LLM models
How to find and host an LLM model on your computer
General recipes
- Club 3090 : recipes for 3090/4090/5090 owners.
- MiaAI-Lab : various recipes for popular models.
- Unsloth
Qwen3.8-27B
Running the Qwen3.8-27B model
- HyperQwen : Qwen3.8-27B on a single 24 GB card with vLLM.
- MiaAI-Lab for one 16-24 GB GPU using the EXL3 quantization format.
- ninfer-4090 : Qwen3.8-27B on one RTX 4090. Based on
ninfer. - MiaAI-Lab (DGX Spark) on one DGX Spark.
Running with llama.cpp or GGUF-compliant engines:
- ISTA: GSQ-RCO : small quants of ISTA lab.
- Swift-1.5-Qwen3.8-27b : UkisAI’s reasoning-efficient derivative of Qwen3.8-27B.
- Unsloth Qwen3.8-27b
Qwen3.8-Flash-Next (Q38FN)
Running the Qwen3.8-Flash-Next model
- Unsloth Q38FN
- MiaAI’s recipe on one DGX Spark with TensorFold.
- blazux’s recipe on a single DGX Spark with vLLM.
- myllmbox’s recipe on a single DGX Spark with vLLM.
- colab-qwen38-g4 : runs Q38FN on a Google Colab G4 GPU (96 GB VRAM).
Deepseek v4.1 Flash
- sybil’s recipe : DeepSeek-V4.1-Flash EXL3 on one 24 GB RTX 3090 + DDR4 + NVMe.
- jayleaton’s recipe : DeepSeek-V4.1-Flash on 2x DGX Spark with TensorFold.
- MiaAI’s recipe : DeepSeek-V4.1-Flash-1017 EXL3 on 2x DGX Sparks with TensorFold.
Gemma4
Speech recognition and text-to-speech (TTS)
Document processing
Biology
Image generation
Last updated on