kelsbeans/qwen3-1.7b-digestive-coach-n390

TEXT GENERATIONPricing:Input $0.32 / Cached $0.064 / Output $1.6Concurrent Unit Cost:1Model Size:2BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Aug 20, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The kelsbeans/qwen3-1.7b-digestive-coach-n390 is a 2 billion parameter Qwen3-1.7B model fine-tuned by kelsbeans using QLoRA on a specialized digestive health dataset. This model is specifically designed as a research artifact for a data-to-behavior fine-tuning project, excelling in parsing and schema validity for digestive wellness scenarios. It serves as a specialized conversational agent for digestive health inquiries, demonstrating high parse rates and schema adherence on its target domain.

Loading preview...

Model Overview

The kelsbeans/qwen3-1.7b-digestive-coach-n390 is a specialized Qwen3-1.7B model, fine-tuned by kelsbeans using QLoRA. It was trained on 390 rows from the kelsbeans/digestive-coach-dataset, focusing on digestive wellness scenarios. This model is provided as a merged fp16 version, loading directly with transformers without requiring peft.

Key Capabilities

  • Specialized Domain Expertise: Fine-tuned specifically for digestive health conversations.
  • High Parse Rate: Achieves a 100% parse rate on a held-out evaluation set, significantly outperforming the base Qwen3-1.7B model (12.5%).
  • Schema Validity: Demonstrates 90% schema validity, a substantial improvement over the base model's 2.5%.
  • Research Artifact: Primarily intended for research into data-to-behavior fine-tuning projects.

Use Cases

  • Digestive Wellness Coaching: Designed to act as a conversational agent for digestive health-related queries.
  • Research and Development: Ideal for researchers exploring fine-tuning techniques on niche datasets and evaluating model behavior in specific domains.

Limitations

  • Not Medical Advice: This model is a research artifact and should not be used for medical advice.
  • Limited Spec Adherence: Achieves 10% spec adherence, indicating room for improvement in strictly following all output specifications compared to higher-N checkpoints.

Finetunes
2 models