opencoti-llamafile โ DEVELOPMENT builds for integration testing
These are NOT releases. This repo holds selected development builds of the opencoti llamafile engine (patch series, flags and usage docs are mirrored in the release repo), published so that downstream projects (xollama and others) can integrate and test against work that has not been released yet.
| Releases (stable, pinned, backed up) | ManniX-ITA/opencoti-llamafile |
| This repo | development snapshots, integration testing only |
What to expect
- Updated on request only โ when a development is ready to be tested in integration. It is not a mirror of every dev build, and there is no update cadence.
- A snapshot can be replaced or removed. Pin by commit revision + sha256, never by
mainalone, if you need a build to stay put. - A dev build carries the previous release tag in its version string (for example
opencoti-0.10.5-c7-<YYMMDDHHMM><NNN>). That is the dev build id, not a claim that it is that release. The next release name is decided when it is cut. - Builds here have passed the gates listed in their
BUILD_INFO.md, and nothing more. Found a bug? Report it with the build id and the DSO sha256 (a CUDA-only iteration does not mint a new id).
Layout
pin/llamafile-binary.txt the pin: repo / rev / tag directives + one row per asset
builds/<build-id>/ one directory per published snapshot
opencoti-<ver>-<tag>-<id> host binary (cosmocc APE: runs on Linux x86_64/aarch64, macOS, Windows as .exe)
ggml-cuda.so side-loadable CUDA DSO (place it next to the binary; the exe dir wins the search)
BUILD_INFO.md source commit, patch-chain head, what is NOT yet captured, gates passed
SHA256SUMS
Pin format
Same format the opencoti download client reads for releases
(pin/llamafile-binary.txt in the release repo):
repo ManniX-ITA/opencoti-llamafile-dev
rev <commit>
tag <dev build id>
<kind> <arch> <hf-path> <sha256> kind = bin | dso
Fetch: https://huggingface.co/ManniX-ITA/opencoti-llamafile-dev/resolve/<rev>/<hf-path> and verify
the sha256 from the pin.
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐ Ask for provider support