{"passport":{"unfragile":{"@version":"1.0","version":"2026-05","artifact":{"id":"hugging-face","slug":"hugging-face","name":"Hugging Face","type":"platform","url":"https://huggingface.co","page_url":"https://unfragile.ai/hugging-face","categories":["deployment-infra","model-training"],"tags":[],"pricing":{"model":"freemium","free":true,"starting_price":null},"status":"active","verified":false},"capabilities":[{"id":"hugging-face__cap_0","uri":"capability://search.retrieval.model.hub.with.unified.discovery.and.metadata.indexing","name":"model hub with unified discovery and metadata indexing","description":"Centralized repository indexing 500K+ pre-trained models across frameworks (PyTorch, TensorFlow, JAX, ONNX) with standardized metadata cards, model cards (YAML + markdown), and full-text search across model names, descriptions, and tags. Uses Git-based version control for model artifacts and enables semantic filtering by task type, language, license, and framework compatibility without requiring manual curation.","intents":["Find a pre-trained model for my specific NLP/vision/audio task without evaluating 100 options","Understand model capabilities, training data, and limitations before downloading","Filter models by license, framework compatibility, and performance benchmarks","Track model versions and access historical checkpoints"],"best_for":["ML practitioners and researchers building on existing models","Teams evaluating multiple model candidates for production","Open-source contributors discovering community-built models"],"limitations":["Metadata quality varies by contributor — some models lack detailed cards or benchmarks","Search ranking is not always aligned with model quality or popularity","No built-in A/B testing framework to compare model outputs directly on the platform"],"requires":["Hugging Face account (free tier available)","Internet connection to access hub","Compatible framework installed locally (PyTorch, TensorFlow, etc.) to use models"],"input_types":["text queries (model names, task descriptions)","filter parameters (framework, task, language)"],"output_types":["model metadata (JSON)","model cards (markdown + YAML)","download links and Git repository URLs"],"categories":["search-retrieval","memory-knowledge"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"hugging-face__cap_1","uri":"capability://data.processing.analysis.dataset.hub.with.streaming.and.lazy.loading","name":"dataset hub with streaming and lazy loading","description":"Hosts 100K+ datasets with streaming-first architecture that enables loading datasets larger than available RAM via the Hugging Face Datasets library. Uses Apache Arrow columnar format for efficient memory usage and supports on-the-fly preprocessing (tokenization, image resizing) without materializing full datasets. Integrates with Parquet, CSV, JSON, and image formats with automatic schema inference and data validation.","intents":["Load a 50GB dataset without downloading the entire file or running out of memory","Apply preprocessing transformations (tokenization, normalization) at load time without creating intermediate files","Share datasets with reproducible splits (train/val/test) and versioning","Discover datasets by task, language, or domain without manual search"],"best_for":["ML engineers training models on large-scale datasets","Researchers sharing datasets with built-in versioning and reproducibility","Teams building data pipelines that require lazy evaluation"],"limitations":["Streaming mode has ~5-10% overhead vs pre-downloaded datasets due to network I/O","Complex custom preprocessing requires writing Python code; no low-code UI for transformations","Dataset schema inference can fail on heterogeneous or poorly-formatted data"],"requires":["Python 3.7+","Hugging Face Datasets library (pip install datasets)","Internet connection for streaming mode","Sufficient disk space for cache (configurable via HF_HOME environment variable)"],"input_types":["dataset identifiers (string)","split names (train/validation/test)","preprocessing functions (Python callables)"],"output_types":["PyArrow Tables","Pandas DataFrames","NumPy arrays","streaming iterables"],"categories":["data-processing-analysis","search-retrieval"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"hugging-face__cap_10","uri":"capability://safety.moderation.safetensors.format.with.malware.detection","name":"safetensors format with malware detection","description":"Secure model serialization format that replaces pickle-based model loading with a safer, human-readable format. Safetensors files are scanned for malware signatures and suspicious code patterns before being made available for download. Format is language-agnostic and enables lazy loading of model weights without deserializing untrusted code.","intents":["Download models without risk of arbitrary code execution from pickle files","Load model weights lazily without materializing full model in memory","Inspect model structure and weight metadata without running code","Ensure models are free of malware