hyrinmansoor/text2frappe-s3-flan-query

🧠 Text2Frappe - Stage 3 Query Generator (FLAN-T5-BASE)

This model is the third stage in the Text2Frappe pipeline, which enables natural language interface to ERPNext by converting questions into executable database queries.

---

🎯 Task

Text2Text Generation – Prompt-based query formulation.

Given:

• A detected ERPNext Doctype (from Stage 1),
• A natural language question,
• A list of selected relevant fields (from Stage 2),

this model generates a valid Frappe ORM-style query (e.g., frappe.get_all(...)) to retrieve the required data.

---

🧩 Input Format

Inputs are JSON-style strings containing:

• doctype: the ERPNext document type.
• question: the user's question in natural language.
• fields: a list of relevant field names predicted by Stage 2.

📥 Example Input

{
  "doctype": "Purchase Invoice Advance",
  "question": "List the reference types used in advance payments made this month.",
  "fields": ["reference_type"]
}

📤 Example Output

frappe.get_all('Purchase Invoice Advance', filters={'posting_date': ['between', ['2024-04-01', '2024-04-30']]}, fields=['reference_type'])