Hugging Face Block

The Hugging Face block generates completions through the Hugging Face Inference API, giving you access to a wide range of open-source models across multiple inference providers. Reach for it when you want chat completions, content generation, or AI-powered conversations from open models with customizable parameters.

Overview

When to Use

• Run chat completions against open-source models hosted on Hugging Face instead of a single proprietary provider.
• Pick a specific inference provider (Novita, Cerebras, Cohere, Fireworks, Together, etc.) for cost, latency, or availability reasons.
• Generate content or conversational responses with a custom system prompt to steer model behavior.
• Tune output with temperature and a maximum token budget for predictable, bounded responses.
• Use a model that is only available through the Hugging Face ecosystem rather than a native provider block.

Configuration

System Prompt

Optional instructions that guide the model's behavior and tone for the request.

User Prompt

The required user message content the model responds to.

Provider

Required dropdown selecting the inference provider. Options: Novita, Cerebras, Cohere, Fal AI, Fireworks, Hyperbolic, HF Inference, Nebius, Nscale, Replicate, SambaNova, Together. Defaults to Novita.

Model

Required model identifier. The model must be available for the selected provider (e.g., deepseek/deepseek-v3-0324, meta-llama/Llama-3.2-3B-Instruct-Turbo).

Temperature

Slider (0–2) controlling response randomness. Defaults to 0.7.

Max Tokens

Optional cap on the number of output tokens generated.

API Token

Required Hugging Face API token (stored as a password field).

Inputs & Outputs

• Inputs:
  • systemPrompt (string) — System instructions
  • content (string) — User message content
  • provider (string) — Model provider
  • model (string) — Model identifier
  • temperature (string) — Response randomness
  • maxTokens (string) — Maximum output tokens
  • apiKey (string) — API access token
• Outputs:
  • content (string) — Generated response
  • model (string) — Model used
  • usage (json) — Token usage stats

Tools

Hugging Face Chat (huggingface_chat) — Sends a chat-completion request to the Hugging Face Inference Router. Builds a messages array from the optional system prompt and required user message, then posts to the provider-specific endpoint under https://router.huggingface.co/<provider>/.... Returns the generated text, the model name that was used, and prompt/completion/total token counts.

YAML Example

huggingface_1:
  type: huggingface
  name: "Hugging Face"
  inputs:
    systemPrompt: "You are a helpful assistant."
    content: "{{start.input}}"
    provider: "novita"
    model: "deepseek/deepseek-v3-0324"
    temperature: 0.7
    maxTokens: 1000
    apiKey: "{{HUGGINGFACE_API_TOKEN}}"
  connections:
    outgoing:
      - target: next-block-id