A developer has created Ophyx, a 350 million parameter Python code model trained from scratch using free compute resources on Kaggle. The model, in its v3 release, achieves a score of 44.5 on HumanEval+ and leads on BigCodeBench among similar-sized open instruct models. Ophyx is built with a custom tokenizer and training pipeline, featuring a decoder-only transformer architecture inspired by Qwen3. The training process involved four stages, including pretraining on a diverse Python dataset and subsequent fine-tuning to improve its performance on coding tasks. AI
IMPACT Demonstrates the feasibility of training capable code models with limited resources, potentially lowering the barrier for individual developers.
RANK_REASON The item describes the training and release of a new, albeit small, language model by an individual developer, detailing its architecture and performance on benchmarks. [lever_c_demoted from research: ic=1 ai=1.0]
- Apache Software License 2.0
- BigCodeBench
- generative pre-trained transformer
- Hugging Face
- HumanEval+
- Kaggle
- llama.cpp
- Ophyx
- Python
- Qwen3
- vLLM
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
