ai-edge-quantizer-nightly 0.3.0.dev20250716__py3-none-any.whl → 0.3.0.dev20250718__py3-none-any.whl
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- ai_edge_quantizer/algorithm_manager.py +2 -0
- ai_edge_quantizer/algorithms/uniform_quantize/common_quantize.py +15 -0
- ai_edge_quantizer/calibrator_test.py +1 -1
- ai_edge_quantizer/default_policy.py +4 -2
- ai_edge_quantizer/qtyping.py +1 -0
- ai_edge_quantizer/utils/tfl_flatbuffer_utils.py +1 -0
- {ai_edge_quantizer_nightly-0.3.0.dev20250716.dist-info → ai_edge_quantizer_nightly-0.3.0.dev20250718.dist-info}/METADATA +1 -1
- {ai_edge_quantizer_nightly-0.3.0.dev20250716.dist-info → ai_edge_quantizer_nightly-0.3.0.dev20250718.dist-info}/RECORD +11 -11
- {ai_edge_quantizer_nightly-0.3.0.dev20250716.dist-info → ai_edge_quantizer_nightly-0.3.0.dev20250718.dist-info}/LICENSE +0 -0
- {ai_edge_quantizer_nightly-0.3.0.dev20250716.dist-info → ai_edge_quantizer_nightly-0.3.0.dev20250718.dist-info}/WHEEL +0 -0
- {ai_edge_quantizer_nightly-0.3.0.dev20250716.dist-info → ai_edge_quantizer_nightly-0.3.0.dev20250718.dist-info}/top_level.txt +0 -0
@@ -116,6 +116,7 @@ MIN_MAX_OP_NAME_MATERIALIZE_FUNC_DICT = {
|
|
116
116
|
_TFLOpName.GATHER_ND: common_quantize.materialize_gather_nd,
|
117
117
|
_TFLOpName.PACK: common_quantize.materialize_pack,
|
118
118
|
_TFLOpName.UNPACK: common_quantize.materialize_unpack,
|
119
|
+
_TFLOpName.DIV: common_quantize.materialize_div,
|
119
120
|
}
|
120
121
|
for op_name, materialize_func in MIN_MAX_OP_NAME_MATERIALIZE_FUNC_DICT.items():
|
121
122
|
register_quantized_op(
|
@@ -258,6 +259,7 @@ _OCTAV_OP_NAME_MATERIALIZE_FUNC_DICT = immutabledict({
|
|
258
259
|
_TFLOpName.GATHER_ND: common_quantize.materialize_gather_nd,
|
259
260
|
_TFLOpName.PACK: common_quantize.materialize_pack,
|
260
261
|
_TFLOpName.UNPACK: common_quantize.materialize_unpack,
|
262
|
+
_TFLOpName.DIV: common_quantize.materialize_div,
|
261
263
|
})
|
262
264
|
|
263
265
|
for op_name, materialize_func in _OCTAV_OP_NAME_MATERIALIZE_FUNC_DICT.items():
|
@@ -794,6 +794,21 @@ def materialize_unpack(
|
|
794
794
|
)
|
795
795
|
|
796
796
|
|
797
|
+
def materialize_div(
|
798
|
+
get_tensor_quant_params_fn: qtyping.GetTensorQuantParamsFuncSignature,
|
799
|
+
op_info: qtyping.OpInfo,
|
800
|
+
graph_info: qtyping.GraphInfo,
|
801
|
+
tensor_name_to_qsv: dict[str, Any],
|
802
|
+
) -> list[qtyping.TensorTransformationParams]:
|
803
|
+
"""Materialize tensors in tfl.div."""
