Huggingface Hub

• You want to find a model matching specific criteria (e.g. "small open-source VLM under 4B parameters") without manually browsing the Hub.
• You need to pull a dataset to a local path or push a fine-tuned model or dataset back to your HF account.
• You want to run SQL queries against a dataset split directly from the assistant.
• You need to deploy, inspect, or tear down inference endpoints or manage Spaces and storage buckets.

• You do not have the hf CLI installed or are not authenticated with a Hugging Face account — the skill will not function without these prerequisites.
• You are working in a locked-down environment where outbound access to huggingface.co is restricted.
• You only need a one-off model lookup and prefer to browse the Hub UI manually.

Search the Hugging Face Hub for models and datasets using natural-language criteria such as task type, parameter count, or license. Results surface directly in the conversation without leaving your workflow.

Pull any public (or accessible gated) model or dataset to a local path with a single instruction. The skill delegates to the hf CLI under the hood, respecting your stored credentials.

Push fine-tuned models, datasets, or Spaces to your HF account and manage repo files, revisions, and metadata — all without leaving the assistant.

Run SQL queries against specific splits of any Hub dataset. Useful for quickly profiling data, checking schema, or filtering rows before committing to a full download.

Deploy and manage Hugging Face inference endpoints from natural language instructions. Monitor status, update configurations, or shut down endpoints on demand.

Create, update, and manage Hugging Face Spaces and associated storage buckets, keeping your deployed demos and persistent storage in sync.

Ask the assistant to find an open-source vision-language model under 4B parameters with a permissive license. The skill queries the Hub and returns ranked candidates with relevant metadata.

Before downloading gigabytes of data, query the train split of a Hub dataset with SQL to inspect its schema, count rows, or filter for relevant examples.

After local training, push your fine-tuned model or curated dataset to your HF account in one step, including repo creation and file upload.

Deploy a model as an inference endpoint for a project, then pause or delete it when the project wraps — all without leaving the assistant or touching the Hub UI.