{"slug": "introducing-zy-s2-a-1-5b-open-source-coding-model-by-zyr-ai", "title": "Introducing ZY-S2: A 1.5B Open-Source Coding Model by ZYR AI", "summary": "ZYR AI released ZY-S2, a 1.5-billion-parameter open-source coding model fine-tuned from Qwen2.5-Coder-1.5B and published on Hugging Face under Apache 2.0. ZYR AI reported that ZY-S2 scored 68.0% on its SLMFrontier coding benchmark and 74.2% on HumanEval+, versus 63.5% and 63.2% for the Qwen2.5-Coder-1.5B baseline, with all models run through the same CPU llama.cpp harness at Q5_K_M quantization. The model is optimized primarily for Python, ships in GGUF and Safetensors formats, and its Q5_K_M GGUF is roughly 1.1 GB, allowing local CPU inference via Ollama; ZYR AI noted the confidence intervals overlap and that math reasoning, tool use, agentic tasks and safety remain weak areas.", "body_md": "Hey Hugging Face community!\n\nWe’re excited to officially introduce **ZY-S2**, a lightweight, open-source AI model developed by **ZYR AI**, focused on coding performance, efficiency, and accessibility.\n\nOur goal is simple: **Small models deserve big capabilities.**\n\n **Model:** [zyr-AGENT/ZY-S2 · Hugging Face](https://huggingface.co/zyr-AGENT/ZY-S2)\n\nZY-S2 is a 1.5-billion-parameter coding model built through fine-tuning Qwen2.5-Coder-1.5B.\n\nRather than focusing on increasing parameter counts, we explored how far targeted post-training could push a compact model.\n\n**Key highlights:**\n\n1.5B parameters\n\nOptimized primarily for Python coding\n\nTrained with a focus on code generation and problem-solving\n\nSupports local CPU inference\n\nAvailable in GGUF and Safetensors formats\n\nOpen-source under Apache 2.0\n\nWe evaluated ZY-S2 against established small language models using our SLMFrontier coding benchmark.\n\n| Model | Coding Score | HumanEval+ | MBPP+ | \n|---|---|---|---|\n| **ZY-S2 (1.5B)** | **68.0%** | **74.2%** | 63.7% | \n| Qwen2.5-Coder-1.5B | 63.5% | 63.2% | 63.7% | \n| Qwen2.5-Coder-0.5B | 47.0% | 52.1% | 43.5% | \n| Llama-3.2-1B | 38.0% | 27.0% | 45.6% | \n\nAll models in this coding comparison were evaluated using the same CPU llama.cpp harness and Q5_K_M quantization.\n\nZY-S2 achieved a **68.0% coding score**, compared with **63.5% for its 1.5B Qwen baseline**.\n\nThese are promising results for a compact model, although confidence intervals overlap and performance on other tasks varies.\n\nYou can run the quantized model using Ollama:\n\n```\nollama run hf.co/zyr-AGENT/ZY-S2:Q5_K_M\n```\n\nThe Q5_K_M GGUF is approximately 1.1 GB, making local experimentation possible without expensive GPU hardware.\n\nZY-S2 is not a general-purpose frontier model. Its main strength is Python coding, while mathematical reasoning, tool use, agentic tasks, and safety remain areas for improvement.\n\nWe believe transparent evaluations, including limitations, are essential to meaningful open-source AI development.\n\nWe’re sharing ZY-S2 with the community to encourage independent testing, feedback, and collaboration.\n\nWe’d love to hear:\n\nHow does ZY-S2 perform on your coding tasks?\n\nWhat should we prioritize in future releases?\n\nWhich benchmarks would you like to see next?\n\nWhat would make compact coding models more useful for real-world developers?\n\n If you find our work interesting, consider supporting the project by liking the model repository or following **ZYR AI** on Hugging Face. Your feedback and support mean a lot!\n\n **Try ZY-S2:** [zyr-AGENT/ZY-S2 · Hugging Face](https://huggingface.co/zyr-AGENT/ZY-S2)\n\n*Built by ZYR AI — Pushing the boundaries of small language models.*", "url": "https://wpnews.pro/news/introducing-zy-s2-a-1-5b-open-source-coding-model-by-zyr-ai", "canonical_source": "https://discuss.huggingface.co/t/introducing-zy-s2-a-1-5b-open-source-coding-model-by-zyr-ai/191260#post_1", "published_at": "2026-10-31 09:21:45+00:00", "updated_at": "2026-10-31 09:49:36.209298+00:00", "lang": "en", "topics": ["large-language-models", "ai-research", "ai-tools", "developer-tools"], "entities": ["ZYR AI", "ZY-S2", "Qwen2.5-Coder-1.5B", "Hugging Face", "Ollama", "Llama-3.2-1B", "SLMFrontier"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/introducing-zy-s2-a-1-5b-open-source-coding-model-by-zyr-ai", "markdown": "https://wpnews.pro/news/introducing-zy-s2-a-1-5b-open-source-coding-model-by-zyr-ai.md", "text": "https://wpnews.pro/news/introducing-zy-s2-a-1-5b-open-source-coding-model-by-zyr-ai.txt", "jsonld": "https://wpnews.pro/news/introducing-zy-s2-a-1-5b-open-source-coding-model-by-zyr-ai.jsonld"}}