|
804
|
+
return common_utils.materialize_standard_op(
|
805
|
+
op_info,
|
806
|
+
graph_info,
|
807
|
+
tensor_name_to_qsv,
|
808
|
+
get_tensor_quant_params_fn,
|
809
|
+
)
|
810
|
+
|
811
|
+
|
797
812
|
def _get_tensor_shape_for_blockwise(
|
798
813
|
tensor_shape: Sequence[int], quantized_dim: int, block_size: int
|
799
814
|
) -> list[int]:
|
@@ -302,7 +302,7 @@ class CalibratorToyGemma2Test(googletest.TestCase):
|
|
302
302
|
self._toy_gemma2_calibration_dataset,
|
303
303
|
model_recipe_manager=recipe_mngr,
|
304
304
|
)
|
305
|
-
self.assertLen(calib.get_model_qsvs(),
|
305
|
+
self.assertLen(calib.get_model_qsvs(), 288)
|
306
306
|
|
307
307
|
|
308
308
|
if __name__ == "__main__":
|
@@ -189,7 +189,8 @@ DEFAULT_JSON_POLICY = """
|
|
189
189
|
"RESIZE_BILINEAR",
|
190
190
|
"GATHER_ND",
|
191
191
|
"PACK",
|
192
|
-
"UNPACK"
|
192
|
+
"UNPACK",
|
193
|
+
"DIV"
|
193
194
|
],
|
194
195
|
"static_wi8_ai8": [
|
195
196
|
"ADD",
|
@@ -227,7 +228,8 @@ DEFAULT_JSON_POLICY = """
|
|
227
228
|
"RESIZE_BILINEAR",
|
228
229
|
"GATHER_ND",
|
229
230
|
"PACK",
|
230
|
-
"UNPACK"
|
231
|
+
"UNPACK",
|
232
|
+
"DIV"
|
231
233
|
],
|
232
234
|
"static_wi4_ai8": ["FULLY_CONNECTED", "CONV_2D", "INPUT", "OUTPUT", "EMBEDDING_LOOKUP"],
|
233
235
|
"static_wi4_ai16": ["FULLY_CONNECTED", "CONV_2D", "INPUT", "OUTPUT", "EMBEDDING_LOOKUP"],
|
ai_edge_quantizer/qtyping.py
CHANGED
@@ -63,6 +63,7 @@ TFL_OP_NAME_TO_CODE = immutabledict.immutabledict({
|
|
63
63
|
_TFLOpName.GATHER_ND: schema.BuiltinOperator.GATHER_ND,
|
64
64
|
_TFLOpName.PACK: schema.BuiltinOperator.PACK,
|
65
65
|
_TFLOpName.UNPACK: schema.BuiltinOperator.UNPACK,
|
66
|
+
_TFLOpName.DIV: schema.BuiltinOperator.DIV,
|
66
67
|
})
|
67
68
|
|
68
69
|
TFL_OP_CODE_TO_NAME = immutabledict.immutabledict(
|
@@ -1,6 +1,6 @@
|
|
1
1
|
Metadata-Version: 2.1
|
2
2
|
Name: ai-edge-quantizer-nightly
|
3
|
-
Version: 0.3.0.
|
3
|
+
Version: 0.3.0.dev20250718
|
4
4
|
Summary: A quantizer for advanced developers to quantize converted AI Edge models.
|
5
5
|
Home-page: https://github.com/google-ai-edge/ai-edge-quantizer
|
6
6
|
Keywords: On-Device ML,AI,Google,TFLite,Quantization,LLMs,GenAI
|
@@ -1,18 +1,18 @@
|
|
1
1
|
ai_edge_quantizer/__init__.py,sha256=4pFSkukSwahYyzwqia0yPRyz8TnFQfGRthVJhYpMWas,793
|
2
|
-
ai_edge_quantizer/algorithm_manager.py,sha256
|
2
|
+
ai_edge_quantizer/algorithm_manager.py,sha256=776VBmJerYII1TYwauxjg5zgUuY38lcVMQy3joR1FG0,12680
|
3
3
|
