-
Notifications
You must be signed in to change notification settings - Fork 243
/
calibration.py
523 lines (474 loc) · 25.9 KB
/
calibration.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
#!/usr/bin/env python
# coding: utf-8
#
# Copyright (c) 2021 Intel Corporation
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
# -------------------------------------------------------------------------
# Copyright (c) Microsoft, Intel Corporation. All rights reserved.
# Licensed under the MIT License. See License.txt in the project root for
# license information.
# --------------------------------------------------------------------------
import copy
import logging
import sys
import numpy as np
import onnx
import onnxruntime
import onnx.numpy_helper as numpy_helper
from onnx import helper, TensorProto, shape_inference
from packaging.version import Version
from neural_compressor.model.onnx_model import ONNXModel
from neural_compressor.adaptor.ox_utils.util import make_dquant_node, is_B_transposed
logger = logging.getLogger()
ONNX18_VERSION = Version("1.8.0")
ORT112_VERSION = Version("1.12.0")
class ONNXRTAugment:
'''augment input model to dump tensor or for calibration'''
def __init__(self, model_wrapper,
dataloader,
dump_op_types,
black_nodes=[],
white_nodes=[],
iterations=[]):
'''
:param model: ONNX model to calibrate
:param dataloader: user implemented object to read in and preprocess calibration dataset
:param op_types: operator types to be calibrated and quantized, default = 'Conv,MatMul'
:param black_nodes: operator names that should not be quantized, default = ''
:param white_nodes: operator names that force to be quantized, default = ''
:param iterations: tensor of which iteration will be collected.
'''
self.model_wrapper = model_wrapper
self.model = model_wrapper.model
ai_onnx_domain = [opset for opset in self.model.opset_import \
if not opset.domain or opset.domain == "ai.onnx"]
self.opset_version = ai_onnx_domain[0].version
self.dataloader = dataloader
self.dump_op_types = dump_op_types
self.black_nodes = black_nodes
self.white_nodes = white_nodes
self.augmented_model = None
self.iterations = iterations
self.augment_nodes = []
self.dequantized_output = {}
self.already_quantized = 'DequantizeLinear' in \
[node.op_type for node in self.model.graph.node]
self.dynamically_quantized = False
self.ort_version = Version(onnxruntime.__version__)
def augment_graph(self, activation_only=False, weight_only=False):
'''
Adds nodes to all quantization_candidates op type nodes in
model and ensures their outputs are stored as part of the graph output
:param activation_only(bool): whether to dump activation tensor only
:param weight_only(bool): whether to dump weight_only
:return: augmented ONNX model
'''
self.dequantized_output.clear()
onnx_version = Version(onnx.__version__)
if onnx_version < ONNX18_VERSION:
logger.warning("Static quantization for NLP model is supported " \
"at onnx 1.8.0 and newer.")
