DeepSeek-V4.1-Flash-1017 (GGUF UD-Q4_K_XL + vision)

DeepSeek (GGUF: unsloth) · DeepSeek
LLMMITpublicTrusted763B params

llama.cpp GGUF build (Unsloth Dynamic Q4_K_XL, MXFP4 experts) of DeepSeek-V4.1-Flash-1017, 763B MoE. The BF16 projector reads images.

DeepSeek-V4.1-Flash-1017 (GGUF UD-Q4_K_XL + vision)

Precision

The routed experts are MXFP4 and the engram table is F8_E4M3, both ggml types of the pinned llama.cpp; the header's file type is MXFP4_MOE, so a task pins Mxfp4. The set has not yet been loaded on a pod.