before using them in production"],"best_for":["Teams prioritizing security in model supply chain","Organizations with strict code review processes","Users downloading models from untrusted sources"],"limitations":["Safetensors scanning is not foolproof; sophisticated attacks may evade detection","Lazy loading adds ~5-10% overhead vs pre-loaded models","Not all models on Hub use Safetensors format; older models may still use pickle","Custom model architectures may not be compatible with Safetensors"],"requires":["Safetensors library (pip install safetensors)","Model must be available in Safetensors format on Hub"],"input_types":["Safetensors files (.safetensors)"],"output_types":["PyTorch/TensorFlow/JAX tensors","model weight metadata (shape, dtype)"],"categories":["safety-moderation","data-processing-analysis"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"hugging-face__cap_11","uri":"capability://tool.use.integration.hugging.face.hub.api.with.programmatic.model.management","name":"hugging face hub api with programmatic model management","description":"REST API for programmatic interaction with Hub (uploading models, creating repos, managing access, querying metadata). Supports authentication via API tokens and enables automation of model publishing workflows. API provides endpoints for model search, metadata retrieval, and file operations (upload, delete, rename) without requiring Git.","intents":["Automate model publishing workflow from CI/CD pipeline","Programmatically search for models matching specific criteria","Upload model checkpoints from training scripts without manual Git operations","Manage model repositories and access control via code"],"best_for":["ML teams automating model publishing in CI/CD","Researchers building tools that interact with Hub","Platforms integrating Hugging Face models into their services"],"limitations":["API rate limits (varies by tier); high-volume operations may require throttling","File upload API has 5GB per file limit; large models require chunked uploads","No built-in retry logic; callers must implement exponential backoff","API documentation is scattered across multiple pages; no OpenAPI spec"],"requires":["Hugging Face API token","HTTP client library (requests, curl, etc.)","Python 3.7+ for huggingface_hub library"],"input_types":["model metadata (JSON)","model files (binary)","search queries (text)"],"output_types":["model metadata (JSON)","search results (JSON array)","file URLs (for download)"],"categories":["tool-use-integration","automation-workflow"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"hugging-face__cap_12","uri":"capability://automation.workflow.transformers.trainer.with.distributed.training.support","name":"transformers trainer with distributed training support","description":"High-level training API that abstracts away boilerplate code for fine-tuning models on custom datasets. Supports distributed training across multiple GPUs/TPUs via PyTorch Distributed Data Parallel (DDP) and DeepSpeed integration. Handles gradient accumulation, mixed-precision training, learning rate scheduling, and evaluation metrics automatically. Integrates with Weights & Biases and TensorBoard for experiment tracking.","intents":["Fine-tune a model on my dataset with minimal code (10-20 lines)","Scale training across multiple GPUs without writing distributed training code","Track training metrics and compare experiments across runs","Use advanced training techniques (mixed precision, gradient accumulation) without manual implementation"],"best_for":["ML engineers fine-tuning models on custom datasets","Teams standardizing on a single training framework","Researchers experimenting with different model architectures"],"limitations":["Trainer is opinionated; customizing training loops requires subclassing and overriding methods","Distributed training setup requires careful configuration of environment variables and launcher scripts","DeepSpeed integration adds complexity and requires additional dependencies","Debugging training issues is difficult due to abstraction layers"],"requires":["Python 3.7+","PyTorch 1.9+","Transformers library (pip install transformers[torch])","NVIDIA GPU with CUDA 11.0+ for GPU training","Optional: DeepSpeed for advanced distributed training"],"input_types":["training dataset (Hugging Face Dataset or PyTorch DataLoader)","model (from Transformers library)","TrainingArguments (configuration dictionary)"],"output_types":["fine-tuned model (saved to disk or Hub)","training metrics (loss, accuracy, F1)","evaluation results (on validation set)"],"categories":["automation-workflow","code-generation-editing"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"hugging-face__cap_13","uri":"capability://data.processing.analysis.model.evaluation.and.benchmarking.framework","name":"model evaluation and benchmarking framework","description":"Standardized evaluation framework for comparing models across common benchmarks (GLUE, SuperGLUE, SQuAD, ImageNet, etc.) with automatic metric computation and leaderboard