Toolchain Docker History

Manual History Version

Toolchain Change log

[v0.32.0]
- Add Einsum defusion in kneronnxopt.
- Support Cast to int64 in knerex and compiler.
- Support HardSwish, Topk and Split nodes in knerex and compiler.
- Update the regression flow log printing. Print success log seperately from errors to avoid confusing.
- Update IP evaluator for DMA with small length.
- Fix the kneronnxopt bug in replace_Gather_with_Slice.
- Fix the knerex bug: node Concat channel mismatch.
- Fix the dynasty float bug in InstanceNorm pad edge mode.
- Fix knerex/compiler bug in CPU node settings for Resize node.
- Verified opset18 operator validity for knerex and compiler.
- Reduce memory usage (especially for large models) for compiler.
[v0.31.1]
- Add const_in_bitwidth_mode option for quantization. The default is int16. Unless the customer particularly desires to increase the speed, it can be changed to int8
- Update analyzer exception log.
- Update kneronnxopt to set expanding dilated Conv to False by default.
- Update kneronnxopt to diable fusing BatchNormalization into Conv by default.
- Update compiler for the deep search memory estimation algorithm.
- Update compiler to extend the timeout for deep search.
- Update compiler to change expt/log/softmax to 16b.
- Fix the ktc bug in some default output path names.
- Fix the kneronnxopt bug in duplicating shared weights.
- Fix the compiler bug in broadcasting.
- Fix the compiler bug that Concat channel axis not supported.
[v0.31.0]
- Introduce quan_config for ModelConfig.analysis for more detailed quantization configuration.
- Add ktc.opt_and_eval command for quick onnx optimization and evaluation.
- Remove deprecated compilerIpevaluator_730.sh and add warning messages to other depecated scritps.
- Add compiler_tiling option for IP evaluator.
- Add --clear-shapes and --replace-avgpool-with-conv flags to kneronnxopt.
- Add --seperate flag to kneronnxopt.onnx_vs_onnx for detailed output comparison.
- Update knerex shared weight combination logic.
- Update knerex and dynasty to support empty Constant nodes.
- Update compiler for better message logging.
- Update dynasty and compiler for softmax support.
- Update regression for longer timeout setting.
- Improve model_fx_report.html readability.
- Speed up compilerfor large model.
- Fix the ktc error message for not supported special characters in model path.
- Fix the ktc bug that logging module not imported.
- Fix the kneronnxopt bug that flip nodes are eliminated incorrectly.
- Fix the kneronnxopt bug that replacing Add/Sub/Mul/Div with BatchNormalization node incorrectly.
- Fix the nef utility bug that 520 nef combination generates invalid nef files.
- Other bug fixes and performance improvements.
[v0.30.0]
- Introduce input_fmt for ModelConfig to specify the input format of the model.
- bie files may not be compatible with previous versions.
- Fix kneronnxopt to duplicate shared weights for not supported cases.
- Update knerex to support alpha&beta hardsigmoid.
- Update webgui to support conda environment selection.
- Bug fixes and performance improvements.
[v0.29.0]
- Introduce mixbw for fixed-point analysis, an automated quantization mode that optimizes 8/16-bit configurations for Conv nodes, balancing accuracy (SNR) and speed (FPS).
- Add onnx_vs_onnx command line entrance for kneronnxopt to compare two onnx models.
- Optimize log printing for ktc.
- Optimize compiler runtime based on partial graph comparison.
- Fix the bug that knerex could not handle last nodes properly in some cases.
- Fix the bug that knerex could not handle Add constant node input properly.
- Fix other known bugs.
[v0.28.2]
- Fix the batch compiler bug that nef files do not contain version information.
- Optimize kneronnxopt for processing large models.
- Fix other bugs.
[v0.28.1]
- Change default miniconda channel due to the license issue.
[v0.28.0]
- Change conda environment due to license issue.
- Remove caffe support.
- Add --opt-matmul flag to kneronnxopt for kneron hardware matmul optimization.
- Add --overwrite-input-shapes and --skip-fuse-qkv flags to kneronnxopt large model processing.
- Support GRU, LSTM, and RNN operators defusion in kneronnxopt.
- Fix bugs.
[v0.27.0]
- Adjust batch compiler internal behavior to improve robustness.
- Optimize compiler to improve feature map cut search speed.
- Optimize data converter to improve speed.
- Fix bugs.
[v0.26.0]
- Optimize compiler for 730 graph cutting.
- Supports the flash attention model.
- Add producer name in kneronnxopt.
- Fix bugs.
[v0.25.1]
- ktc supports non-str platform conversion.
- Fix kneronnxopt argument name.
- Update conda environment
  - base:
    - numpy-1.21.0
    - pandas-1.2.0
  - onnx1.13:
    - numpy-1.26.4
    - pandas-2.2.2
- Fix bugs.
[v0.25.0]
- IP evaluator add arguments weight_bandwidth and dma_bandwidth.
- 730 toolchain full upgrade.
- Optimize batch compiler efficiency.
- Optimize batch compiler memory management algorithm.
- Support fmap over 4D.
- Support more operators (Abs, Log, Pow, Sign).
- Support mix accuracy mode.
- Update environment packages.
- Change default environment to onnx1.13.
- Fix bugs.
[v0.24.0]
- ktc.ModelConfig.analysis rename 'fm_cut' to 'compiler_tiling'.
- ktc.ModelConfig.analysis change datapath_bitwidth_mode and weight_bitwidth_mode to support mix-bitwidth mode "mix light" and "mix balance". (exclude 520).
- Support ONNX opset 18.
- Optimize customized node support.
- Support multiple output operators.
- Support per-channel quantization
- WebGUI Auto-detect model input node information, make convert onnx or tflite to nef more easier.
- Docker update the parallel package version.
- Fix known bugs and improve performance.
