“vllm” 관련 오픈소스 프로젝트

Serving DeepSeek-V4.1-Flash (552B MoE) across four NVIDIA DGX Sparks with TensorFold and vLLM, with launch scripts, patches and measured results.

Python library that detects LLM decoding corruption such as repetition loops and language drift mid-stream, so bad tokens can be cut before users see them.

A 9B-parameter open-weight multimodal model for visual understanding, spatial reasoning, tool use, and embodied-AI research, built on Qwen3.5 and C-RADIOv4-H.

Spark-X2.5 is an open-weight language model series (4B and 1.7B) aimed at agentic coding and tool use on local and on-device hardware.