if self.already_quantized and any([i.dims in [1, 2] for i in \
self.model_wrapper.initializer() if i.name.endswith('_scale')]):
if self.opset_version < 13 and self.ort_version >= ORT112_VERSION:
logger.warning("Please use onnxruntime < 1.12.0 or upgrade model opset " \
"version to 13 or higher to inspect per-channel quantized weight")
model = copy.deepcopy(self.model)
model_nodes_names = [node.name for node in model.graph.node]
added_nodes = []
added_outputs = []
tensors_to_dump = set()
for augment_node_type in self.augment_nodes:
if augment_node_type not in ['DequantizeLinear']: # pragma: no cover
raise ValueError("Unexpected augment_node {} only DequantizeLinear is " \
"supported".format(augment_node_type))
if self.already_quantized:
# mapping between fp32 node and int8 node
new_white_nodes = []
for white_node in self.white_nodes:
new_white_node = white_node + "_quant"
assert new_white_node in model_nodes_names, "no quantized {} in the " \
"graph".format(white_node)
new_white_nodes.append(new_white_node)
self.white_nodes = new_white_nodes
initializers = {i.name: i.data_type for i in model.graph.initializer}
node_outputs = []
for node in model.graph.node: # pylint: disable=no-member
node_outputs.extend(node.output)
should_be_dump = ((node.op_type in self.dump_op_types) and
(node.name not in self.black_nodes)) or \
(node.name in self.white_nodes)
if should_be_dump:
if not weight_only and not activation_only:
tensors_to_dump.update(node.input)
tensors_to_dump.update(node.output)
elif weight_only:
for input in node.input:
if self.already_quantized and \
input.replace('_dequantized', '_quantized') in initializers:
tensors_to_dump.add(input)
elif not self.already_quantized and input in initializers:
tensors_to_dump.add(input)
elif activation_only:
tensors_to_dump.update(node.output)
for tensor in tensors_to_dump:
if self.augment_nodes:
if tensor not in node_outputs and tensor not in initializers:
continue
for augment_node_type in self.augment_nodes:
if augment_node_type in ['DequantizeLinear']:
# insert DequantizeLinear node as output
if tensor.endswith('_scale') or tensor.endswith('_zero_point') or \
tensor.endswith('_QuantizeLinear') or \
tensor.endswith('_QuantizeInput_quantized'):
continue
if not self.dynamically_quantized:
tensor = tensor.replace('_QuantizeInput', '_quantized') if \
tensor.endswith('_QuantizeInput') else tensor
else:
tensor = tensor.replace('_output_quantized', '') if \
tensor.endswith('_output_quantized') else tensor
augment_node_name = tensor + "_new_" + augment_node_type
scale, zero_point = self.model_wrapper.get_scale_zero(tensor)
if scale:
# the tensor is in INT8 dtype
nodes, output = self._dequantize(tensor, scale, zero_point)
if output:
added_nodes.extend(nodes)
added_outputs.append(helper.make_tensor_value_info(
output, # pylint: disable=no-member
TensorProto.FLOAT, ())) # pylint: disable=no-member
else:
# the tensor is in FP32 dtype
if tensor not in [t.name for t in model.graph.output]:
added_tensor = helper.ValueInfoProto()
added_tensor.name = tensor
added_outputs.append(added_tensor)
else:
if tensor not in [t.name for t in model.graph.output]:
added_tensor = helper.ValueInfoProto()
added_tensor.name = tensor
added_outputs.append(added_tensor)
if self.augment_nodes:
model.graph.node.extend(added_nodes) # pylint: disable=no-member
model.graph.output.extend(added_outputs) # pylint: disable=no-member
self.augmented_model = model
def get_intermediate_outputs(self, calib_mode=None):
'''
Gather intermediate model outputs after running inference
:return: dictionary mapping: {node output tensor names: node output tensor }
'''
# conduct inference session and get intermediate outputs
so = onnxruntime.SessionOptions()
if sys.version_info < (3,10): # pragma: no cover
from onnxruntime_extensions import get_library_path
so.register_custom_ops_library(get_library_path())
session = onnxruntime.InferenceSession(self.augmented_model.SerializeToString(), so)
intermediate_outputs = []
len_inputs = len(session.get_inputs())
inputs_names = [session.get_inputs()[i].name for i in range(len_inputs)]
output_dicts = {}
node_output_names = [output.name if output.name not in self.dequantized_output \
else self.dequantized_output[output.name] \
for output in session.get_outputs()]
for idx, (inputs, labels) in enumerate(self.dataloader):
ort_inputs = {}
if len_inputs == 1:
ort_inputs.update(
inputs if isinstance(inputs, dict) else {inputs_names[0]: inputs}
)
else:
assert len_inputs == len(inputs), \
'number of input tensors must align with graph inputs'
if isinstance(inputs, dict): # pragma: no cover
ort_inputs.update(inputs)
else:
for i in range(len_inputs):
if not isinstance(inputs[i], np.ndarray): # pragma: no cover
