← Dashboard  /  News

2026-09-01

AI/Tech Digest 2026-09-01

1. unsloth/gpt-oss-20b-GGUF · Hugging Face [web]
   Features native MXFP4 quantization for MoE layers, enabling large models to run on single H100 GPUs.
   https://huggingface.co/unsloth/gpt-oss-20b-GGUF (huggingface.co)
2. GitHub - xigh/open-weight-models: Curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no EU restrictions. · GitHub [web]
   Includes Mellum 2, a 12B/A2.5B MoE model post-trained with SFT + RL, relevant to mid-size MoE architecture interests.
   https://github.com/xigh/open-weight-models (github.com)
3. AgentRace: Benchmarking Efficiency in LLM Agent Frameworks | OpenReview [web]
   Introduces the first benchmark specifically designed to systematically evaluate the efficiency of LLM agent frameworks.
   https://openreview.net/forum?id=eUuxWAQA5F (openreview.net)