ai_edge_quantizer/algorithm_manager_api.py,sha256=u903TG0s1uIDhJqfeJne3CFl8A93phZrwgV2-hwdcXU,9247
|
4
4
|
ai_edge_quantizer/algorithm_manager_api_test.py,sha256=w6bSONvXkX6bzXAGc0-7b6gNDt9oz9ieq97KP8Sg_JU,7666
|
5
5
|
ai_edge_quantizer/calibrator.py,sha256=Sms7_AIHPH9G5xFaz5Ef3a5gPhxuIWQI8d2LUM8C96I,12071
|
6
|
-
ai_edge_quantizer/calibrator_test.py,sha256=
|
6
|
+
ai_edge_quantizer/calibrator_test.py,sha256=ejKc5YC7id8J1Ll9HAYCzMnKzxd0FUENSD06zkSSV0c,11900
|
7
7
|
ai_edge_quantizer/conftest.py,sha256=SxCz-5LlRD_lQm4hQc4c6IGG7DS8d7IyEWY9gnscPN0,794
|
8
|
-
ai_edge_quantizer/default_policy.py,sha256=
|
8
|
+
ai_edge_quantizer/default_policy.py,sha256=JkvwaQEqvBh1BKBNwS179wbgJp7C5M-bkhSJoN_wDZQ,11366
|
9
9
|
ai_edge_quantizer/model_modifier.py,sha256=teGa8I6kGvn6TQY6Xv53YFIc_pQEhNvM9Zb4bvhezyw,7110
|
10
10
|
ai_edge_quantizer/model_modifier_test.py,sha256=cJd04SLOG-fQZZNZPcisoBLx3cLtWEwGqUBbLb-pif4,4751
|
11
11
|
ai_edge_quantizer/model_validator.py,sha256=Hj0_5o-Oa3dSlJ3ryVjRhvsyelHNyek1GrtG9buMczg,13153
|
12
12
|
ai_edge_quantizer/model_validator_test.py,sha256=EeqOP_mrZsnZ3rug756s0ryDDqd2KgIDld5Lm_gDuWY,13020
|
13
13
|
ai_edge_quantizer/params_generator.py,sha256=hcgMHJlERZERUyIAEi6AHJcLJ8gsKIBAEojzFFz-tqk,20098
|
14
14
|
ai_edge_quantizer/params_generator_test.py,sha256=RDYoRZDJfEZRtjlTAU2kZ_4t3JHOqEHxfJX9V4ETAhg,40597
|
15
|
-
ai_edge_quantizer/qtyping.py,sha256=
|
15
|
+
ai_edge_quantizer/qtyping.py,sha256=J81aZXCe5jUDsQLP4T6slCmVIfN1R-sduGBXOLGuaRA,16654
|
16
16
|
ai_edge_quantizer/quantizer.py,sha256=g3DMqFMrMpt9jQttCE0WcdNbMtk0JZnmN5MmCHrNdyM,13202
|
17
17
|
ai_edge_quantizer/quantizer_test.py,sha256=K_HBA56JkFI3HL8VLWCqGEfC0ISh5ldMKoNyBdGRAJg,20368
|
18
18
|
ai_edge_quantizer/recipe.py,sha256=FR0uJceumZrnle2VRSOQZ1uXup4S1cTYKRH-N53mWRo,2919
|
@@ -28,7 +28,7 @@ ai_edge_quantizer/algorithms/nonlinear_quantize/__init__.py,sha256=lpq1g2ayg3lCP
|
|
28
28
|
ai_edge_quantizer/algorithms/nonlinear_quantize/float_casting.py,sha256=Bs9CK7wZAw6jNaZ8xEtbwO2vM34VYXNZSMVWvxJo9nw,9297
|
29
29
|
ai_edge_quantizer/algorithms/nonlinear_quantize/float_casting_test.py,sha256=EqIHGEZ1LgUrTN7zf880RuAzEv3Qy7kgh5ivObJGHSo,22646
|
30
30
|
ai_edge_quantizer/algorithms/uniform_quantize/__init__.py,sha256=lpq1g2ayg3lCPLy79t2VicYcnGKw64FfYIj1V7J-4m8,676
|
31
|
-
ai_edge_quantizer/algorithms/uniform_quantize/common_quantize.py,sha256=
|
31
|
+
ai_edge_quantizer/algorithms/uniform_quantize/common_quantize.py,sha256=LH6tBhgWeZWwhSL3hG8kUgh5bmtHAs_issHFsEwagkM,32221