ranking. Supports custom evaluation datasets and metrics via pluggable evaluation functions. Results are tracked in model cards and contribute to community leaderboards for transparency.","intents":["Evaluate my fine-tuned model against standard benchmarks without manual metric computation","Compare my model's performance against other models on the same task","Track evaluation results across multiple model versions","Contribute to community leaderboards and benchmarks"],"best_for":["Researchers publishing models with benchmark results","Teams comparing model candidates for production","Communities maintaining leaderboards for specific tasks"],"limitations":["Benchmark selection is limited to popular tasks; domain-specific benchmarks not supported","Leaderboard rankings can be gamed via cherry-picking datasets or metrics","No built-in statistical significance testing; results may be noisy","Evaluation infrastructure is separate from training; requires manual integration"],"requires":["Model compatible with Transformers library","Benchmark dataset (automatically downloaded)","Evaluation metrics library (e.g., scikit-learn, seqeval)"],"input_types":["model predictions (logits or labels)","ground truth labels","benchmark dataset"],"output_types":["evaluation metrics (accuracy, F1, BLEU, etc.)","leaderboard rankings","comparison plots"],"categories":["data-processing-analysis","planning-reasoning"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"hugging-face__cap_2","uri":"capability://tool.use.integration.inference.api.with.multi.provider.task.routing","name":"inference api with multi-provider task routing","description":"Serverless inference endpoint that routes requests to appropriate model inference backends (CPU, GPU, TPU) based on model size and task type. Supports 20+ task types (text classification, token classification, question answering, image classification, object detection, etc.) with automatic model selection and batching. Uses HTTP REST API with request queuing and auto-scaling based on load; responses cached for identical inputs within 24 hours.","intents":["Run inference on a model without provisioning servers or managing GPU resources","Get predictions in <500ms for common tasks without cold-start delays","Batch multiple inference requests to amortize latency","Test a model's output before committing to production deployment"],"best_for":["Developers prototyping ML features without DevOps expertise","Teams needing on-demand inference without capacity planning","Researchers testing model behavior across different inputs"],"limitations":["Inference API has variable latency (100ms-5s) depending on model size and queue depth; not suitable for sub-100ms SLA requirements","Free tier has rate limits (1 request/second) and no SLA guarantees","Caching is opaque — identical inputs cached for 24h but no control over cache invalidation","Cold starts occur when models are unloaded from GPU after inactivity (~5-30s latency)"],"requires":["Hugging Face API token (free tier available)","HTTP client (curl, Python requests, etc.)","Model must be publicly available on Hugging Face Hub or in user's account"],"input_types":["JSON payloads with task-specific fields (text, images, audio)","batch requests (multiple inputs in single HTTP call)"],"output_types":["JSON responses with predictions, scores, and metadata","structured outputs (labels, bounding boxes, token classifications)"],"categories":["tool-use-integration","automation-workflow"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"hugging-face__cap_3","uri":"capability://automation.workflow.inference.endpoints.with.custom.docker.and.auto.scaling","name":"inference endpoints with custom docker and auto-scaling","description":"Managed inference service that deploys models to dedicated, auto-scaling infrastructure with support for custom Docker images, GPU/TPU selection, and request-based scaling. Provides private endpoints (no public internet exposure), request authentication via API tokens, and monitoring dashboards with latency/throughput metrics. Supports batch inference jobs and real-time streaming via WebSocket connections.","intents":["Deploy a model to production with guaranteed latency SLAs and auto-scaling","Run inference on proprietary models without exposing them to public API","Monitor model performance (latency, throughput, error rates) in real-time","Scale inference capacity automatically based on request volume"],"best_for":["Teams deploying models to production with SLA requirements","Organizations with proprietary models requiring private endpoints","ML platforms needing white-label inference infrastructure"],"limitations":["Minimum cost ~$0.06/hour per endpoint; not cost-effective for sporadic inference","Custom Docker images require manual optimization for inference performance","Auto-scaling has ~30-60 second latency to provision new instances during traffic spikes","No built-in model versioning or canary deployment — requires manual traffic routing"],"requires":["Hugging Face Pro account or higher","Model accessible