Non-autoregressive System 1 decision engine. Typed choice, score and yes/no decisions over any text in a single forward pass, in 100+ languages, with a router that picks the right checkpoint per request.
Python calibrationclassificationdecision-modelhuggingfacejevmodernbertmultilingualnlppythonpytorchroutingtyped-decisionszero-shot
docs(docker): the GPU image installs cu130 wheels, not CUDA 12.8
Adeshina Lasisi806be5ba304b
diff --git a/Dockerfile b/Dockerfileindex 76bda0ce..e1b8a26a 100644--- a/Dockerfile+++ b/Dockerfile@@ -11,9 +11,9 @@ ENV PIP_NO_CACHE_DIR=1 \RUN python -m venv --upgrade-deps /opt/venvENV PATH="/opt/venv/bin:$PATH"-# CPU by default on AMD64 and ARM64. The CUDA override selects cu128 and the-# DGX Spark override selects cu130. Bump TORCH_VERSION deliberately; the check-# below fails the build if the wheel does not match the requested index.+# CPU by default on AMD64 and ARM64. Both the CUDA override and the DGX Spark+# override select cu130. Bump TORCH_VERSION deliberately; the check below fails+# the build if the wheel does not match the requested index.ARG TORCH_INDEX=cpuARG TORCH_VERSION=2.14.0COPY docker/check_torch.py /opt/check_torch.pydiff --git a/docs/docker-platforms.md b/docs/docker-platforms.mdindex 36c2d86d..9743d6fc 100644--- a/docs/docker-platforms.md+++ b/docs/docker-platforms.md@@ -7,7 +7,7 @@ an ARM64 CPU does not imply an NVIDIA GPU.| --- | --- | --- || Linux AMD64 CPU | `compose.yaml` | CI build and checks; CPU inference || Linux ARM64 CPU | `compose.yaml`, built on the ARM64 host | CI build and checks on a native ARM64 runner; CPU inference |-| Linux AMD64 NVIDIA | add `compose.cuda.yaml` (CUDA 12.8) | CUDA inference on an RTX 4070 Ti with the base quickstart |+| Linux AMD64 NVIDIA | add `compose.cuda.yaml` (CUDA 13.0) | CUDA inference on an RTX 4070 Ti with the base quickstart || DGX Spark | add `compose.spark.yaml` (ARM64, CUDA 13.0), with or without `compose.http.yaml` | CI builds and loads the CUDA libraries without a GPU; Spark inference not yet reported || Apple Silicon | Linux ARM64 container on CPU | See [Apple Silicon](#apple-silicon) |diff --git a/docs/docker.md b/docs/docker.mdindex 99a43e67..dcabf649 100644--- a/docs/docker.md+++ b/docs/docker.md@@ -26,7 +26,7 @@ For ARM64 hosts, DGX Spark and Apple Silicon, seeInstall a compatible NVIDIA driver and configure Docker with the[NVIDIA Container Toolkit](https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/latest/install-guide.html).-The GPU image uses PyTorch CUDA 12.8 wheels. Check your GPU's compute capability+The GPU image uses PyTorch CUDA 13.0 wheels. Check your GPU's compute capabilityand driver against [PyTorch's supported builds](https://pytorch.org/get-started/locally/);older cards may require a different build. Allow additional disk space for CUDAlayers. VRAM needs depend on the checkpoint, batch size and input length.