TOSA reference model v0.50.0 release tag
- Note that in this release, the specification has added the concept of
levels, but the reference model and conformance tests do not support
testing of levels
- Added FP16 and BF16 data type support, renamed FLOAT to FP32
- Started support for TOSA Main Inference conformance. Still a work in
progress
- Added the IModelRunner interface to allow running the reference model
via an API
- Added simple testing against Numpy
- Tests created/fixed: prelu, CAST, RESCALE, TILE, SIGMOID
- Upgraded Eigen submodule to version 3.4.0
- Misc. bugfixes
« Previous 630c17c Add negative testing to cond_if, while_loop by Matthew Haddon · 2 years, 9 months ago bb5676f Add ERROR_IF checks to operators without specific ERROR_IFs by Matthew Haddon · 2 years, 9 months ago 25fbe52 Cleanup CMakeLists.txt by Kevin Cheng · 2 years, 8 months ago d4ea8f1 Update reference model version to 0.24.0 draft by Eric Kunze · 2 years, 8 months ago 5ffbc94 Remove draft tag for 0.23 release by Eric Kunze · 2 years, 8 months ago v0.23 v0.23.0 f3e016f more ERROR_IF fixes by Kevin Cheng · 2 years, 8 months ago 8079480 Fix for tensor_ops.cc by Kevin Cheng · 2 years, 8 months ago c0b24f0 RESIZE test generator updates by Jeremy Johnson · 2 years, 8 months ago fe392ce Changes for 0.23.0 release by Kevin Cheng · 2 years, 8 months ago 1009674 Add version to reference model by Kevin Cheng · 2 years, 8 months ago 8c06a65 Limit tensor values for COND_IF/WHILE_LOOP tests to stop saturation by Jeremy Johnson · 2 years, 8 months ago 6e52866 Fix typo in Concat ERROR_IF by Kevin Cheng · 2 years, 8 months ago 58026d6 Fix pool2d generation and check to match specification by Jeremy Johnson · 2 years, 8 months ago 5d00c69 Add ERROR_IF to control flow ops. by Kevin Cheng · 2 years, 8 months ago cc61be3 More ERROR_IF supports by Kevin Cheng · 2 years, 9 months ago e807aae Add Negative tests for pad, reshape, slice, transpose by Matthew Haddon · 2 years, 9 months ago c4cf037 Add negative testing support to fully_connected, matmul, argmax by Matthew Haddon · 2 years, 9 months ago c202521 Add negative testing support to RESCALE by Matthew Haddon · 2 years, 9 months ago 8a0a663 Remove tosa_serializer.py and use the one from serialization_lib now. by Kevin Cheng · 2 years, 9 months ago 7eb93d7 Catch ERROR_IF on AVG_POOL2D and MAX_POOL2D by Kevin Cheng · 2 years, 9 months ago 6040b4d COND_IF int8, int16 tests by Les Bell · 2 years, 9 months ago b6b59e3 Add negative testing for avg_pool2d, max_pool2d by Matthew Haddon · 2 years, 9 months ago d6ce725 Add negative testing support for reduce operators by Matthew Haddon · 2 years, 9 months ago 03bec73 Fix rank and dtype filtering for ops like conv3d & fully_connected by Jeremy Johnson · 2 years, 9 months ago e4ecdb2 Add negative testing support for ew_unary operators by Matthew Haddon · 2 years, 9 months ago eacff9a Add negative testing support to (most) EW Binary Ops by Matthew Haddon · 2 years, 9 months ago 693ba9e Add ERROR_IF checks for mismatched batch/channel by Matthew Haddon · 2 years, 9 months ago 1c00b71 Separate positive and negative test generation and refactor by Matthew Haddon · 2 years, 9 months ago 960985a Fix shape generation with num-const-inputs-concat option by Jeremy Johnson · 2 years, 9 months ago ec5586c Fix reduction ERROR_IF cases by Kevin Cheng · 2 years, 9 months ago 478101b Couple of reference model fixes by Kevin Cheng · 2 years, 9 months ago 848efb4 Add shared ERROR_IF statements by Matthew Haddon · 2 years, 10 months ago 17e9202 Fix CONST test. by Kevin Cheng · 2 years, 9 months ago c72b59c Fixes to pass NEGATE op test. by Kevin Cheng · 2 years, 9 months ago 903763c Add SUBGRAPH_ERROR_IF() to catch graph-level error. by Kevin Cheng · 2 years, 9 months ago 7fb8fa1 Remove duplicate output tensor in cond_if_binary test by Kevin Cheng · 2 years, 9 months ago 269e898 Remove ReluN op. by Kevin Cheng · 2 years, 9 months ago c42addc Removing rank 0 broadcast in binary op. by Kevin Cheng · 2 years, 9 months ago 6097c3d Op that violates rank requirement now runs to the end and return ERROR instead of bailing out. by Kevin Cheng · 2 years, 9 months ago e86fd34 Add ERROR_IF support for RESIZE by Matthew Haddon · 2 years, 10 months ago 7aa69f4 Extended tests for 2D/3D tensor functions by Les Bell · 2 years, 9 months ago 6132cfe Add stride larger than dimension error_if check for resize by Matthew Haddon · 2 years, 9 months ago 7e7fccd Unit test runner ergonomics by Jared Smolens · 2 years, 9 months ago 1918e8a Add maximum dimension check ERROR_IF in RESIZE. by Kevin Cheng · 2 years, 10 months ago c94e63d Compatibility with flatbuffers 1.12.0 and 2.0.0 by Jared Smolens · 2 years, 9 months ago f414b3c Generate more CONV3D tests by Les Bell · 2 years, 10 months ago 1533b85 Implement Conv3D kernel. by Kevin Cheng · 2 