119 likesNew · 4d old
Uncensored 763B DeepSeek runs on a 12 GB GPU
What is DeepSeek-V4.1-Flash-1017-Uncensored-GSQ-RCO-GGUF?
A 763-billion-parameter DeepSeek model has been quantized so the dense layers no longer refuse unsafe prompts, while the expert layers stay unchanged. The GGUF files stream the large Engram table from an SSD, letting the whole model fit on a single 12 GB GPU via the Strata system.
Who it is for
Developers and researchers who want to experiment with massive language models locally, without internet or built-in safety filters, can finally do it on consumer hardware.
How to try it
Download the GGUF files from Hugging Face and load them with llama.cpp (or any Strata-compatible backend) on a 12 GB GPU.