ort_inputs.update({inputs_names[i]: np.array(inputs[i])})
else:
ort_inputs.update({inputs_names[i]: inputs[i]})
if self.iterations != []:
if idx > max(self.iterations):
break
if idx in self.iterations:
for output_idx, output in enumerate(session.run(None, ort_inputs)):
if calib_mode == 'naive' and output.size != 0:
output_dicts.setdefault(node_output_names[output_idx], \
[]).append([output.min(), output.max()])
elif calib_mode == None:
output_dicts.setdefault(node_output_names[output_idx], \
[]).append(output)
else:
for output_idx, output in enumerate(session.run(None, ort_inputs)):
if calib_mode == 'naive' and output.size != 0:
output_dicts.setdefault(node_output_names[output_idx], \
[]).append([output.min(), output.max()])
elif calib_mode == None:
output_dicts.setdefault(node_output_names[output_idx], \
[]).append(output)
return list(output_dicts.keys()), output_dicts
def _dequantize(self, tensor, scale_tensor, zo_tensor):
''' helper function to dequantize tensor
'''
int_tensor = self.model_wrapper.get_initializer(tensor)
if int_tensor: # weight tensor
return self._dequantize_weight(tensor, scale_tensor, zo_tensor)
else:
return self._dequantize_activation(tensor, scale_tensor, zo_tensor)
def _dequantize_activation(self, activation_tensor_name, scale_tensor, zo_tensor):
''' helper funtion to dequantize activation'''
added_nodes, added_output = self._add_dequantize_node(activation_tensor_name, \
scale_tensor, zo_tensor)
self.dequantized_output[added_output] = activation_tensor_name
return added_nodes, added_output
def _dequantize_weight(self, weight_tensor_name, scale_tensor, zo_tensor):
''' helper function to dequantize weight'''
weight_tensor = self.model_wrapper.get_initializer(weight_tensor_name)
if len(scale_tensor.dims) in [1, 2] and weight_tensor.dims[0] == max(scale_tensor.dims):
logger.debug("weight {} is quantized with per channel granularity."
.format(weight_tensor_name))
if self.opset_version < 13 and self.ort_version >= ORT112_VERSION:
logger.warning("Skip dequantizing weight {}, please use onnxruntime < 1.12.0 " \
"or upgrade model opset version to 13 or higher".format(weight_tensor_name))
return [], None
node = self.model_wrapper.input_name_to_nodes[weight_tensor_name][0]
if 'Conv' in node.op_type or \
('Gemm' in node.op_type and is_B_transposed(node)):
added_nodes, added_output = self._add_dequantize_transpose_node(
weight_tensor_name,
scale_tensor, zo_tensor,
len(weight_tensor.dims))
else:
added_nodes, added_output = self._add_dequantize_node(
weight_tensor_name,
scale_tensor,
zo_tensor,
axis=1 if self.opset_version > 12 else None)
else:
added_nodes, added_output = self._add_dequantize_node(weight_tensor_name,
scale_tensor,\
zo_tensor)
self.dequantized_output[added_output] = weight_tensor_name
return added_nodes, added_output
def _add_dequantize_node(self, tensor_name, scale_tensor, zo_tensor, axis=None):
'''helper function to generate dequantize node'''
dequantize_node = make_dquant_node(tensor_name + '_DequantizeLinear',
[tensor_name,
scale_tensor.name,
zo_tensor.name],
[tensor_name + '_output'],
axis)
return [dequantize_node], tensor_name + '_output'
def _add_dequantize_transpose_node(self, tensor_name, scale_tensor, zo_tensor, dim):
pre_transpose_node = onnx.helper.make_node(
'Transpose',
inputs=[tensor_name],
outputs=[tensor_name + '_transposed'],
perm=(1,0,2,3) if dim == 4 else (1,0),
name=tensor_name + '_pre_transpose')
dequantize_node = make_dquant_node(
tensor_name + '_DequantizeLinear',
[tensor_name + '_transposed',
scale_tensor.name,
zo_tensor.name],
[tensor_name + '_DequantizeLinear'],
axis=1 if self.opset_version > 12 else None)
post_transpose_node = onnx.helper.make_node(
'Transpose',
inputs=[tensor_name + '_DequantizeLinear'],
outputs=[tensor_name + '_output'],
perm=(1,0,2,3) if dim == 4 else (1,0),
name=tensor_name + '_post_transpose')
added_nodes = [pre_transpose_node, dequantize_node, post_transpose_node]
return added_nodes, tensor_name + '_output'
def _map_calibration(self, node_output_names, output_dicts, calib_mode='naive'):
merged_dict = {}
for name, minmaxs in output_dicts.items():
for minmax in minmaxs:
merged_dict.setdefault(name + '_Min', []).append(minmax[0])
merged_dict.setdefault(name + '_Max', []).append(minmax[1])
# Characterizing distribution of a node's values across test data sets
clean_merged_dict = dict((i, merged_dict[i]) for i in merged_dict)
if calib_mode == 'naive':
pairs = [
tuple([
float(min(clean_merged_dict[name + '_Min'])),
float(max(clean_merged_dict[name + '_Max']))
]) for name in node_output_names
]
else:
raise ValueError('Unknown value for calib_mode. \
Currently only naive mode is supported.')
