This combines the unchanged IQ2_XXS backbone of ISTA-DASLab/DeepSeek-V4.1-Flash-1017-GSQ-RCO-GGUF with the Engram tables at the full original precision of deepseek-ai/DeepSeek-V4.1-Flash-1017 (MXFP8, 202.8 GB) in place of ISTA's 4-bit IQ4_NL Engram.
The official FP8 E4M3 values and E8M0 per-32 scales (revision 1bae0e10) were copied byte-for-byte into ISTA's Engram GGUF layout as ggml type 43 (MXFP8), verified byte-exact over all 768M rows and checked with Strata's pack_gguf.py --engram; scripts in tools/, results in verification.log.