jujuloaiza/Qwen3-1.7B-riddles is a 1.7 billion parameter model fine-tuned from Qwen/Qwen3-1.7B using QLoRA (4-bit) with a context length of 32768 tokens. This model is specifically designed to generate riddles based on various topics and targeted age groups, providing an answer that is not immediately revealed. It is intended for use in the CS-394/594 class at DigiPen as a test model for riddle generation and answer evaluation.
Loading preview...
Model Overview
jujuloaiza/Qwen3-1.7B-riddles is a specialized language model, fine-tuned from the Qwen/Qwen3-1.7B architecture using QLoRA (4-bit) with supervised fine-tuning. It has 1.7 billion parameters and a context length of 32768 tokens. This model was developed as a test model for the CS-394/594 class at DigiPen.
Key Capabilities
- Riddle Generation: Generates various riddles based on a list of topics, targeting different age groups.
- Answer Concealment: Provides an answer to the generated riddle but does not immediately reveal it.
- Answer Evaluation: Can evaluate user-provided answers, responding with "Correct" or "Incorrect."
Training Details
The model was trained on the jujuloaiza/riddletraining dataset over 3 epochs, utilizing a LoRA rank of 16 and an alpha of 32, with a learning rate of 0.0002.
Limitations
- Limited Topics: The model currently has a limited range of riddle topics.
- Short Answers: Primarily trained on short answers for riddles.
- Riddle Length: Tends to generate mostly short riddles.
- Strict Evaluation: The model is strict in its answer evaluation and may not robustly assess user answers.