|
32
32
|
ai_edge_quantizer/algorithms/uniform_quantize/common_quantize_test.py,sha256=GGf_n3wIeg3GB_eGsmyNJ0fTcxgpeMMbugTMRONK6TQ,3553
|
33
33
|
ai_edge_quantizer/algorithms/uniform_quantize/dequantized_weight_recovery.py,sha256=BDdn_uBZakfHyzdMJPKadsOqxqyC-s6W2ZzFH99L4fE,8652
|
34
34
|
ai_edge_quantizer/algorithms/uniform_quantize/dequantized_weight_recovery_test.py,sha256=sT5eX5TLZEHTtPfnSkCPDlS0sQxlTFWbCsbvOuj--yY,8889
|
@@ -62,14 +62,14 @@ ai_edge_quantizer/utils/__init__.py,sha256=lpq1g2ayg3lCPLy79t2VicYcnGKw64FfYIj1V
|
|
62
62
|
ai_edge_quantizer/utils/calibration_utils.py,sha256=e3dG7Nm94Ix0hkTWTWPUhEG6a8QR_cAM3PSwblfJV5g,15106
|
63
63
|
ai_edge_quantizer/utils/calibration_utils_test.py,sha256=4BlksXl7b4yptL8xPR67hmJCnjhN9V10a2PunzfHrUE,9372
|
64
64
|
ai_edge_quantizer/utils/test_utils.py,sha256=spqUmSNciOKPQHCBkHE7Zo34eMFq_BfBCAnMT3jAulU,8615
|
65
|
-
ai_edge_quantizer/utils/tfl_flatbuffer_utils.py,sha256=
|
65
|
+
ai_edge_quantizer/utils/tfl_flatbuffer_utils.py,sha256=vCrFYE3lbnn4eP_L-P6ymjehV8KYpQyqQjoqn2OxOGo,11042
|
66
66
|
ai_edge_quantizer/utils/tfl_flatbuffer_utils_test.py,sha256=K1SbK8q92qYVtiVj0I0GtugsPTkpIpEKv9zakvFV_Sc,8555
|
67
67
|
ai_edge_quantizer/utils/tfl_interpreter_utils.py,sha256=SKSu1nqhGGzVei_DxmzXK-bbOE7G1vKnPDc5skce-yY,14322
|
68
68
|
ai_edge_quantizer/utils/tfl_interpreter_utils_test.py,sha256=6fjkM-rycZ95L4yfvlr0TN6RlrhfPzxNUYrZaYO_F0A,12013
|
69
69
|
ai_edge_quantizer/utils/validation_utils.py,sha256=oYw33Sg547AqtGw-choPUJmp9SAKkV46J_ddqSsum2Q,3950
|
70
70
|
ai_edge_quantizer/utils/validation_utils_test.py,sha256=V_qNDikPD4OPB-siOLQCWNVWTAu87h2IgNYt7teFd-o,2934
|
71
|
-
ai_edge_quantizer_nightly-0.3.0.
|
72
|
-
ai_edge_quantizer_nightly-0.3.0.
|
73
|
-
ai_edge_quantizer_nightly-0.3.0.
|
74
|
-
ai_edge_quantizer_nightly-0.3.0.
|
75
|
-
ai_edge_quantizer_nightly-0.3.0.
|
71
|
+
ai_edge_quantizer_nightly-0.3.0.dev20250718.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
|
72
|
+
ai_edge_quantizer_nightly-0.3.0.dev20250718.dist-info/METADATA,sha256=CW12kg_vsxO2JAEknvHU0DtPcCsS6vK0fM73CajJ-ms,1528
|
73
|
+
ai_edge_quantizer_nightly-0.3.0.dev20250718.dist-info/WHEEL,sha256=tZoeGjtWxWRfdplE7E3d45VPlLNQnvbKiYnx7gwAy8A,92
|
74
|
+
ai_edge_quantizer_nightly-0.3.0.dev20250718.dist-info/top_level.txt,sha256=8QTfPnFXNVUhScFLaa-NWZMFWMn72M50DVPubpwWB1g,18
|
75
|
+
ai_edge_quantizer_nightly-0.3.0.dev20250718.dist-info/RECORD,,
|
File without changes
|
File without changes
|