GInfer model catalog
catalog.json is the list of GInfer model packages that GChat offers for download. GChat reads this one file from
this repository; you do not need to open it yourself.
What it contains
Each model entry lists its packages. For every package the catalog gives:
- the package name and the model identity;
- the download URL, the exact size in bytes and the SHA-256 of the file, which GChat checks after downloading;
- the tensor-parallel size (1 for one GPU, 2 for two GPUs);
- where it runs: the NVIDIA GPU architectures and the least memory per GPU it was tested with, or the processor types (AVX2, AVX-512, AMX) for a package that runs on the CPU;
- what the model supports (for example tools and vision);
- the tested launch settings, which GChat offers when you load the model.
Format
The file is JSON. manifest_version and schema_version are both 1; updated_at is when the file was generated;
models is the list of model entries, each with a releases list. GChat accepts only schema version 1.
Downloads
Every URL points to one file in one of the SectileLabs model repositories and is pinned to that repository's commit
(.../resolve/<commit>/<file>), so a download is always the exact file listed here, even if the repository gets new
commits later. The size and SHA-256 in this catalog equal the file's Hugging Face metadata.
The packages run only on GInfer and GChat. They are not safetensors or GGUF files. The licenses of the models are in each model repository.