build-small-hackathon/pozify-coach-summary1

TEXT GENERATIONConcurrent Unit Cost:1Model Size:7.6BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jun 14, 2026Architecture:Transformer Featherless Exclusive Cold

The build-small-hackathon/pozify-coach-summary1 is a 7.6 billion parameter LoRA adapter fine-tuned from Qwen/Qwen2.5-7B-Instruct. Developed by build-small-hackathon, this model is specifically designed for generating grounded coach summaries in JSON format from structured evidence and knowledge cards. It specializes in producing structured outputs for coaching applications, leveraging a 32768 token context length.

Loading preview...

Model Overview

The build-small-hackathon/pozify-coach-summary1 is a 7.6 billion parameter LoRA adapter, built upon the Qwen/Qwen2.5-7B-Instruct base model. It is specifically fine-tuned for generating structured coach_summary.json outputs.

Key Capabilities

  • Grounded Summary Generation: Specializes in creating coach summaries based on provided structured evidence and knowledge cards.
  • Structured Output: Designed to produce JSON-formatted outputs, ensuring consistency for downstream applications.
  • Domain-Specific Fine-tuning: Trained on a small, focused dataset of coach summaries and public fitness style data, indicating a specialization in coaching-related content generation.

Training Details

The model was fine-tuned using a limited dataset of 22 training rows and 5 evaluation rows, with 4 style rows mixed in. It is packaged as a merged, inference-ready Transformers checkpoint.

Limitations

Initial evaluations show low rates for JSON validity (0.4), verifier pass rate (0.0), and section completeness (0.0), suggesting that the model's output quality for these metrics is currently limited.