Skip to content

Ignore, caused by diffbase #429

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
copybara-service wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Ignore, caused by diffbase #429

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
78 changes: 50 additions & 28 deletions ai_edge_quantizer/algorithms/nonlinear_quantize/float_casting.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -141,23 +141,38 @@ def materialize_fc_conv(
weight_tensor,
graph_info.buffers,
)
assert weight_content is not None
if not (
quant_params := tensor_quant_params_cache.lookup(
weight_tensor.buffer, _FP16_QUANT_CONFIG
if weight_content is None:
# If the weight data is not stored in the flatbuffer (e.g. dynamic weights),
# skip quantization for this tensor.
op2weight_params = qtyping.OpToTensorParams(
subgraph_op_id=op_info.subgraph_op_index,
transformations=[_QuantTransformation.NO_QUANTIZE],
)
else:
if not (
quant_params := tensor_quant_params_cache.lookup(
weight_tensor.buffer, _FP16_QUANT_CONFIG
)
):
quant_params = qtyping.NonLinearQuantParams(
num_bits=16, quantized_data=weight_content.astype(np.float16)
)
):
tensor_quant_params_cache.insert(
weight_tensor.buffer, _FP16_QUANT_CONFIG, quant_params
)
op2weight_params = qtyping.OpToTensorParams(
subgraph_op_id=op_info.subgraph_op_index,
parameters=quant_params,
transformations=[_QuantTransformation.ADD_DEQUANTIZE],
)
quant_params = qtyping.NonLinearQuantParams(
num_bits=16, quantized_data=weight_content.astype(np.float16)
num_bits=16, quantized_data=weight_content.astype(np.float16) # pytype: disable=attribute-error
)
tensor_quant_params_cache.insert(
weight_tensor.buffer, _FP16_QUANT_CONFIG, quant_params
op2weight_params = qtyping.OpToTensorParams(
subgraph_op_id=op_info.subgraph_op_index,
parameters=quant_params,
transformations=[_QuantTransformation.ADD_DEQUANTIZE],
)
op2weight_params = qtyping.OpToTensorParams(
subgraph_op_id=op_info.subgraph_op_index,
parameters=quant_params,
transformations=[_QuantTransformation.ADD_DEQUANTIZE],
)
op_tensor_params.append(
qtyping.TensorTransformationParams(
tensor_name=tfl_flatbuffer_utils.get_tensor_name(weight_tensor),
Expand Down Expand Up @@ -241,23 +256,30 @@ def materialize_conv2d_transpose(
weight_tensor,
graph_info.buffers,
)
assert weight_content is not None
if not (
quant_params := tensor_quant_params_cache.lookup(
weight_tensor.buffer, _FP16_QUANT_CONFIG
)
):
quant_params = qtyping.NonLinearQuantParams(
num_bits=16, quantized_data=weight_content.astype(np.float16)
if weight_content is None:
# If the weight data is not stored in the flatbuffer (e.g. dynamic weights),
# skip quantization for this tensor.
op2weight_params = qtyping.OpToTensorParams(
subgraph_op_id=op_info.subgraph_op_index,
transformations=[_QuantTransformation.NO_QUANTIZE],
)
tensor_quant_params_cache.insert(
weight_tensor.buffer, _FP16_QUANT_CONFIG, quant_params
else:
if not (
quant_params := tensor_quant_params_cache.lookup(
weight_tensor.buffer, _FP16_QUANT_CONFIG
)
):
quant_params = qtyping.NonLinearQuantParams(
num_bits=16, quantized_data=weight_content.astype(np.float16)
)
tensor_quant_params_cache.insert(
weight_tensor.buffer, _FP16_QUANT_CONFIG, quant_params
)
op2weight_params = qtyping.OpToTensorParams(
subgraph_op_id=op_info.subgraph_op_index,
parameters=quant_params,
transformations=[_QuantTransformation.ADD_DEQUANTIZE],
)
op2weight_params = qtyping.OpToTensorParams(
subgraph_op_id=op_info.subgraph_op_index,
parameters=quant_params,
transformations=[_QuantTransformation.ADD_DEQUANTIZE],
)
op_tensor_params.append(
qtyping.TensorTransformationParams(
tensor_name=tfl_flatbuffer_utils.get_tensor_name(weight_tensor),
Expand Down
109 changes: 109 additions & 0 deletions ai_edge_quantizer/algorithms/nonlinear_quantize/float_casting_test.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,6 +14,7 @@
# ==============================================================================

