mjpsm/activity-generation-v1.2-qwen0.5b-merged

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:0.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Sep 2, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The mjpsm/activity-generation-v1.2-qwen0.5b-merged model is a 0.5 billion parameter Qwen2.5 causal language model, fine-tuned from Qwen/Qwen2.5-0.5B-Instruct. It is specifically designed for generating a single, small, next learning activity within the MyVillage learning ecosystem. The model takes learner context (goal, previous activity, knowledge submission, and villager wisdom) and outputs structured JSON containing a title, description, and instructions for the next micro-activity. Its primary differentiation lies in its specialized fine-tuning for educational activity generation, focusing on short, focused tasks and intelligent handling of vague learner submissions.

Loading preview...

Activity Generation V1.2 - Qwen2.5-0.5B Merged

This model, mjpsm/activity-generation-v1.2-qwen0.5b-merged, is a specialized 0.5 billion parameter language model built upon Qwen/Qwen2.5-0.5B-Instruct. It has been fine-tuned using Supervised Fine-Tuning with LoRA, specifically for the MyVillage Activity Generation V1.2 pipeline. Its core function is to generate a single, concise next learning activity based on four key pieces of learner context: a village_goal, previous_activity_title, knowledge_submission, and villager_wisdom.

Key Capabilities and Design Principles

  • Structured Output: The model consistently produces output in a precise JSON format, including title, description, and instructions for the activity.
  • Contextual Activity Generation: It intelligently uses learner context to suggest relevant next steps, prioritizing the overall learning goal and learner's demonstrated understanding.
  • Handling Vague Submissions: A key feature is its ability to avoid assuming mastery from vague submissions (e.g., "done"), instead generating clarification-oriented activities.
  • Micro-Activity Focus: Activities are designed to be short, focused, and contain one main learner action, avoiding large projects or complex assignments.
  • Wisdom Integration: It incorporates "Villager Wisdom" as secondary guidance without directly quoting or citing it.
  • Efficiency: As a 0.5B parameter model, it supports CPU inference, though GPU is recommended for lower latency.

Intended Use Cases

This model is ideal for educational platforms requiring automated generation of personalized, bite-sized learning activities. It is particularly suited for systems that need to:

  • Dynamically adapt learning paths based on learner progress.
  • Generate actionable, short-term learning tasks.
  • Ensure activities align with overarching learning goals.
  • Handle diverse learner inputs, including incomplete or vague responses, by prompting for clarification rather than advancing prematurely.

Limitations

As a small, specialized model, it may produce malformed JSON on out-of-distribution inputs, repetitive activities, or exhibit weaker reasoning on complex learner histories. Applications should implement validation and guardrails for generated outputs.