GuruVarshini/supportlm-qwen2.5-3b

TEXT GENERATIONConcurrent Unit Cost:1Model Size:3.1BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jun 22, 2026Architecture:Transformer0.0K Featherless Exclusive Cold

GuruVarshini/supportlm-qwen2.5-3b is a 3.09 billion parameter Qwen2.5-3B-Instruct model fine-tuned for Indian e-commerce customer support. It specializes in processing code-switched Hinglish and Tanglish messages, classifying issue types, extracting order IDs, and generating empathetic responses. This model is optimized for specific customer service automation tasks in multilingual Indian contexts.

Loading preview...

SupportLM: Specialized Customer Support for Indian E-commerce

GuruVarshini/supportlm-qwen2.5-3b is a fine-tuned Qwen2.5-3B-Instruct model specifically designed for customer support in the Indian e-commerce sector. Its primary differentiation lies in its ability to understand and respond to code-switched Hinglish and Tanglish messages, which are common in the region.

Key Capabilities

  • Issue Classification: Accurately categorizes customer queries into 8 predefined types (e.g., order_status, return_request, payment_issue). Achieves 79.9% classification accuracy on held-out test data, outperforming GPT-4o-mini.
  • Order ID Extraction: Reliably extracts order IDs from customer messages with 99.4% accuracy.
  • Empathetic Response Generation: Crafts empathetic responses in the same code-switched language used by the customer, maintaining a 100% language match rate.

Training and Performance

The model was trained on 1277 synthetic Hinglish/Tanglish examples using QLoRA and DPO methods. It leverages the Qwen/Qwen2.5-3B-Instruct as its base model. The SFT (Supervised Fine-Tuning) version was selected for deployment due to its superior classification accuracy. This model is ideal for automating initial customer support interactions for Indian e-commerce platforms.