",
"limitation": "important",
"safetyRiskAssessment": "important",
"typeOfModel": "important",
"modelExplainability": "supplementary",
"standardCompliance": "supplementary",
"domain": "supplementary",
"energyQuantity": "supplementary",
"energyUnit": "supplementary",
"informationAboutTraining": "supplementary",
"informationAboutApplication": "supplementary",
"metric": "supplementary",
"metricDecisionThreshold": "supplementary",
"modelDataPreprocessing": "supplementary",
"autonomyType": "supplementary",
"useSensitivePersonalInformation": "supplementary"
},
"missing_fields": {
"critical": [],
"important": ["modelCard.quantitativeAnalysis", "energyConsumption", "safetyRiskAssessment"],
"supplementary": ["modelExplainability", "standardCompliance", "energyQuantity", "energyUnit",
"metric", "metricDecisionThreshold", "modelDataPreprocessing",
"autonomyType", "useSensitivePersonalInformation"]
},
"completeness_profile": {
"name": "standard",
"description": "Comprehensive fields for proper documentation",
"satisfied": True
},
"penalty_applied": False,
"penalty_reason": None,
"recommendations": [
{
"priority": "medium",
"field": "modelCard.quantitativeAnalysis",
"message": "Missing important field: modelCard.quantitativeAnalysis",
"recommendation": "Add quantitative analysis information to the model card"
},
{
"priority": "medium",
"field": "energyConsumption",
"message": "Missing important field: energyConsumption - helpful for environmental impact assessment",
"recommendation": "Consider documenting energy consumption metrics for better transparency"
},
{
"priority": "medium",
"field": "safetyRiskAssessment",
"message": "Missing important field: safetyRiskAssessment",
"recommendation": "Add safety risk assessment information to improve documentation"
}
]
}
@app.post("/generate", response_class=HTMLResponse)
async def generate_form(
request: Request,
model_id: str = Form(...),
include_inference: bool = Form(False),
use_best_practices: bool = Form(True)
):
sbom_count = get_sbom_count() # Get count early for context
# --- Input Sanitization ---
sanitized_model_id = html.escape(model_id)
# --- Input Format Validation ---
if not is_valid_hf_input(sanitized_model_id):
error_message = "Invalid input format. Please provide a valid Hugging Face model ID (e.g., 'owner/model') or a full model URL (e.g., 'https://huggingface.co/owner/model') ."
logger.warning(f"Invalid model input format received: {model_id}") # Log original input
# Try to display sanitized input in error message
return templates.TemplateResponse(
"error.html", {"request": request, "error": error_message, "sbom_count": sbom_count, "model_id": sanitized_model_id}
)
# --- Normalize the SANITIZED and VALIDATED model ID ---
normalized_model_id = _normalise_model_id(sanitized_model_id)
# --- Check if the ID corresponds to an actual HF Model ---
try:
hf_api = HfApi()
logger.info(f"Attempting to fetch model info for: {normalized_model_id}")
model_info = hf_api.model_info(normalized_model_id)
logger.info(f"Successfully fetched model info for: {normalized_model_id}")
except RepositoryNotFoundError:
error_message = f"Error: The provided ID \"{normalized_model_id}\" could not be found on Hugging Face or does not correspond to a model repository."
logger.warning(f"Repository not found for ID: {normalized_model_id}")
return templates.TemplateResponse(
"error.html", {"request": request, "error": error_message, "sbom_count": sbom_count, "model_id": normalized_model_id}
)
except Exception as api_err: # Catch other potential API errors
error_message = f"Error verifying model ID with Hugging Face API: {str(api_err)}"
logger.error(f"HF API error for {normalized_model_id}: {str(api_err)}")
return templates.TemplateResponse(
"error.html", {"request": request, "error": error_message, "sbom_count": sbom_count, "model_id": normalized_model_id}
)
# --- End Model Existence Check ---
# --- Main Generation Logic ---
try:
# Try different import paths for AIBOMGenerator
generator = None
try:
from src.aibom_generator.generator import AIBOMGenerator
generator = AIBOMGenerator()
except ImportError:
try:
from aibom_generator.generator import AIBOMGenerator
generator = AIBOMGenerator()
except ImportError:
try:
from generator import AIBOMGenerator
generator = AIBOMGenerator()
except ImportError:
logger.error("Could not import AIBOMGenerator from any known location")
raise ImportError("Could not import AIBOMGenerator from any known location")
# Generate AIBOM (pass SANITIZED ID)
aibom = generator.generate_aibom(
model_id=sanitized_model_id, # Use sanitized ID
include_inference=include_inference,
use_best_practices=use_best_practices
)
enhancement_report = generator.get_enhancement_report()
# Save AIBOM to file, use industry term ai_sbom in file name
# Corrected: Removed unnecessary backslashes around '/' and '_'
# Save AIBOM to file using normalized ID
filename = f"{normalized_model_id.replace('/', '_')}_ai_sbom.json"
filepath = os.path.join(OUTPUT_DIR, filename)
with open(filepath, "w") as f:
json.dump(aibom, f, indent=2)
# --- Log Generation Event ---
log_sbom_generation(sanitized_model_id) # Use sanitized ID
sbom_count = get_sbom_count() # Refresh count after logging
# --- End Log ---
download_url = f"/output/{filename}"
