forked from fastmachinelearning/qonnx
-
Notifications
You must be signed in to change notification settings - Fork 0
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Merge branch 'fastmachinelearning:main' into dev
- Loading branch information
Showing
34 changed files
with
2,152 additions
and
205 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,151 @@ | ||
| # Copyright (c) 2023 Advanced Micro Devices, Inc. | ||
| # All rights reserved. | ||
| # | ||
| # Redistribution and use in source and binary forms, with or without | ||
| # modification, are permitted provided that the following conditions are met: | ||
| # | ||
| # * Redistributions of source code must retain the above copyright notice, this | ||
| # list of conditions and the following disclaimer. | ||
| # | ||
| # * Redistributions in binary form must reproduce the above copyright notice, | ||
| # this list of conditions and the following disclaimer in the documentation | ||
| # and/or other materials provided with the distribution. | ||
| # | ||
| # * Neither the name of Xilinx nor the names of its | ||
| # contributors may be used to endorse or promote products derived from | ||
| # this software without specific prior written permission. | ||
| # | ||
| # THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" | ||
| # AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE | ||
| # IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE | ||
| # DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE | ||
| # FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL | ||
| # DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR | ||
| # SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER | ||
| # CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, | ||
| # OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE | ||
| # OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. | ||
|
|
||
| import clize | ||
| import numpy as np | ||
| import os | ||
| from matplotlib import pyplot as plt | ||
| from tqdm import tqdm | ||
|
|
||
| from qonnx.core.modelwrapper import ModelWrapper | ||
| from qonnx.util.range_analysis import ( | ||
| REPORT_MODE_RANGE, | ||
| REPORT_MODE_STUCKCHANNEL, | ||
| REPORT_MODE_ZEROSTUCKCHANNEL, | ||
| report_mode_options, | ||
| report_modes, | ||
| ) | ||
|
|
||
|
|
||
| def update_tensor_stats(tensor, axes, ret_dict={}): | ||
| shp = tensor.shape | ||
| if ret_dict == {}: | ||
| ret_dict["shape"] = shp | ||
| else: | ||
| assert ret_dict["shape"] == shp | ||
| for axis in axes: | ||
| tensor_new = np.moveaxis(tensor, axis, 0).reshape(shp[axis], -1) | ||
| ret_axis = { | ||
| "min": np.min(tensor_new, axis=1), | ||
| "max": np.max(tensor_new, axis=1), | ||
| } | ||
| axis_name = "axis%d" % axis | ||
| if axis_name in ret_dict: | ||
| ret_dict[axis_name] = ret_axis | ||
| ret_axis["min"] = np.minimum(ret_axis["min"], ret_dict[axis_name]["min"]) | ||
| ret_axis["max"] = np.maximum(ret_axis["max"], ret_dict[axis_name]["max"]) | ||
| ret_dict[axis_name] = ret_axis | ||
| return ret_dict | ||
|
|
||
|
|
||
| def tensor_stats( | ||
| modelwrapper_or_filename, | ||
| act_dump_dir: str, | ||
| output_stats_dir: str, | ||
| *, | ||
| axes="1", | ||
| plot=True, | ||
| report_mode: report_mode_options = REPORT_MODE_RANGE | ||