final_dict = dict(zip(node_output_names, pairs))
return final_dict
def dump_minmax(self, calib_mode='naive'):
self.augment_graph()
node_output_names, output_dicts = self.get_intermediate_outputs(calib_mode)
return self._map_calibration(node_output_names, output_dicts,
calib_mode=calib_mode)
def dump_calibration(self, calib_mode='naive'):
'''
Gather calibration params for quantization
parameter calib_mode: type 'naive' gives (Min, Max) pairs
for each intermediate model output across
test data sets, where the first element is
a minimum of all values and the
second element is a maximum of all values;
:return: dictionary mapping: {added node names: (ReduceMin, ReduceMax) pairs }
'''
return self.calculate_quantization_params(self.dump_minmax(calib_mode))
def calculate_quantization_params(self, quantization_thresholds):
'''
Given quantization thresholds, calculate the quantization params.
:param quantization_thresholds:
Dictionary specifying the min and max values for outputs of conv and matmul nodes.
The quantization_thresholds should be specified in the following format:
{
"param_name": [min, max]
}
example:
{
'Conv_3:0': [np.float32(0), np.float32(0.5)],
'Conv_4:0': [np.float32(1), np.float32(3.5)]
}
:return: Dictionary containing the zero point and
scale values for outputs of conv and matmul nodes.
The dictionary format is
{
"param_name": [zero_point, scale]
}
'''
if quantization_thresholds is None:
raise ValueError(
'quantization thresholds is required to calculate quantization \
params (zero point and scale)')
quantization_params = {}
model = self.model
input_name_to_nodes = self.model_wrapper.input_name_to_nodes
output_name_to_nodes = self.model_wrapper.output_name_to_node
for tensor_name in quantization_thresholds.keys():
child = None
if tensor_name in input_name_to_nodes:
children = input_name_to_nodes[tensor_name]
if len(children) == 1:
child = children[0]
parent = None
if tensor_name in output_name_to_nodes:
parent = output_name_to_nodes[tensor_name]
node_thresholds = quantization_thresholds[tensor_name]
node_params = self.calculate_scale_zeropoint(parent, child, node_thresholds[0],
node_thresholds[1])
quantization_params[tensor_name] = node_params
return quantization_params
def dump_tensor(self, activation=True, weight=False):
if "QuantizeLinear" in [node.op_type for node in self.model.graph.node] or \
"DynamicQuantizeLinear" in [node.op_type for node in self.model.graph.node]:
self.augment_nodes = ["DequantizeLinear"]
self.already_quantized = True
self.dynamically_quantized = \
"DynamicQuantizeLinear" in [node.op_type for node in self.model.graph.node]
self.augment_graph(activation_only=not weight, weight_only=not activation)
_, output_dicts = self.get_intermediate_outputs()
iters = len(list(output_dicts.values())[-1])
map_node_activation = [{} for _ in range(iters)]
map_node_weight = {}
self.white_nodes = [node.replace('_quant', '') for node in self.white_nodes]
augmengted_wrapper = ONNXModel(self.augmented_model)
map_output = augmengted_wrapper.output_name_to_node
map_input = augmengted_wrapper.input_name_to_nodes
model_output_names = [t.name for t in self.model.graph.output]
model_input_names = [t.name for t in self.model.graph.input]