# Create download and UI interaction scripts
download_script = f"""
<script>
function downloadJSON() {{
const a = document.createElement('a');
a.href = '{download_url}';
a.download = '{filename}';
document.body.appendChild(a);
a.click();
document.body.removeChild(a);
}}
function switchTab(tabId) {{
// Hide all tabs
document.querySelectorAll('.tab-content').forEach(tab => {{
tab.classList.remove('active');
}});
// Deactivate all tab buttons
document.querySelectorAll('.aibom-tab').forEach(button => {{
button.classList.remove('active');
}});
// Show the selected tab
document.getElementById(tabId).classList.add('active');
// Activate the clicked button
event.currentTarget.classList.add('active');
}}
function toggleCollapsible(element) {{
element.classList.toggle('active');
var content = element.nextElementSibling;
if (content.style.maxHeight) {{
content.style.maxHeight = null;
content.classList.remove('active');
}} else {{
content.style.maxHeight = content.scrollHeight + "px";
content.classList.add('active');
}}
}}
</script>
"""
# Get completeness score or create a comprehensive one if not available
# Use sanitized_model_id
completeness_score = None
if hasattr(generator, 'get_completeness_score'):
try:
completeness_score = generator.get_completeness_score(sanitized_model_id)
logger.info("Successfully retrieved completeness_score from generator")
except Exception as e:
logger.error(f"Completeness score error from generator: {str(e)}")
# If completeness_score is None or doesn't have field_checklist, use comprehensive one
if completeness_score is None or not isinstance(completeness_score, dict) or 'field_checklist' not in completeness_score:
logger.info("Using comprehensive completeness_score with field_checklist")
completeness_score = create_comprehensive_completeness_score(aibom)
# Ensure enhancement_report has the right structure
if enhancement_report is None:
enhancement_report = {
"ai_enhanced": False,
"ai_model": None,
"original_score": {"total_score": 0, "completeness_score": 0},
"final_score": {"total_score": 0, "completeness_score": 0},
"improvement": 0
}
else:
# Ensure original_score has completeness_score
if "original_score" not in enhancement_report or enhancement_report["original_score"] is None:
enhancement_report["original_score"] = {"total_score": 0, "completeness_score": 0}
elif "completeness_score" not in enhancement_report["original_score"]:
enhancement_report["original_score"]["completeness_score"] = enhancement_report["original_score"].get("total_score", 0)
# Ensure final_score has completeness_score
if "final_score" not in enhancement_report or enhancement_report["final_score"] is None:
enhancement_report["final_score"] = {"total_score": 0, "completeness_score": 0}
elif "completeness_score" not in enhancement_report["final_score"]:
enhancement_report["final_score"]["completeness_score"] = enhancement_report["final_score"].get("total_score", 0)
# Add display names and tooltips for score sections
display_names = {
"required_fields": "Required Fields",
"metadata": "Metadata",
"component_basic": "Component Basic Info",
"component_model_card": "Model Card",
"external_references": "External References"
}
tooltips = {
"required_fields": "Basic required fields for a valid AIBOM",
"metadata": "Information about the AIBOM itself",
"component_basic": "Basic information about the AI model component",
"component_model_card": "Detailed model card information",
"external_references": "Links to external resources"
}
weights = {
"required_fields": 20,
"metadata": 20,
"component_basic": 20,
"component_model_card": 30,
"external_references": 10
}
# Render the template with all necessary data, with normalized model ID
return templates.TemplateResponse(
"result.html",
{
"request": request,
"model_id": normalized_model_id,
"aibom": aibom,
"enhancement_report": enhancement_report,
"completeness_score": completeness_score,
"download_url": download_url,
"download_script": download_script,
"display_names": display_names,
"tooltips": tooltips,
"weights": weights,
"sbom_count": sbom_count,
"display_names": display_names,
"tooltips": tooltips,
"weights": weights
}
)
# --- Main Exception Handling ---
except Exception as e:
logger.error(f"Error generating AI SBOM: {str(e)}")
sbom_count = get_sbom_count() # Refresh count just in case
# Pass count, added normalized model ID
return templates.TemplateResponse(
"error.html", {"request": request, "error": str(e), "sbom_count": sbom_count, "model_id": normalized_model_id}
)
@app.get("/download/{filename}")
async def download_file(filename: str):
"""
Download a generated AIBOM file.
This endpoint serves the generated AIBOM JSON files for download.
"""
file_path = os.path.join(OUTPUT_DIR, filename)
if not os.path.exists(file_path):
raise HTTPException(status_code=404, detail="File not found")
return FileResponse(
file_path,
media_type="application/json",
filename=filename
)
# If running directly (for local testing)
if __name__ == "__main__":
import uvicorn
# Ensure HF_TOKEN is set for local testing if needed
if not HF_TOKEN:
print("Warning: HF_TOKEN environment variable not set. SBOM count will show N/A and logging will be skipped.")
uvicorn.run(app, host="0.0.0.0", port=8000)