via Hugging Face Hub or custom Docker image","API token for authentication","GPU/TPU quota (varies by region and account tier)"],"input_types":["JSON payloads (same format as Inference API)","batch job definitions (CSV/JSON with multiple inputs)","streaming requests (WebSocket)"],"output_types":["JSON responses","streaming predictions (WebSocket)","batch job results (downloadable files)"],"categories":["automation-workflow","tool-use-integration"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"hugging-face__cap_4","uri":"capability://planning.reasoning.autotrain.with.automatic.hyperparameter.tuning","name":"autotrain with automatic hyperparameter tuning","description":"No-code training service that automatically selects model architecture, hyperparameters, and training strategy based on dataset characteristics and task type. Uses Bayesian optimization to search hyperparameter space (learning rate, batch size, epochs) and early stopping to prevent overfitting. Supports text classification, token classification, question answering, image classification, object detection, and tabular regression with automatic data splitting and validation.","intents":["Fine-tune a model on my custom dataset without writing training code","Automatically find good hyperparameters without manual experimentation","Train a model end-to-end in minutes without ML expertise","Compare multiple model architectures and select the best performer"],"best_for":["Non-technical users and product managers building ML features","Teams without ML engineering resources","Rapid prototyping and MVP development"],"limitations":["Limited control over training process — no access to training code or custom loss functions","Hyperparameter search space is predefined; cannot optimize task-specific parameters","Training time can be slow (hours to days) for large datasets; no distributed training across multiple GPUs","Model interpretability and debugging are limited — black-box optimization without visibility into training dynamics"],"requires":["Hugging Face account with AutoTrain credits","Dataset in supported format (CSV, JSON, image folders)","Task type must be one of: text classification, token classification, QA, image classification, object detection, tabular regression"],"input_types":["CSV/JSON files with text and labels","image folders with class subdirectories","tabular data (CSV with numerical and categorical features)"],"output_types":["fine-tuned model (uploaded to Hugging Face Hub)","training metrics (accuracy, F1, loss curves)","model card with hyperparameters and performance"],"categories":["planning-reasoning","automation-workflow"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"hugging-face__cap_5","uri":"capability://automation.workflow.spaces.with.git.based.deployment.and.persistent.storage","name":"spaces with git-based deployment and persistent storage","description":"Serverless hosting for interactive ML demos (Gradio, Streamlit, Docker) with Git-based deployment (push to deploy), automatic HTTPS, and persistent storage via mounted volumes. Supports CPU and GPU hardware selection, environment variable secrets management, and automatic scaling based on concurrent users. Demos are publicly shareable via URL with optional authentication.","intents":["Deploy an interactive demo of my model without managing servers","Share a working example of my model with collaborators or users","Host a Gradio/Streamlit app with GPU acceleration for real-time inference","Version control my demo code alongside my model"],"best_for":["Researchers sharing reproducible demos with papers","Teams showcasing model capabilities to stakeholders","Open-source projects needing free hosting for demos"],"limitations":["Free tier has CPU-only hardware and limited concurrent users (~5); GPU requires paid tier","Cold starts occur after inactivity (~30 seconds to spin up container)","Persistent storage is limited (10GB on free tier); not suitable for large datasets","No built-in authentication beyond simple password; not suitable for sensitive applications"],"requires":["Hugging Face account (free tier available)","Git repository with Gradio/Streamlit app or Dockerfile","Python 3.7+ for Gradio/Streamlit apps"],"input_types":["Git repository URL","Dockerfile (for custom environments)","Gradio/Streamlit Python scripts"],"output_types":["Public URL to interactive demo","shareable link with optional password protection"],"categories":["automation-workflow","tool-use-integration"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"hugging-face__cap_6","uri":"capability://code.generation.editing.transformers.library.integration.with.model.caching","name":"transformers library integration with model caching","description":"Official Python library providing unified interface to 500K+ models with automatic downloading, caching, and tokenizer management. Uses local file-based caching (HF_HOME directory) to avoid re-downloading models; supports lazy loading of model weights via SafeTensors format for memory efficiency. Integrates with