Who it is forAn operator who wants the Flash-Next IQ3_XXS model served by strata and has the upstream GGUF shards on hand.
The problemThe strata engine needs a pack directory and a draft directory that are not made by the front end, and the steps are easy to get wrong.
The outcomeA pack directory and MTP draft directory ready to pass to strata through the --pack and --mtp options.
What you need first
- Linux x86-64 machine with one NVIDIA GPU (Pascal or newer) and CUDA 12.x
- git and a C++ build toolchain able to run make build
- GGUF model files copied to the machine by hand, since the fork does not download them
- Optional: the mmproj projector file from the same repository as the model, for picture input
Steps
- Install the Python tool requirements listed in tools/requirements.txt (numpy and regex), with gguf-py from the submodule.
- Run
python tools/iq_pack.py --gguf --outwith your GGUF path and output directory (add --compat-bf16 for Unsloth files). - Copy the published MTP directory from https://huggingface.co/robertkarl/UnbiasedStrataPacks instead of fetching and quantizing it, or run
STRATA_SOURCE=huggingface python tools/mtp_fetch.py fetch --out. - Run
python tools/mtp_pack.py --src --experts q2_0 --out /mtp-q2_0.ggufandpython tools/mtp_rt.py --gguf /mtp-q2_0.gguf --out /rt, thencp data/draft_vocab.bin /rt/draft_vocab.bin.
Prompt for your agent
In the UnbiasedStrata repo at <repo-path>, prepare the Flash-Next pack from the GGUF shard at <gguf-path> and report the output directory and its file list. Verify that draft_vocab.bin is present in the MTP directory before reporting success.
Caveat. The README's commands leave some arguments blank, so fill in the paths yourself, and the published MTP files are the ones the author hashed, so check your copy against the listed hashes.
Get 5 new use cases every morning. Free, one short email, unsubscribe anytime.
Get them freeShare: X LinkedIn Hacker News Email Copy link
Disclaimer
Deduped lists open-source repositories that belong to other people. We do not own, maintain, audit, or vouch for any of them, and a repo appearing here is not an endorsement or a safety check. The use cases, steps, and prompts are written by AI and may be incomplete, outdated, or just wrong. Where a use case is marked "Seen in the wild", we link to public posts where someone described something similar, but we have not verified those posts and the use case itself is still our AI's own write-up. Use all of this at your own risk. Only run code, commands, or agents on systems and accounts you are authorized to use, and read the code and every command before you run it, especially anything that touches credentials, files, or the network. Nothing here is professional, security, legal, or financial advice. All trademarks and project names belong to their respective owners, and their use here does not imply affiliation. Content is provided as is, without warranties of any kind. Found a mistake, a safety problem, or something you want removed? Email [email protected] and we will look at it promptly. Full disclaimer page