| ): | ||
| assert report_mode in report_modes, "Unknown report_mode" | ||
| if not isinstance(modelwrapper_or_filename, ModelWrapper): | ||
| model = ModelWrapper(modelwrapper_or_filename) | ||
| else: | ||
| model = modelwrapper_or_filename | ||
| if not isinstance(axes, list): | ||
| axes = [int(x.strip()) for x in axes.split(",")] | ||
| if not os.path.isdir(output_stats_dir): | ||
| os.makedirs(output_stats_dir) | ||
|
|
||
| all_tensor_dump_files = [] | ||
| all_tensor_dump_files = [f for f in os.listdir(act_dump_dir) if os.path.isfile(os.path.join(act_dump_dir, f))] | ||
| all_tensor_dump_files = [f for f in all_tensor_dump_files if f.endswith(".npy")] | ||
| tensorwise_stats = {} | ||
| stuck_chans = {} | ||
| for outp in tqdm(model.graph.output, "Tensors"): | ||
| tname = outp.name | ||
| t_files = [f for f in all_tensor_dump_files if f.startswith(tname)] | ||
| tensorwise_stats[tname] = {} | ||
| for f in tqdm(t_files, "Batches"): | ||
| t_file = np.load(os.path.join(act_dump_dir, f)) | ||
| tensorwise_stats[tname] = update_tensor_stats(t_file, axes=axes, ret_dict=tensorwise_stats[tname]) | ||
| tstats = tensorwise_stats[tname] | ||
| for axis in axes: | ||
| axis_name = "axis%d" % axis | ||
| axis_data = tstats[axis_name] | ||
| axis_min = axis_data["min"] | ||
| axis_max = axis_data["max"] | ||
| tensor_stuck_chans = np.nonzero(axis_min == axis_max)[0] | ||
| if report_mode in [REPORT_MODE_STUCKCHANNEL, REPORT_MODE_ZEROSTUCKCHANNEL]: | ||
| if len(tensor_stuck_chans) > 0: | ||
| list_stuck_chans = list(tensor_stuck_chans) | ||
| list_stuck_values = list(axis_min[tensor_stuck_chans]) | ||
| stuck_chans[tname] = list(zip(list_stuck_chans, list_stuck_values)) | ||
| if plot: | ||
| for axis in axes: | ||
| axis_name = "axis%d" % axis | ||
| data = tensorwise_stats[tname][axis_name] | ||
| axis_min = data["min"] | ||
| axis_max = data["max"] | ||
| axis_range = axis_max - axis_min | ||
| chans = [i for i in range(len(axis_min))] | ||
| plt.clf() | ||
| plt.figure(constrained_layout=True, figsize=(5, len(axis_min) / 3)) | ||
| bars = plt.barh(chans, axis_range, left=axis_min) | ||
| bar_labels = [str((axis_min[i], axis_max[i])) for i in range(len(axis_min))] | ||
| plt.bar_label(bars, bar_labels) | ||
| plt.yticks([x for x in range(len(axis_range))]) | ||
| plt.xlabel("Channel number") | ||
| plt.ylabel("Channel range") | ||
| plt.title("Observed range for %s_%s" % (tname, axis_name)) | ||
|
|
||
| plt.savefig(output_stats_dir + "/%s_%s.png" % (tname, axis_name)) | ||
|
|
||
| if report_mode == REPORT_MODE_RANGE: | ||
| ret = tensorwise_stats | ||
| elif report_mode == REPORT_MODE_STUCKCHANNEL: | ||
| ret = stuck_chans | ||
| elif report_mode == REPORT_MODE_ZEROSTUCKCHANNEL: | ||
| # only leave channels that are stuck at zero | ||
| # value info removed since implicitly 0 | ||
| ret = stuck_chans | ||
| new_ret = {} | ||
| for tname, schans in ret.items(): | ||
| schans_only_zero = set([x[0] for x in schans if x[1] == 0]) | ||
| if len(schans_only_zero) > 0: | ||
| new_ret[tname] = schans_only_zero | ||
| ret = new_ret | ||
| return ret | ||
|
|
||
|
|
||
| def main(): | ||
| clize.run(tensor_stats) | ||
|
|
||
|
|
||
| if __name__ == "__main__": | ||
| main() |
File renamed without changes.
Binary file not shown.
