mjpsm/activity-generation-v1.3

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:0.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Sep 7, 2026Architecture:Transformer Featherless Exclusive Cold

mjpsm/activity-generation-v1.3 is a 0.5 billion parameter instruction-tuned causal language model, fine-tuned from Qwen/Qwen2.5-0.5B-Instruct by mjpsm. Optimized for the MyVillage ecosystem, this model specializes in generating the smallest useful next activity based on a village goal, previous activity, knowledge submission, and wisdom. It excels at micro-progression, providing concise, single-action steps and handling vague submissions by requesting clarification.

Loading preview...

Overview

mjpsm/activity-generation-v1.3 is a specialized 0.5 billion parameter language model, fine-tuned from Qwen/Qwen2.5-0.5B-Instruct using LoRA. Its core function is to generate a single, small, and actionable next activity within the MyVillage learning ecosystem. The model was trained on approximately 1.2K synthetic examples, focusing on micro-progression rather than full curriculum generation.

Key Capabilities

  • Micro-Progression: Generates the "smallest useful next activity" based on four contextual inputs: village_goal, previous_activity_title, knowledge_submission, and wisdom.
  • Contextual Activity Generation: Utilizes provided context to create relevant, concise activity titles, descriptions, and single-action instructions.
  • Vague Submission Handling: Designed to ask for clarification or evidence when knowledge submissions are insufficient, preventing the invention of unproven progress.
  • JSON Output Consistency: Trained to consistently produce output in a specific JSON format with activity_title, activity_description, and activity_instructions fields.
  • Assistant-Only Loss Masking: Training focused solely on the activity output fields, enhancing generation quality.

When to Use This Model

  • MyVillage Ecosystem: Ideal for applications within the MyVillage framework requiring dynamic, context-aware activity generation.
  • Micro-Learning Paths: Suitable for systems that guide users through small, incremental steps rather than large assignments.
  • Interactive Learning: Useful for scenarios where a model needs to respond intelligently to user progress, including vague or incomplete updates.

Limitations

As a 0.5B parameter model, it has limited capacity compared to larger LLMs. It may occasionally produce activities that are too broad or introduce unsupported details. It is not intended for high-stakes educational decisions, grading, or full curriculum generation.

Finetunes
2 models