Downloads · 30 days
5
23% of all-time downloads
R3n3r0/dapack-catalog
dapack-catalog is a machine learning model from R3n3r0. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as mit.
The glue that turns the three dapack packs into one routed catalogue:
Downloads · 30 days
5
23% of all-time downloads
All-time downloads
22
Public
Repo size
146 MB
Likes
0
Public
Click a slice to open those files.
.gguf146 MB · 100%
From the Hugging Face model README
The glue that turns the three dapack packs into one routed catalogue:
| file | what it is |
|---|---|
manifest.json | domains, measured capability manifests, TF-IDF detector, embedding-router centroids, file hashes |
embed/nomic-embed.gguf | the embedding model for the ensemble router (nomic-embed-text-v1.5 Q8, 140 MB) |
Download the three packs and this repo's files into one directory:
mycatalog.dapack/
manifest.json <- from this repo
embed/nomic-embed.gguf <- from this repo
gguf/math.gguf <- R3n3r0/dapack-math (graded_math.gguf)
gguf/language.gguf <- R3n3r0/dapack-language (graded_language.gguf)
gguf/code.gguf <- R3n3r0/dapack-code (graded_code.gguf)
Then run it with the dapack runtime (binaries under Releases, nothing to compile):
./dapack serve mycatalog.dapack --addr :8080
--router tfidf bag-of-words, zero extra memory, ~0.1 ms
--router embed dense centroids, ~145 MB VRAM, ~5 ms per request
--router ensemble both mixed 30/70 — 44% → 89% routing accuracy on a
frozen holdout; the two err on different prompts
--router auto (default) ensemble when this repo's files are present,
tfidf otherwise
Whichever router runs, the capability gate is identical: a request needing a capability a pack has measurably lost is routed to a pack that has it, or refused with a reason.
Runtime, measurements and docs: https://github.com/R3n3r0/dapack