Qrzysztof/functiongemma-270m-it-prepaid-cards-v2

Overview

This model is a fine-tuned version of google/functiongemma-270m-it (Gemma 3 270M) specifically designed for tool-calling related to prepaid cards. It recognizes user intents in chat conversations and generates appropriate tool calls for purchase_card, get_card_balance, and get_transaction_history functions.

Key Capabilities

• Multilingual Support: Trained on 107 languages, including multi-turn conversations and realistic user noise (typos, text-speak, scrambled word order).
• Robust Intent Recognition: Effectively handles noisy and complex user inputs to identify prepaid card-related requests.
• Function Calling: Emits structured tool calls following the standard FunctionGemma protocol, enabling integration with backend systems.
• Small Footprint: At 270M parameters, it's a compact model suitable for distillation targets and efficient deployment.

Good For

• Chat Agents: Ideal for building conversational AI agents that automate prepaid card services, such as purchasing, balance inquiries, and transaction history retrieval.
• Multi-turn Interactions: Excels in scenarios where users provide information across multiple messages or require clarification.
• Global Applications: Its extensive language training makes it suitable for international deployments where diverse linguistic inputs are expected.

Limitations

• Trained on synthetic data, not real user traffic.
• Language quality is uneven, with English and major languages being most robust.
• The model only emits tool calls; it does not execute actions or access real-time data itself.
• Requires server-side validation of tool arguments due to potential mis-parsing of card numbers under heavy noise.