7 posts tagged by Digg
AI
Harness Hub reports higher success rates for its Pi-specific Qwen3.8-27B adapter, with thinking off and benchmark-informed checkpoint selection.
Technology
Nous reports Chinese-state framing fell from 22% to 0.9% in its own evaluation of the fine-tuned model.
AI
A user says a compressed MLA copy fits an RTX 4090 and holds about 3.5x more context than base Qwen.
AI
SGLang says it turned Qwen3.8-27B into a multimodal decision model and added a /v1/decisions endpoint for using language and vision-language models in classification and scoring.
AI
tinygrad says its small size and pure-Python code make it particularly easy to build on.
AI
The 27B model is now available on Ollama for agentic coding and professional tasks.
AI
Official account shares video for model focused on coding and cowork tasks.