Pinned Loading

1. prismaquant prismaquant Public

   Mixed-precision quantization for LLMs. Every layer refracts into a different format based on its sensitivity. Native compressed-tensors export, validated on Qwen3.6-35B-A3B MoE with MTP speculative…

   Python 103 17

2. tessera tessera Public

   Tessera — embedded-native weight coding. Sub-4.5 bpp weight format terminating on native FP4 tensor cores. Schema prismaquant.tessera.v1, exact-byte accountant, fail-closed parser.

   Python 2 1

3. prismabuild prismabuild Public

   PrismaBuild — deterministic distributed build/dispatch for quantization work. Content-addressed actions, worker attestation, pull-queue transport across a heterogeneous GPU fleet.

   Python 1 1