[v0.23.1]
- Compiler may raise error on unsupported CPU nodes.
- Bug fixes for compiler, dynasty and ktc.
[v0.23.0]
- ktc.kneron_inference now requires input_names which is optional previously.
- ktc.ModelConfig.evaluate() now only generate html report instead of the previous txt report.
- bie format update for better supporting new models. Due to this update, the previous generated bie files are no longer supported.
- Batch compiler now support 730 hardware.
- Provide more information in IP evaluator report.
- Optimize the quantization snr calculation.
- Optimize the compiler for better on-chip performance.
- Supporting latest models with loop node inside.
- Bug fixes and other improvements.
[v0.22.0]
- ktc.ModelConfig.evaluate
- Change function parameter output_folder to output_dir.
- Change default output location to /data1/kneronflow.
- ktc.ModelConfig.analysis
- Change function parameter output_bie to output_dir, which should be a folder path instead of the file path before. Defaults to /data1/kneronflow.
- Remove deprecated parameters bitwidth_mode, outlier and skip_verify.
- Add parameters model_in_bitwidth_mode, model_out_bitwidth_mode, datapath_bitwidth_mode and weight_bitwidth_mode.
- ktc.ModelConfig
- Add radix_json_path to ktc.ModelConfig.__init__ which provides ktc.compile ability to compile with onnx and a special json file. Defaults to None.
- Add debug to ktc.ModelConfig.__init__ to avoid removing debug output files. Defaults to False.
- ktc.compile, ktc.encrpyt_compile
- Change default output location to /data1/kneronflow.
- Add debug parameter to avoid removing debug output files. Defaults to False.
- Provided a function ktc.convert_channel_last_to_first to help converting channel last inputs to channel first(image only).
- Remove unessential output files. Some functions add debug parameters to avoid removing debug output files. ktc.kneron_inference can use dump parameter avoid removing debug output files.
- Bug fixes and performance improvements.
[v0.21.0]
- Change input format from channel last to the same input shape of ONNX.
- Change compiler generated ioinfo.csv into ioinfo.json for platforms other than 520.
- Remove deprecated command line scripts, e.g. fpAnalyserCompilerIpevaluator_520.py, fpAnalyserBatchCompile_520.py.
- Add html report for analysis API.
- Add helper utilities under compiler folder.
- Add parallel and w3m into docker environment.
- Bug fixes.
- Documentation updates.
[v0.20.2]
- Fix combining nef error.
[v0.20.1]
- Update toolchain example to MobileNet v2.
- Fix knerex bias adjustment.
- Fix knerex shared weight with same name support.
- Fix other bugs.
[v0.20.0]
- Support text procssing models.
- Set flatbuffer as the default 720 compiling mode.
- Refactor compiler and analyser inner structure.
- Due to the structure change, batch compiler do not backwark support previous bie files.
- Refactor toolchain manual.
- Bug fixes.
[v0.19.0]
- Add 730 hardware support.
- Minor bug fixes.
[v0.18.2]
- ktc: Add mode, optimize, export_dynasty_dump argument to analysis.
- ktc: Set skip_verify in analysis as deprecated.
- regression: Add optimize option for optimization level selection.
- regression: Fix interface to asure platform is integer.
- converter: Add 720 batch process with --opt-720 flag.
- converter: Add enable shared weight duplication flag -d. By default, shared weights are no longer duplicated.
- converter: Remove -s flag since it is now the default behavious.
- converter: Optimize debug output.
- compiler: Fix RDMA not correctly executed.
- E2E simulator: Change dynasty library fetching method.
- Minor bug fixes.
[v0.18.1]
- docker: Update numpy from 1.18.5 to 1.21.
- ktc: Add km_cut argument to analysis.
- converter: Add operator checking before optimizaiton.
- E2E simulator: Change dynasty library fetching method.
- Minor bug fixes.
[v0.18.0]
- ONNX is updated to 1.7.0.
- Introduce WebGUI.
- Adjust 720 and 530 IP Evaluator default hardware specification.
- Add more analysis options.
[v0.17.0]
- Optimize analysis API. Now we verify the model while analysing the fixed point performance.
- E2E simulator no longer requires radix parameter.
[v0.16.0]
- Introduce 530 toolchain.
- Optimizer supports pixel modification for wider input range adjustment.
- FP analysis supports mmse mode.
- Compiler supports hardware_cut_opt option.
- Compiler supports combining nef files.
[v0.15.0]
- Document now is written for Python API. The original script document can be found in Command Line Script Tools.
- Compiler now support weight compress option.
[v0.14.0]
- ONNX is updated to 1.6.0
- Pytorch is updated to 1.7.1
- Introduce toolchain Python API.
[v0.13.0]
- 520 toolchain and 720 toolchain now is combined into one. But the scripts names and paths are the same as before. You don't need to learn it again.
- E2E simulator has been updated to a new version. Usage changed. Please check its document.
[v0.12.0] Introduce convert_model.py which simplify the conversion process.
[v0.11.0] Batch compile now generate .nef files to simplify the output.
[v0.10.0]
- input_params.json and batch_input_params.json have been simplified a lot. Please check the document for details.
- simulator.sh, emulator.sh and draw yolo image scripts are no longer available from /workspace/scripts. They have been moved to E2E simulator. Please check its document.
[v0.9.0] In the example, the mount folder /docker_mount is separated from the interactive folder /data1 to avoid unexpected file changes. Users need to copy data between the mount folder and the interactive folder. Of course you can still mount on /data1. But please be careful that the results folder under /data1 may be overwritten.