PyTorch, TensorFlow, and JAX with automatic device placement (CPU/GPU/TPU) and mixed-precision training support.","intents":["Load a pre-trained model with a single line of code: from_pretrained()","Automatically download and cache models locally to avoid repeated downloads","Use the same code across different frameworks (PyTorch, TensorFlow, JAX)","Fine-tune a model with automatic mixed-precision training for faster convergence"],"best_for":["ML engineers building applications with Hugging Face models","Researchers fine-tuning models for custom tasks","Teams standardizing on a single model loading interface"],"limitations":["Large models (>10GB) require significant disk space for caching; no automatic cache eviction","First load of a model requires downloading from internet (can take minutes for large models)","Mixed-precision training requires NVIDIA GPUs with compute capability 7.0+; not available on CPU","Tokenizer mismatches between model versions can cause silent failures if not carefully managed"],"requires":["Python 3.7+","PyTorch 1.9+, TensorFlow 2.4+, or JAX 0.2.0+","Hugging Face Transformers library (pip install transformers)","Internet connection for initial model download"],"input_types":["model identifiers (string, e.g., 'bert-base-uncased')","text inputs (strings or tokenized tensors)","configuration dictionaries (for model initialization)"],"output_types":["PyTorch/TensorFlow/JAX tensors","model outputs (logits, hidden states, attention weights)","tokenized inputs (token IDs, attention masks)"],"categories":["code-generation-editing","memory-knowledge"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"hugging-face__cap_7","uri":"capability://text.generation.language.model.card.generation.and.documentation.standards","name":"model card generation and documentation standards","description":"Standardized template system for documenting models with YAML frontmatter (metadata) and markdown sections (description, intended use, limitations, training data, evaluation results). Enforces documentation best practices via optional validation and provides templates for common model types. Model cards are rendered as web pages on the Hub and included in model repositories for version control.","intents":["Document my model's capabilities, limitations, and training data in a standardized format","Understand potential biases and failure modes of a model before using it","Share evaluation results and benchmarks with the community","Ensure reproducibility by documenting training procedures and hyperparameters"],"best_for":["Model creators sharing models responsibly with the community","Teams ensuring model governance and compliance documentation","Researchers publishing models alongside papers"],"limitations":["Model card completion is optional; many models lack detailed documentation","No automated validation of claims (e.g., accuracy numbers, training data descriptions)","Template is generic and may not capture domain-specific documentation needs","No built-in tools for tracking model updates or deprecation notices"],"requires":["Hugging Face account with model repository","Markdown and YAML knowledge (basic)","Model metadata (training data, evaluation metrics, hyperparameters)"],"input_types":["YAML frontmatter (model metadata)","markdown text (description, evaluation results, limitations)"],"output_types":["rendered model card (HTML on Hub)","markdown file (in model repository)"],"categories":["text-generation-language","safety-moderation"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"hugging-face__cap_8","uri":"capability://text.generation.language.community.discussions.and.model.feedback.system","name":"community discussions and model feedback system","description":"Threaded discussion interface on each model and dataset page enabling users to ask questions, report issues, and provide feedback. Discussions are indexed and searchable, allowing users to find answers to common questions without contacting model authors directly. Model authors can pin important discussions and provide official responses, creating a FAQ-like knowledge base.","intents":["Ask questions about a model's performance or usage without contacting the author directly","Report bugs or unexpected behavior in a model","Find answers to common questions from other users","Provide feedback on model quality and suggest improvements"],"best_for":["Users troubleshooting model issues","Model authors gathering community feedback","Teams building knowledge bases around popular models"],"limitations":["Discussions are unmoderated; misinformation can spread without correction","No SLA for response time from model authors","Search is basic (full-text only); no semantic search across discussions","No built-in notification system for discussion replies"],"requires":["Hugging Face account","Model or dataset page on Hub"],"input_types":["text messages (discussion posts)","optional code snippets or error logs"],"output_types":["threaded discussion threads","searchable discussion