Exporters: run exported models (`ExportedModel`) #50032
IlyasMoutawwakil
wants to merge
18 commits into
exporters-artifacts
from
exporters-forward-runtime
echarlaix
approved these changes
on 2026-10-28
Exporters: run single-graph exports (ExportedModel, AutoExportedModel)
54eb6f1d
Exporters: sort the runtimes import, return only tensors from the Exe…
942a0a73
Exporters: rebuild the model's outputs from the traced specs, drop Au…
7de7cb11
IlyasMoutawwakil
changed the title
Exporters: register the cache classes a saved export's specs name bef…
4f164d4e
Exporters: keep ONNX Runtime CUDA inputs and outputs on the GPU (IO b…
c8818751
Exporters: run ONNX through IO binding on every device, save a runtim…
fa127e11
Exporters: bind only tensor inputs on ONNX, copy OpenVINO outputs once
6b9fba9b
Exporters: feed ExecuTorch string constants as None, keep non-string …
69c33a98
Exporters: let runtime() pass runner options, ONNX Runtime providers …
b7a1d1b4
Exporters: compile OpenVINO in the export's precision on CPU, take co…
c4f40cf5
Exporters: choose per call whether a runner shares its inputs and out…
ef7dacfa
Exporters: give a torch program the scalars it was traced with, in it…
3b861dd7
Exporters: record a None kwarg as a constant too
62191414
Exporters: bind ONNX inputs by name, run bf16 convolutions in fp32 on…
6d80fa4b
Exporters: feed SDPA's boolean mask to ONNX as an additive one
fc78285f
Merge branch 'exporters-artifacts' into exporters-forward-runtime
fc65352a
Exporters: read bf16 OpenVINO outputs as bf16
41c68b16
Exporters: split the CPU between torch and OpenVINO (`torch_threads`)
90024083
Login to write a write a comment.
Login via GitHub