deepseek-v4.1POST
Running V4.1 Flash at EXL3 3.75bpw on 3 DGX Sparks: 4M context, no OOM
A user runs DeepSeek-V4.1-Flash-1017 as EXL3 3.75bpw on TensorFold TP3 across three DGX Sparks, getting near-full weights, a stable 4M context with no OOM, and uses it as a daily driver with a full recipe.
Thomas Eastham
@easthambuildsArchived 2 days agoOriginally posted 2026.10.27
Thomas Eastham @easthambuildsOriginally posted 2026.10.27
@marc_jeffrey18 @MiaAI_lab Works today if you're OK with a quant: I run DeepSeek-V4.1-Flash-1017 as EXL3 3.75 bpw on TensorFold TP3 across 3 Sparks. Full recipe: https://github.com/easthambuilds/DeepSeek-V4.1-Flash-1017-EXL3-3.75bpw-3x-DGX-Sparks-TensorFold Nearly full weights and still gives good context size of 4M. stable without OOM issues. This is my daily driver.
x.com/easthambuilds/status/2114840389928587299Archived 2026.10.27 12:09