machalek29/qwen3-0.6b-state-lifetime-tutor-n500-v2

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:0.8BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Aug 19, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The machalek29/qwen3-0.6b-state-lifetime-tutor-n500-v2 is a 0.8 billion parameter Qwen3-based model with a 32768 token context length, specifically fine-tuned to act as a Python state-lifetime tutor. It identifies mutable-state lifetime bugs in short Python programs and asks a single, non-compound question about object creation, ownership, or shared references. This model is designed for educational assistance in debugging Python code related to object state and lifetime.

Loading preview...

Python State-Lifetime Tutor

This model, machalek29/qwen3-0.6b-state-lifetime-tutor-n500-v2, is a specialized 0.8 billion parameter Qwen3-based language model designed to tutor users on Python state-lifetime bugs. It operates by identifying a single mutable-state lifetime bug within a provided Python program and then formulating exactly one non-compound question related to the object's creation, ownership, or shared references. The model explicitly avoids providing corrected code or stating the correction directly.

Key Capabilities

  • Bug Identification: Pinpoints relevant declarations, assignments, or mutations associated with a mutable-state lifetime bug.
  • Targeted Questioning: Generates a single, focused question to guide the user's understanding of the bug.
  • Specialized Tutoring: Acts as a tutor for Python object state and lifetime concepts, rather than a code corrector.
  • High Adherence: Achieved 96% spec adherence on 24 clean held-out scenarios and 100% robustness on 12 adversarial scenarios during evaluation.

Training and Usage

The model was fine-tuned using LoRA on the machalek29/state-lifetime-tutor-v2 dataset, specifically the first 500 examples. It requires a specific system prompt and greedy decoding (do_sample=False) with thinking off for optimal performance, as it was trained under these conditions. The base model used was Qwen/Qwen3-0.6B.