Local deploy notes: DeepSeek V4.1 Flash ISTA IQ2_XS pack runs in Maya unchanged
A user documents copying the deepseek-v4.1-flash-ist-das-2.25bpw pack into Maya with a new config, running across four GPUs with no code or file conversion.
notes from my agent ❯ what did we do to take itsa model into maya - it just worked if i recall? Yes, it just worked. Maya needed no code changes and no conversion of the ISTA files. Here's what the setup involved: 1. Copied Strata's pack folder. Strata's setup step had already built a small index folder for the model, Strata-data/packs/deepseek-v4.1-flash-ist-das-2.25bpw. It says where each tensor sits in the ISTA GGUF files, and it holds the tokenizer. I copied it to maya-models/ista-iq2xs-pack, because Strata's engine is a fork of Maya and still uses Maya's pack format. The copy points at the original GGUFs in strata-models/, so the weights weren't duplicated. 2. Wrote a new config, maya-src/maya-ista.json, based on the existing maya-ds4.json. It points --pack and the tokenizer at that folder, sets the model name to deepseek-v4.1-flash-ista-iq2xs, gives it its own log file, and uses all four GPUs. I also removed the old fixed layer split and env overrides, so Maya picks its own split for this model.