Change Logs¶

0.7.6¶

#193: validates with 4.53.3
#189: support for task mask-generation
#192: add support for Gemma-3, add serialization for HybridCache, changes to support transformers>=4.54

#186: add parameter –output_names to command line validate to change the output names of the onnx exported model
#185: remove the use of _seen_tokens in DynamicCache (removed in transformers>4.53), updates dummpy inputs for feature-extraction
#184: implements side-by-side

#178: add a patch for eager_mask to handle assert len(flat_dynamic_shapes) == num_placeholders - num_lifted_inputs
#177: changes for the next version of onnx, fixes all_dynamic_shape_from_inputs

#159: supports for models with custom code in huggingface
#158: fix uses of pretrained version
#156, #157: add plots and other options to deal with the unpredictable
#155: better aggregation of historical data
#151, #153: adds command line agg, class CubeLogsPerformance to produce timeseries
#152: add a function to compute fully dynamic shapes given any inputs

#149: supports for StaticCache
#147: simplified log processing
#146: patch for IdeficsAttention, IdeficsEmbedding
#145: patch for _compute_dynamic_ntk_parameters (Phi3RotaryEmbedding)
#144: support for second inputs with different dimension, rename test_helper into validate, support interpolate_pos_encoding for VitModel, update model builder helpers for this PR Use ONNX IR for model builder
#143: compares intermediate results,

#128: patch for Phi3RotaryEmbedding
#126: add repeat and warmup to command line validate
#125: handles sequences in TorchOnnxEvaluator
#123: add subgraphs to TorchOnnxEvaluator
#122: add local functions to TorchOnnxEvaluator
#120: enables TorchOnnxEvaluator in command line python -m onnx_diagnostic validate ...
#115, #116, #117, #118, #119, #127: first steps for TorchOnnxEvaluator
#114: extends the list of known rewritings
#113: fixes a couple of issues with ModelBuilder

#111: support ModelBuilder with command line validate
#108, #109, #110: first version of an algorithm rendering small onnx graph in ascii, patch for torch.vmap

#105: more options to tune control flow rewriting
#104: add summarization task, add rewrite to command line validate
#101: first draft to rewrite loops
#100: implements a context to automatically rewrite methods or function with control flows
#96: implements is_stealing, steal_append to complement steal_forward
#95: fixzq Scan implementation for OnnxruntimeEvaluator
#93: introduces patched expressions to get around annoying export issues
#92: supports errors distribution in max_diff
#91: enables strings in guess_dynamic_shapes
#88, #89: extends steal_forward to dump input, outputs in onnx models
#83, #85: improves the automated rewriting of control flow (test)

#75: renames bypass_export_some_patches into torch_export_patches, keeps the old name
#74: increases the list of class/architectures

#65: support SlidingWindowCache
#63: support option --trained
#61: improves dynamic shapes for EncoderDecoderCache
#58: add function use_dyn_not_str to replace string by torch.export.Dim.DYNAMIC, use string instead of torch.export.Dim.DYNAMIC when returning the dynamic shapes for a specific models, it is a valid definition for torch.onnx.export which can reuse the names
#55: add support for text-classification
#54: add support for fill-mask, refactoring
#52: add support for zero-shot-image-classification
#50: add support for onnxruntime fusion
#48: add support for EncoderDecoderCache, test with openai/whisper-tiny
#45: improve change_dynamic_dimension to fix some dimensions

#43: uses custom patches
#38: uses the registered serialization functions when it is available
#30, #31: adds command to test a model id, validate the export
#29: adds helpers to measure the memory peak and run benchmark on different processes
#28: adds command line to print out the configuration for a model id, support image-text-to-text
#26: creates a folder helpers to gather all the functions used in many places
#25: improve patches for DynamicCache (issue with register_pytree_flatten_spec being deprecated)
#24: dummy inputs for text2text-generation, add new function convert_dynamic_axes_into_dynamic_shapes to convert dynamic axes into dynamic shapes, add support for T5ForConditionalGeneration
#23: dummy inputs for image-classification
#22, #27: api to create untrained model copying the architecture of the trained models and dummy inputs for them, support for text-generation

#16: refactors patches, add model Phi2, implements a tweak to raise an exception with a dynamic dimension becomes static when exporting a model

first version