Binary file not shown.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,59 @@ | ||
| # Copyright (c) 2023 Advanced Micro Devices, Inc. | ||
| # All rights reserved. | ||
| # | ||
| # Redistribution and use in source and binary forms, with or without | ||
| # modification, are permitted provided that the following conditions are met: | ||
| # | ||
| # * Redistributions of source code must retain the above copyright notice, this | ||
| # list of conditions and the following disclaimer. | ||
| # | ||
| # * Redistributions in binary form must reproduce the above copyright notice, | ||
| # this list of conditions and the following disclaimer in the documentation | ||
| # and/or other materials provided with the distribution. | ||
| # | ||
| # * Neither the name of Xilinx nor the names of its | ||
| # contributors may be used to endorse or promote products derived from | ||
| # this software without specific prior written permission. | ||
| # | ||
| # THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" | ||
| # AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE | ||
| # IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE | ||
| # DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE | ||
| # FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL | ||
| # DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR | ||
| # SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER | ||
| # CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, | ||
| # OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE | ||
| # OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. | ||
|
|
||
| from qonnx.core.modelwrapper import ModelWrapper | ||
| from qonnx.transformation.base import Transformation | ||
|
|
||
|
|
||
| class ChangeBatchSize(Transformation): | ||
| """Change the batch size dimension to the given value for the entire graph | ||
| by changing it for the global input/output and removing all intermediate | ||
| shapes (will need a call to shape inference to restore shapes). | ||
| Will attempt to handle any Reshape nodes with constant shape parameters by | ||
| changing the batch size dimension value in the parameter.""" | ||
|
|
||
| def __init__(self, bsize): | ||
| super().__init__() | ||
| self.bsize = int(bsize) | ||
|
|
||
| def apply(self, model: ModelWrapper): | ||
| onnx_model = model.model | ||
| bsize = self.bsize | ||
| onnx_model.graph.input[0].type.tensor_type.shape.dim[0].dim_value = bsize | ||
| onnx_model.graph.output[0].type.tensor_type.shape.dim[0].dim_value = bsize | ||
| while len(onnx_model.graph.value_info) > 0: | ||
| onnx_model.graph.value_info.remove(onnx_model.graph.value_info[0]) | ||
| reshape_nodes = model.get_nodes_by_op_type("Reshape") | ||
| for reshape_node in reshape_nodes: | ||
| rs_param_name = reshape_node.input[1] | ||
| rs_param = model.get_initializer(rs_param_name) | ||
| if rs_param is not None: | ||
| rs_param = rs_param.copy() | ||
| rs_param[0] = bsize | ||
| model.set_initializer(rs_param_name, rs_param) | ||
| return (model, False) |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,66 @@ | ||
| # Copyright (c) 2023, Advanced Micro Devices, Inc. | ||
| # All rights reserved. | ||
| # | ||
| # Redistribution and use in source and binary forms, with or without | ||
| # modification, are permitted provided that the following conditions are met: | ||
| # | ||
| # * Redistributions of source code must retain the above copyright notice, this | ||
| # list of conditions and the following disclaimer. | ||
| # | ||
| # * Redistributions in binary form must reproduce the above copyright notice, | ||
| # this list of conditions and the following disclaimer in the documentation | ||
| # and/or other materials provided with the distribution. | ||
| # | ||
| # * Neither the name of QONNX nor the names of its | ||
| # contributors may be used to endorse or promote products derived from | ||
| # this software without specific prior written permission. | ||
| # | ||
| # THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" | ||
| # AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE | ||
| # IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE | ||
| # DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE | ||
| # FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL | ||
| # DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR | ||
| # SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER | ||
| # CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, | ||
| # OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE | ||
| # OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. | ||
|
|
||
| from qonnx.core.modelwrapper import ModelWrapper | ||
| from qonnx.transformation.base import Transformation | ||
|
|
||
|
|
||
| class ExposeIntermediateTensorsLambda(Transformation): | ||
| def __init__(self, tensor_filter=lambda tname, model: True): | ||
| super().__init__() | ||
| self.tensor_filter = tensor_filter | ||
|
|
||
| def apply(self, model: ModelWrapper): | ||
| all_tensor_names = model.get_all_tensor_names() | ||
| for tname in all_tensor_names: | ||
| if self.tensor_filter(tname, model): | ||
| # check whether this tensor is already in the outputs | ||
| if tname in [x.name for x in model.graph.output]: | ||
| # already part of outputs, skip | ||
| continue | ||
| else: | ||
| # append ValueInfo to outputs | ||
| tensor_vi = model.get_tensor_valueinfo(tname) | ||
| model.graph.output.append(tensor_vi) | ||
| # remove existing ValueInfo to avoid duplicate | ||
| model.graph.value_info.remove(tensor_vi) | ||
|
|
||
| return (model, False) | ||
|
|
||
|
|
||
| class ExposeIntermediateTensorsPatternList(ExposeIntermediateTensorsLambda): | ||
| def pattern_filter(self, tname, model): | ||
| if self.dynamic_only: | ||
| return any([(pat in tname) and (model.get_initializer(tname) is None) for pat in self.pattern_list]) | ||
| else: | ||
| return any([(pat in tname) for pat in self.pattern_list]) | ||
|
|
||
| def __init__(self, pattern_list, dynamic_only=True): | ||
| self.pattern_list = pattern_list | ||
| self.dynamic_only = dynamic_only | ||
| super().__init__(tensor_filter=self.pattern_filter) |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Oops, something went wrong.