model_initializer_names = [t.name for t in self.model.graph.initializer]
for tensor_name, tensors in output_dicts.items():
if tensor_name.replace('_dequantized', '_quantized') in model_initializer_names:
nodes = [node for node in map_input[tensor_name] \
if node.name.replace('_quant', '') in self.white_nodes]
elif tensor_name.replace('_quantized', '') in model_input_names:
continue
else:
nodes = [map_output[tensor_name]]
for node in nodes:
node_name = node.name.replace('_quant', '')
if tensor_name in model_output_names and node_name not in self.white_nodes:
continue
while node_name not in self.white_nodes and self.already_quantized:
node = augmengted_wrapper.get_parents(node, output_name_to_node=map_output)[0]
node_name = node.name.replace('_quant', '')
if node_name not in self.white_nodes:
continue
if node_name not in map_node_weight:
map_node_weight[node_name] = {}
if tensor_name not in model_initializer_names:
for i in range(iters):
map_node_activation[i][node_name] = \
{tensor_name.replace('_quantized', ''): tensors[i]}
else:
map_node_weight[node_name].update({tensor_name.replace('_quantized', ''): \
tensors[0]})
dumped_tensors_map = {}
if weight:
dumped_tensors_map.update({"weight": map_node_weight})
if activation:
dumped_tensors_map.update({"activation": map_node_activation})
return dumped_tensors_map
def calculate_scale_zeropoint(self, last_node, next_node, rmin, rmax):
'''
Given the source and destination node of tensor, \
return calculated zero point and scales.
:param last_node: the source of the tensor
:param next_node: the destination of the tensor
:param rmin: min threshold of the tensor
:param rmax: max threshold of the tensor
:return (List): zero_point and scale
'''
zp_and_scale = []
# adjust rmin and rmax such that 0 is included in the range. This is required
# to make sure zero can be uniquely represented.
rmin = min(rmin, 0)
rmax = max(rmax, 0)
if next_node:
if next_node.op_type == 'Relu':
if rmin < 0:
rmin = 0
elif next_node.op_type == 'Clip' and len(next_node.input) ==3:
if rmin < numpy_helper.to_array(
self.model_wrapper.get_initializer(next_node.input[1])):
rmin = numpy_helper.to_array(
self.model_wrapper.get_initializer(next_node.input[1]))
if rmax > numpy_helper.to_array(
self.model_wrapper.get_initializer(next_node.input[2])):
rmax = numpy_helper.to_array(
self.model_wrapper.get_initializer(next_node.input[2]))
if last_node:
if last_node.op_type in ['Conv', 'FusedConv']:
attrs = [attr for attr in last_node.attribute]
attrs_names = [attr.name for attr in last_node.attribute]
if 'activation' in attrs_names:
if attrs[attrs_names.index('activation')].s == b'Relu':
rmin = max(rmin, 0)
if attrs[attrs_names.index('activation')].s == b'Clip':
assert 'activation_params' in attrs_names, "the model contains no \
params for clip node \
{}".format(last_node)
clip_params = attrs[attrs_names.index('activation_params')].floats
rmin = min(rmin, clip_params[0], clip_params[1])
rmax = max(rmax, clip_params[0], clip_params[1])
scale = np.float32((rmax - rmin) / 255 if rmin != rmax else 1)
initial_zero_point = (0 - rmin) / scale
zero_point = np.uint8(round(max(0, min(255, initial_zero_point))))
zp_and_scale.append(zero_point)
zp_and_scale.append(scale)
return zp_and_scale