kelsbeans/qwen3-1.7b-digestive-coach-n1102

TEXT GENERATIONPricing:Input $0.32 / Cached $0.064 / Output $1.6Concurrent Unit Cost:1Model Size:2BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Aug 22, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

kelsbeans/qwen3-1.7b-digestive-coach-n1102 is a 1.7 billion parameter Qwen3 model fine-tuned by kelsbeans for digestive wellness coaching. This model specializes in generating responses related to digestive health, demonstrating significantly improved parsing, schema validity, and spec adherence compared to its base model. It is designed as a research artifact for data-to-behavior fine-tuning projects, not for medical advice.

Loading preview...

Model Overview

kelsbeans/qwen3-1.7b-digestive-coach-n1102 is a specialized Qwen3-1.7B model, fine-tuned by kelsbeans using Unsloth QLoRA on a custom dataset of 1102 rows from the kelsbeans/digestive-coach-dataset. This model is designed to function as a digestive wellness coach, providing structured and relevant responses within this domain. It is provided as a merged fp16 model, loadable directly with transformers.

Key Capabilities & Performance

This fine-tuned model shows substantial improvements over the base Qwen3-1.7B on a held-out, family-disjoint evaluation set:

  • Parse Rate: Achieves 100.0% compared to 12.5% for the base model.
  • Schema Validity: Reaches 85.0% versus 2.5% for the base model.
  • Spec Adherence: Demonstrates 25.0% adherence, a significant increase from 0.0%.
  • Field Accuracy: Improved to 44.0% from 26.7%.

Judge evaluations (using Anthropic's Claude-Sonnet-5) indicate high scores for integrated voice (4.58), honest uncertainty (4.30), and no manufactured agreement (4.36). The model also shows improved robustness and buried-red-flag recall compared to earlier iterations.

Use Cases

This model is primarily intended as a research artifact for exploring data-to-behavior fine-tuning in specific domains. It is suitable for developers and researchers interested in:

  • Developing specialized conversational agents for health and wellness topics.
  • Studying the impact of targeted fine-tuning on model behavior and response quality.
  • Experimenting with domain-specific instruction following and structured output generation.

Important Note: This model is not intended to provide medical advice.

Finetunes
2 models