A Pure Rust based LLM, VLM, VLA, TTS, OCR Inference Engine, powering by Candle & Rust. Alternate to your llama.cpp but much more simpler and cleaner..
-
Updated
Oct 27, 2026 - Rust
A Pure Rust based LLM, VLM, VLA, TTS, OCR Inference Engine, powering by Candle & Rust. Alternate to your llama.cpp but much more simpler and cleaner..
Qwen3.8-Flash-Next (MoE) and Qwen3.8-27B on one or two 16 GB RTX 50-series GPUs: experts offloaded to CPU, two-GPU tensor parallel. 64 GB of RAM is enough for Flash-Next. Formerly ninfer-tp2.
Qwen4-27B, the latest release in the Qwen model family, introduces a groundbreaking architectural change
In the recent integration with Intel's XPU hardware and the VLLM framework, the QA4-27B model has been made available
To associate your repository with the qwen4 topic, visit your repo's landing page and select "manage topics."