years, 10 months ago 93a1628 Rename attribute: Pool2d, Conv2d, TransposeConv2d -> Pool, Conv, TransposeConv by Kevin Cheng · 2 years, 10 months ago e3d6a8f Add a top level license file by Eric Kunze · 2 years, 10 months ago ef509a4 Fix for ADD/SUB saturation in tests by Jeremy Johnson · 2 years, 10 months ago b227ae5 Fix while_loop test. by Kevin Cheng · 2 years, 10 months ago 9034747 Add saturation REQUIREs for ADD,SUB,MUL,INTDIV by Jeremy Johnson · 2 years, 10 months ago b724efc Remove invalid tests from test generator by Matthew Haddon · 2 years, 10 months ago 7456709 Allow user to specify test type generated by Matthew Haddon · 3 years ago 4b2881a Fix batch size if target shape set for SCATTER operator by Matthew Haddon · 2 years, 10 months ago 459443c Rename DIV operator to INTDIV by Matthew Haddon · 2 years, 10 months ago 82ad4d6 Set TILE multiple to one if input dimension is large by Matthew Haddon · 2 years, 10 months ago a9d5395 Restore expectedFailure to test description files by Jared Smolens · 2 years, 10 months ago acb550f Replace node level check ASSERT_MSG_NODE()/FATAL_ERROR_NODE() with REQUIRE() or ERROR_IF() by Kevin Cheng · 3 years ago 68e7aee Make MATMUL output shape random by Matthew Haddon · 2 years, 11 months ago 818ab90 Produce Concat tests with multiple input tensors by Matthew Haddon · 3 years ago a901740 Support int4 weights read. Added conv2d int8xint4 in test generation. by Kevin Cheng · 2 years, 11 months ago 7ffccce test gen PAD agPad fix to cover all pad values by Les Bell · 2 years, 11 months ago cac4ee9 Add support for UINT8 by Matthew Haddon · 3 years ago 2a29dc6 fully_connected test OC independent of input shape by Les Bell · 2 years, 11 months ago 33d837e fix output type for resize in test gen by Les Bell · 2 years, 11 months ago 30e4680 fix quantization zero-point generation by Les Bell · 3 years ago 18e2666 Fix random INT4 and INT8 dtype ranges and clamp attrs by Jeremy Johnson · 3 years ago a44ac5e Fix bug causing test generator to create unused input tensor by Matthew Haddon · 3 years ago f54d8a2 Add INT8 table operator support to test generator. by Jeremy Johnson · 3 years ago 16aac57 Update serialization_lib hash. by Kevin Cheng · 3 years ago 43e3719 Make operator tests follow consistent naming scheme by Matthew Haddon · 3 years ago 97eb75f Allow selection of higher rank tests using --target-rank by Jeremy Johnson · 3 years ago 5fc4e68 Fix bug causing reshape rank not to match test description. by Matthew Haddon · 3 years ago 2ad047d Fix bug causing identical reshape permutations by Matthew Haddon · 3 years ago d593414 Adding option -Coutput_dir= to control output location. by Kevin Cheng · 3 years ago a618557 Fix transpose test gen of permutations & rank by Jeremy Johnson · 3 years ago 82507d7 Update to use new serialization_lib API. by Kevin Cheng · 3 years ago 2d60f00 adding batch dimension to MatMul. by Kevin Cheng · 3 years, 1 month ago cd79f0e Rewrite model frontend to be json-driven. by Kevin Cheng · 3 years, 1 month ago 571f718 Support 8-bit TABLE op. by Kevin Cheng · 3 years, 1 month ago 47315e1 Fix typo in DIV unit test generator by Kevin Cheng · 3 years, 1 month ago 14d7f7a Update to v0.22.0 by Kevin Cheng · 3 years, 2 months ago d267dd9 Regenerate examples/ tests by Kevin Cheng · 3 years, 2 months ago 9257fd5 Fix OpMul for 32-bit x 32-bit inputs by Kevin Cheng · 3 years, 2 months ago 0f87c95 Support 16-bit Rescale by Kevin Cheng · 3 years, 3 months ago ad15dfa Concat takes variadic inputs by Kevin Cheng · 3 years, 4 months ago 573ecd4 Re-order TOSA unit tests to match the spec ordering by Jared Smolens · 3 years, 4 months ago 989cb05 Support mixed-precision input tensors for TOSA unit test. by Kevin Cheng · 3 years, 2 months ago 550ccc5 Replace serialization/ and verif/ with MLPlatform's serialization_lib submodule by Kevin Cheng · 3 years, 4 months ago cf6224e Fix bug that NegateOp doesn't clip output value by Kevin Cheng · 3 years, 2 months ago 2a76ad2 Update DTypes for TOSA ops, test rig fixes by Jared Smolens · 3 years, 4 months ago df86269 Fix operator initialization bug when rank/dtype fails to initialize with Usage arrays by Kevin Cheng · 3 years, 4 months ago 3a47857 Remove AINT8 by Kevin Cheng · 3 years, 5 months ago e2e13cd TOSA Unit Test updates by Jared Smolens · 3 years, 5 months ago 4025315 Numpy utility supports scalar (rank 0) .npy by Kevin Cheng · 3 years, 6 months ago 77d0f76 Update reference model/serialization library to 0.21.0 with unit tests added/updated by Kevin Cheng · 3 years, 7 months ago aee1fac Implement and add unit tests for MUL and ARITHMETIC_RIGHT_SHIFT by Kevin Cheng · 3 years, 8 months ago 99bea14 Update apply_scale_32() by Kevin Cheng · 3 years, 8 months ago e5e2676 Initial checkin of TOSA reference_model and tests by Eric Kunze · 3 years, 9 months ago Next »