v23.05.00

What’s Changed

⚠ Breaking Changes

• update feast op to handle feast 0.31 - latest @jperez999 (#344)
• Condense the WorkflowRunner sub-classes @karlhigley (#319)
• Convert Workflow outputs to Triton-friendly format using TensorTable @bschifferer (#317)
• Fix Tensorflow serving to make Merlin example tests pass again @karlhigley (#298)
• Switch from __lengths to __offsets for ragged list features @karlhigley (#296)

🐜 Bug Fixes

• Fix the code that aligns returned data with the output schema @karlhigley (#335)
• Convert Workflow outputs to Triton-friendly format using TensorTable @bschifferer (#317)
• Make minor adjustments to the way Workflow outputs are formatted @karlhigley (#314)
• Revert "Remove the Triton-specific Workflow execution code (#309)" @karlhigley (#313)
• remove numpy conversion for cpu based data @jperez999 (#311)
• Fix PredictTensorflow args in session-based serving integration test @karlhigley (#310)
• Adjust setup_faiss to use METRIC_INNER_PRODUCT @karlhigley (#308)
• Remove the extra columns validation from WorkflowRunner @karlhigley (#303)
• Fix Tensorflow serving to make Merlin example tests pass again @karlhigley (#298)

🚀 Features

• update feast op to handle feast 0.31 - latest @jperez999 (#344)
• Support Ragged inputs and outputs in TransformWorkflow with Triton @oliverholworthy (#324)
• add AUTO kind for deployment of tritonserver and support no gpu environment @jperez999 (#321)
• Update SoftmaxSampling to support dynamic dtypes @oliverholworthy (#304)
• Add initial support and integration test for Merlin Models transformer models @karlhigley (#299)
• Fix Tensorflow serving to make Merlin example tests pass again @karlhigley (#298)

🔧 Maintenance

• minor syntax fixes in integration tests @nv-alaiacano (#351)
• add test requirements to postmerge gpu @nv-alaiacano (#347)
• Send slack notification on failed GHA jobs @nv-alaiacano (#346)
• Include tracebacks in errors returned from failing requests @karlhigley (#343)
• Add workflows to check base branch and set stable branch @oliverholworthy (#341)
• Add GitHub Workflow to publish conda package @oliverholworthy (#340)
• Update tag pattern in GitHub Workflows @oliverholworthy (#342)
• Update feast test dependency from 0.19.4 to 0.18.1 @oliverholworthy (#339)
• Raise TritonModelException if the Triton model has an error @oliverholworthy (#333)
• Refactor tensor representation adjustments to align with a schema @karlhigley (#332)
• working embedding in systems ensemble @jperez999 (#329)
• working padding op in systems ensemble @jperez999 (#330)
• Support serving in container without tensorflow, torch, or cudf @oliverholworthy (#326)
• use concurrency setting to kill already-running jobs when new commits come in @nv-alaiacano (#327)
• Condense the WorkflowRunner sub-classes @karlhigley (#319)
• add AUTO kind for deployment of tritonserver and support no gpu environment @jperez999 (#321)
• Check offsets instead of is_ragged for ragged representation @oliverholworthy (#318)
• Use ragged names in workflow triton request if column is ragged @oliverholworthy (#315)
• Remove the Triton-specific Workflow execution code @karlhigley (#309)
• Downgrade unused columns error to a warning @karlhigley (#306)
• Remove ensemble specific code @oliverholworthy (#305)
• Minor fixes to post-merge test actions @karlhigley (#302)
• Set up pre- and post-merge CI test runs @karlhigley (#301)
• add detection swap for cpu based to numpy @jperez999 (#300)
• Switch from __lengths to __offsets for ragged list features @karlhigley (#296)
• Update default value of cpu to None in dataset fixture @oliverholworthy (#294)
• update conftest for backwards compat and new to parquet api @jperez999 (#297)
• Adjust Systems to use TensorTable and schema shapes @karlhigley (#293)