index"],"categories":["text-generation-language","search-retrieval"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"hugging-face__cap_9","uri":"capability://tool.use.integration.private.model.repositories.with.access.control","name":"private model repositories with access control","description":"Ability to create private model repositories with fine-grained access control (read-only, write, admin) for team members. Private repos are not indexed in public search and require authentication to access. Supports the same Git-based versioning and model card system as public repos, enabling teams to share proprietary models internally.","intents":["Share a proprietary model with my team without exposing it publicly","Control who can download, modify, or deploy my model","Maintain version history of proprietary models with Git","Collaborate on model improvements within a team"],"best_for":["Teams building proprietary ML products","Organizations with IP protection requirements","Enterprises sharing models across departments"],"limitations":["Private repos require paid Hugging Face account tier","No built-in encryption at rest; relies on Hugging Face infrastructure security","Access control is coarse-grained (no per-file permissions)","No audit logs for tracking who accessed or modified models"],"requires":["Hugging Face Pro account or higher","Team members with Hugging Face accounts"],"input_types":["model files (same as public repos)","access control lists (user emails and permission levels)"],"output_types":["private model repository URL","access tokens for programmatic access"],"categories":["tool-use-integration","safety-moderation"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"hugging-face__headline","uri":"capability://deployment.infra.ai.model.hub.and.dataset.repository","name":"ai model hub and dataset repository","description":"Hugging Face is the leading platform for hosting and sharing AI models and datasets, serving as a central hub for the open-source AI ecosystem with over 500K models and 100K datasets available for developers and researchers.","intents":["best AI model repository","AI models for machine learning projects","top platforms for sharing datasets","AI model hosting services","best tools for AI model deployment"],"best_for":[],"limitations":[],"requires":[],"input_types":[],"output_types":[],"categories":["deployment-infra","model-training"],"confidence":0.5,"matches":0,"success_rate":0}],"trust":{"score":60,"verified":false,"data_access_risk":"high","permissions":["Hugging Face account (free tier available)","Internet connection to access hub","Compatible framework installed locally (PyTorch, TensorFlow, etc.) to use models","Python 3.7+","Hugging Face Datasets library (pip install datasets)","Internet connection for streaming mode","Sufficient disk space for cache (configurable via HF_HOME environment variable)","Safetensors library (pip install safetensors)","Model must be available in Safetensors format on Hub","Hugging Face API token"],"failure_modes":["Metadata quality varies by contributor — some models lack detailed cards or benchmarks","Search ranking is not always aligned with model quality or popularity","No built-in A/B testing framework to compare model outputs directly on the platform","Streaming mode has ~5-10% overhead vs pre-downloaded datasets due to network I/O","Complex custom preprocessing requires writing Python code; no low-code UI for transformations","Dataset schema inference can fail on heterogeneous or poorly-formatted data","Safetensors scanning is not foolproof; sophisticated attacks may evade detection","Lazy loading adds ~5-10% overhead vs pre-loaded models","Not all models on Hub use Safetensors format; older models may still use pickle","Custom model architectures may not be compatible with Safetensors","builder identity is not verified yet","no observed match outcomes yet"],"rank_breakdown":{"adoption":0.7,"quality":1,"ecosystem":0.25,"match_graph":0.25,"freshness":0.75,"weights":{"adoption":0.3,"quality":0.25,"ecosystem":0.15,"match_graph":0.25,"freshness":0.05}},"observed_outcomes":{"matches":0,"success_rate":0,"avg_confidence":0,"top_intents":[],"last_matched_at":null},"maintenance":{"status":"active","updated_at":"2026-05-24T12:16:23.327Z","last_scraped_at":null,"last_commit":null},"community":{"stars":null,"forks":null,"weekly_downloads":null,"model_downloads":null,"model_likes":null}},"distribution":{"claim_url":"https://unfragile.ai/submit?claim=hugging-face","compare_url":"https://unfragile.ai/compare?artifact=hugging-face"}},"signature":"yt174AkSvRrKRs5yKdAkAkWq28bIvPy8IHs5ZEReNO9FJto8BVAzW0IIxkVH30qKiCZXn8hScGj1dz+KcKMoCQ==","signedAt":"2026-06-20T12:09:53.622Z","signedBy":"unfragile.ai","version":1},"_links":{"self":"https://unfragile.ai/api/v1/passport/hugging-face","artifact":"https://unfragile.ai/hugging-face","verify":"https://unfragile.ai/api/v1/verify?slug=hugging-face","publicKey":"https://unfragile.ai/api/v1/trust-passport-public-key","spec":"https://unfragile.ai/trust","schema":"https://unfragile.ai/schema.json","docs":"https://unfragile.ai/docs"}}