All news

Matching stories Newest first

1. Oct 20
   Lr/LocalLLaMACommunity

   Reddit user runs Qwen 3.8 Flash Next locally on an 8GB GPU under CachyOS

   The poster reports about 30 tokens per second on average and bursts up to 48 tokens per second, using a 3-bit Unsloth quant with a 42k context that nearly exhausts VRAM.

   • Lab: Alibaba
   • Model: Qwen 3.8 Flash Next
   • Topic: product