Skip to content
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.

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

[TFLite] Fix tests so that TensorFlow 2.9 is supported #12130

Merged
merged 1 commit into from
Aug 23, 2022
Merged
Changes from all commits
Commits
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
8 changes: 5 additions & 3 deletions python/tvm/relay/frontend/keras.py
Original file line number Diff line number Diff line change
@@ -635,9 +635,11 @@ def _convert_pooling(
_op.nn.global_max_pool2d(inexpr, **global_pool_params), keras_layer, etab, data_layout
)
if pool_type == "GlobalAveragePooling2D":
return _convert_flatten(
_op.nn.global_avg_pool2d(inexpr, **global_pool_params), keras_layer, etab, data_layout
)
global_avg_pool2d = _op.nn.global_avg_pool2d(inexpr, **global_pool_params)
keep_dims = len(keras_layer.input.shape) == len(keras_layer.output.shape)
if keep_dims:
return global_avg_pool2d
return _convert_flatten(global_avg_pool2d, keras_layer, etab, data_layout)
pool_h, pool_w = keras_layer.pool_size
stride_h, stride_w = keras_layer.strides
params = {
33 changes: 25 additions & 8 deletions tests/python/frontend/tflite/test_forward.py
Original file line number Diff line number Diff line change
@@ -950,6 +950,10 @@ def representative_data_gen():
input_node = subgraph.Tensors(model_input).Name().decode("utf-8")

tflite_output = run_tflite_graph(tflite_model_quant, data)
if tf.__version__ < LooseVersion("2.9"):
input_node = data_in.name.replace(":0", "")
else:
input_node = "serving_default_" + data_in.name + ":0"
tvm_output = run_tvm_graph(tflite_model_quant, data, input_node)
tvm.testing.assert_allclose(
np.squeeze(tvm_output[0]), np.squeeze(tflite_output[0]), rtol=1e-2, atol=1e-2
@@ -1982,10 +1986,12 @@ def _test_abs(data, quantized, int_quant_dtype=tf.int8):
# TFLite 2.6.x upgrade support
if tf.__version__ < LooseVersion("2.6.1"):
in_node = ["serving_default_input_int8"]
else:
elif tf.__version__ < LooseVersion("2.9"):
in_node = (
["serving_default_input_int16"] if int_quant_dtype == tf.int16 else ["tfl.quantize"]
)
else:
in_node = "serving_default_input"

tvm_output = run_tvm_graph(tflite_model_quant, data, in_node)
tvm.testing.assert_allclose(
@@ -2013,8 +2019,10 @@ def _test_rsqrt(data, quantized, int_quant_dtype=tf.int8):
tf.math.rsqrt, data, int_quant_dtype=int_quant_dtype
)
tflite_output = run_tflite_graph(tflite_model_quant, data)
in_node = ["tfl.quantize"]

if tf.__version__ < LooseVersion("2.9"):
in_node = ["tfl.quantize"]
else:
in_node = "serving_default_input"
tvm_output = run_tvm_graph(tflite_model_quant, data, in_node)
tvm.testing.assert_allclose(
np.squeeze(tvm_output[0]), np.squeeze(tflite_output[0]), rtol=1e-5, atol=1e-2
@@ -2095,7 +2103,10 @@ def _test_cos(data, quantized, int_quant_dtype=tf.int8):
tf.math.cos, data, int_quant_dtype=int_quant_dtype
)
tflite_output = run_tflite_graph(tflite_model_quant, data)
in_node = ["tfl.quantize"]
if tf.__version__ < LooseVersion("2.9"):
in_node = ["tfl.quantize"]
else:
in_node = "serving_default_input"
tvm_output = run_tvm_graph(tflite_model_quant, data, in_node)
tvm.testing.assert_allclose(
np.squeeze(tvm_output[0]), np.squeeze(tflite_output[0]), rtol=1e-5, atol=1e-2
@@ -3003,7 +3014,6 @@ def _test_quantize_dequantize(data):
add = tf.keras.layers.Add()([data_in, relu])
concat = tf.keras.layers.Concatenate(axis=0)([relu, add])
keras_model = tf.keras.models.Model(inputs=data_in, outputs=concat)
input_name = data_in.name.split(":")[0]

# To create quantized values with dynamic range of activations, needs representative dataset
def representative_data_gen():
@@ -3013,7 +3023,11 @@ def representative_data_gen():
tflite_model_quant = _quantize_keras_model(keras_model, representative_data_gen, True, True)

tflite_output = run_tflite_graph(tflite_model_quant, data)
tvm_output = run_tvm_graph(tflite_model_quant, data, input_name)
if tf.__version__ < LooseVersion("2.9"):
in_node = data_in.name.split(":")[0]
else:
in_node = "serving_default_" + data_in.name + ":0"
tvm_output = run_tvm_graph(tflite_model_quant, data, in_node)
tvm.testing.assert_allclose(
np.squeeze(tvm_output[0]), np.squeeze(tflite_output[0]), rtol=1e-5, atol=1e-2
)
@@ -3030,7 +3044,6 @@ def _test_quantize_dequantize_const(data):
add = tf.keras.layers.Add()([data, relu])
concat = tf.keras.layers.Concatenate(axis=0)([relu, add])
keras_model = tf.keras.models.Model(inputs=data_in, outputs=concat)
input_name = data_in.name.split(":")[0]

# To create quantized values with dynamic range of activations, needs representative dataset
def representative_data_gen():
@@ -3040,7 +3053,11 @@ def representative_data_gen():
tflite_model_quant = _quantize_keras_model(keras_model, representative_data_gen, True, True)

tflite_output = run_tflite_graph(tflite_model_quant, data)
tvm_output = run_tvm_graph(tflite_model_quant, data, input_name)
if tf.__version__ < LooseVersion("2.9"):
in_node = data_in.name.split(":")[0]
else:
in_node = "serving_default_" + data_in.name + ":0"
tvm_output = run_tvm_graph(tflite_model_quant, data, in_node)
tvm.testing.assert_allclose(
np.squeeze(tvm_output[0]), np.squeeze(tflite_output[0]), rtol=1e-5, atol=1e-2
)