A new fine-tuned model named Sabi-Noema, based on Qwen 3.8 27B, has been released, focusing on enhanced capability and efficiency for models under 30 billion parameters. This fine-tune demonstrates improved performance in coding, agentic tasks, and tool-calling, outperforming other models in its size class on benchmarks like DeepSWE and TB2.1. Additionally, Sabi-Noema-MLA, a variant, significantly reduces token generation and memory usage through MLA attention, enabling much larger context windows on consumer hardware. AI
IMPACT This fine-tune demonstrates significant advancements in smaller model capabilities, potentially lowering the barrier to entry for sophisticated AI applications.
RANK_REASON Release of a fine-tuned open-source model with benchmark results. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