import pathlib
from unittest import mock

from absl.testing import absltest
from absl.testing import parameterized
Expand Down Expand Up @@ -335,6 +336,50 @@ def test_conv2d_weight_only_succeeds(self):
float_casting.materialize_fc_conv,
)

def test_conv2d_dynamic_weight_succeeds(self):
# Read from Model Explorer.
subgraph0 = self._test_model.subgraphs[0]
subgraph_op_id = 0
op = subgraph0.operators[subgraph_op_id]

op_info = qtyping.OpInfo(
op=op,
op_name=_TFLOpName.CONV_2D,
subgraph_op_index=subgraph_op_id,
op_quant_config=qtyping.OpQuantizationConfig(
weight_tensor_config=_TensorQuantConfig(
num_bits=16, dtype=qtyping.TensorDataType.FLOAT
),
compute_precision=_ComputePrecision.FLOAT, # WEIGHT_ONLY.
explicit_dequantize=True,
),
)

with mock.patch.object(
tfl_flatbuffer_utils, "get_tensor_data", return_value=None
):
tensor_quant_params = float_casting.materialize_fc_conv(
op_info,
self._graph_info,
self._tensor_name_to_qsv,
tensor_quant_params_cache=common_utils.TensorQuantParamsCache(),
)

# 4 tensors (input, weight, bias, output)
self.assertLen(tensor_quant_params, 4)
# The weight tensor is at index 1.
weight_params = tensor_quant_params[1]
self.assertEqual(weight_params.tensor_name, "sequential/conv2d/Conv2D")
self.assertLen(weight_params.consumers, 1)

# Transformation should be NO_QUANTIZE since we mocked get_tensor_data to
# None.
op_params = weight_params.consumers[0]
self.assertSequenceEqual(
op_params.transformations, [_QuantTransformation.NO_QUANTIZE]
)
self.assertIsNone(op_params.parameters)

@parameterized.named_parameters(
dict(
testcase_name="invalid_fc",
Expand Down Expand Up @@ -468,6 +513,70 @@ def test_conv2d_transpose_weight_only_succeeds(self):
is_inbounding_tensor=True,
)

def test_conv2d_transpose_dynamic_weight_succeeds(self):
# Read from Model Explorer.
test_model_path = str(
pathlib.Path(_TEST_DATA_PREFIX_PATH)
/ "single_conv2d_transpose_bias.tflite"
)

test_model = tfl_flatbuffer_utils.read_model(test_model_path)
# The test model has one subgraph for now.
graph_info = qtyping.GraphInfo(
subgraph_tensors=test_model.subgraphs[0].tensors,
buffers=test_model.buffers,
)

subgraph0 = test_model.subgraphs[0]
subgraph_op_id = 0
op = subgraph0.operators[subgraph_op_id]

op_info = qtyping.OpInfo(
op=op,
op_name=_TFLOpName.CONV_2D_TRANSPOSE,
subgraph_op_index=subgraph_op_id,
op_quant_config=qtyping.OpQuantizationConfig(
weight_tensor_config=_TensorQuantConfig(
num_bits=16, dtype=qtyping.TensorDataType.FLOAT
),
compute_precision=_ComputePrecision.FLOAT, # WEIGHT_ONLY.
explicit_dequantize=True,
),
)

with mock.patch.object(
tfl_flatbuffer_utils, "get_tensor_data", return_value=None
):
tensor_quant_params = float_casting.materialize_conv2d_transpose(
op_info,
graph_info,
self._tensor_name_to_qsv,
tensor_quant_params_cache=common_utils.TensorQuantParamsCache(),
)

_, _, bias_tensor, _ = tfl_flatbuffer_utils.parse_fc_bmm_conv_tensors(
op_info.op, graph_info.subgraph_tensors
)

num_configs = 4 if bias_tensor is not None else 3
self.assertLen(tensor_quant_params, num_configs)

# The weight tensor is at index 1.
weight_params = tensor_quant_params[1]
self.assertEqual(
weight_params.tensor_name,
"sequential_5/conv2d_transpose_3/conv2d_transpose",
)
self.assertLen(weight_params.consumers, 1)

# Transformation should be NO_QUANTIZE since we mocked get_tensor_data to
# None.
op_params = weight_params.consumers[0]
self.assertSequenceEqual(
op_params.transformations, [_QuantTransformation.NO_QUANTIZE]
)
self.assertIsNone(op_params.parameters)

def test_depthwise_conv2d_weight_only_succeeds(self):
# Read from Model Explorer.
test_model_path = str(
Expand Down
Loading