Every verb is a batch over buffer-protocol inputs, and every output whose length the call decides lands in an arena you reuse.
uform.Modelencodes text, images, video and audio into pooled embeddings, per-token states and token vectors, and scores calibrated choices over texts.uform.Trainertrains one on GPUs, joined across hosts by auform.Mesh.uform.Executorowns the threads or streams every call runs on, apart from any model.
Topics
| Installation1 | The wheels carry the CPU library. |
|---|---|
| Quickstart | |
| Receivers and Deferred Calls15 | A receiver carries what every C argument structure puts before its inputs: the executor and the completion. |
| Executors39 | uform.Executor.forram(threads=0, , memorydomains=None, cpuclass=None) creates an executor of host threads, and uform.Executor.forvram(devices=None, , scratchlimit=0, gpuclass=None) opens one stream per device. |
| Arenas7 | Every output whose length only the call knows, the embeddings, the states and their offsets, lands in a uform.Arena you pass, and the verb answers views over it without copying. |
| Encoding20 | |
| Choices4 | |
| Models24 | uform.Model.loadram(path) and uform.Model.loadvram(path, devices=[0]) copy a safetensors checkpoint into host memory or onto GPUs, and loadrambuffer and loadvrambuffer take the checkpoint's bytes instead. |
| Training98 | Training is GPU-only. |
| Errors16 | |
| Design |