AI Briefing
KO

Qwen 3.6 27B local 20 TPS

·2026.04.24 16:13

Key point

On a combination of m720q i7, 64GB RAM, and T1000 8GB, Qwen 3.6 27B recorded 20 TPS.

Details

They shared that running Qwen 3.6 27B locally on a combination of an m720q i7 system, 64GB RAM, and Nvidia T1000 8GB VRAM achieved 20 TPS.

The run settings are as follows.

  • -c 131072
  • -n 32768
  • --no-context-shift
  • --temp 0.6
  • --top-p 0.95
  • --top-k 20
  • --repeat-penalty 1.00
  • --presence-penalty 0.00
  • --fit on
  • -fa on
  • -ctk q8_0 -ctv q8_0
  • --chat-template-kwargs '{"preserve_thinking": true}'

They added that for long-running use, they are running Pi 24/24.

qwengguflocal-llmllama-cpp
Read originalShare on Threads

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.

Related

RedditDynamic Quantiser Released, Improving Accuracy Over Standard Quantization in VRAM-Constrained EnvironmentsHuggingFace BlogHugging Face Adds GGUF Model Support to TransformersRedditHugging Face Introduces Native GGUF Support in TransformersHacker NewsApple Releases LensVLM-9B for Processing Compressed Text ImagesTLDR AIQwen Releases 'RecreationWorld' for Training AI Agents to Rebuild AppsHacker NewsOllaya Launches Local Runtime for Open-Source Decision Models with TypeSafe SDK Compatibility