orange-lmsci 0.1.0

Orange widgets for Language Model interaction

Orange LMSci

LM Task — Orange Data Mining Widget

An Orange 3 widget that connects to a locally running Ollama server and lets you run LLM prompts against your data.

---

Usage

With an input table

1. Connect a Data table to the widget's input.

2. Write a prompt template using {column_name} placeholders matching your table's column names:

   Classify the sentiment of this review:
   
   {review_text}
   
   Reply with only: positive, negative, or neutral.
   

3. Choose Output mode:

   • Text — each LLM response is emitted on the Text output one by one.
   • Table — when all rows are processed, a copy of the input table is sent on the Data output with an extra llm column containing each row's LLM response.

4. The widget processes every row automatically when data is connected.

Standalone (no input table)

• The Query button and Query on load checkbox are enabled.
• Write any prompt (without placeholders, or with literal {braces} — they won't be substituted).
• Click Query to send it once.
• Check Query on load to run the prompt automatically every time the workflow is opened.

---

Controls

Control	Description
URL	Ollama server base URL (default: http://localhost:11434)
Model	Dropdown populated from /api/tags; editable for manual entry
↺ (Refresh)	Re-fetches the model list from the server
Output: Text / Table	Selects output mode
Query	Sends the standalone prompt (disabled when table input is connected)
Query on load	Auto-query on workflow load when no input is connected
Cancel	Cancels the current in-progress query batch

---

Requirements

• Orange 3 (orange3)
• A running Ollama server (e.g. ollama serve)
• No extra Python dependencies — uses only stdlib urllib and json