-
Notifications
You must be signed in to change notification settings - Fork 21.4k
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
[ONNX] Support converting fx graph with symbolic shape to ONNX #96350
Commits on Mar 8, 2023
-
Configuration menu - View commit details
-
Copy full SHA for 4c1d6e0 - Browse repository at this point
Copy the full SHA 4c1d6e0View commit details -
Configuration menu - View commit details
-
Copy full SHA for 22b362f - Browse repository at this point
Copy the full SHA 22b362fView commit details
Commits on Mar 9, 2023
-
Configuration menu - View commit details
-
Copy full SHA for 7ccc994 - Browse repository at this point
Copy the full SHA 7ccc994View commit details -
Configuration menu - View commit details
-
Copy full SHA for 6bc993a - Browse repository at this point
Copy the full SHA 6bc993aView commit details -
Configuration menu - View commit details
-
Copy full SHA for 8a6d9ad - Browse repository at this point
Copy the full SHA 8a6d9adView commit details
Commits on Mar 10, 2023
-
Configuration menu - View commit details
-
Copy full SHA for 9bd0919 - Browse repository at this point
Copy the full SHA 9bd0919View commit details
Commits on Mar 13, 2023
-
Configuration menu - View commit details
-
Copy full SHA for 455a3fb - Browse repository at this point
Copy the full SHA 455a3fbView commit details -
Configuration menu - View commit details
-
Copy full SHA for d41603e - Browse repository at this point
Copy the full SHA d41603eView commit details -
Configuration menu - View commit details
-
Copy full SHA for 54bb992 - Browse repository at this point
Copy the full SHA 54bb992View commit details -
Update on "[ONNX] Support dynamic axes"
Need microsoft/onnxscript#484 [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for c957d8a - Browse repository at this point
Copy the full SHA c957d8aView commit details -
Update on "[ONNX] Support dynamic axes"
Need microsoft/onnxscript#484 [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for 2b09207 - Browse repository at this point
Copy the full SHA 2b09207View commit details -
Update on "[ONNX] Support dynamic axes"
Need microsoft/onnxscript#484 [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for 3ea2434 - Browse repository at this point
Copy the full SHA 3ea2434View commit details -
Update on "[ONNX] Support dynamic axes"
Need microsoft/onnxscript#484 [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for 06fdede - Browse repository at this point
Copy the full SHA 06fdedeView commit details
Commits on Mar 14, 2023
-
Update on "[ONNX] Support dynamic axes"
Need microsoft/onnxscript#484 Support dynamic export on fx-ONNX exporter. Essentially, we set inputs size and nodes all dynamic in torchscript, and leverage on `aten::sym_size` to catch dynamic size between each Op. 1. Add `dynamic_axes` switch to avoid affecting what we have on static export. 2. Add test_fx_dynamic_onnruntime.py to test on some addhoc we have from old exporter. 3. Since `aten::sym_size` are operated with built-in function, a built-in function mapping is added. 4. sym_size output value is also fx.Node, and can be found in `fx_name_to_onnxscipt_value`, so it's operation stays the same as other ONNX ops in ONNX graph. [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for 5fdae17 - Browse repository at this point
Copy the full SHA 5fdae17View commit details -
Update on "[ONNX] Support dynamic axes"
Need microsoft/onnxscript#484 Support dynamic export on fx-ONNX exporter. Essentially, we set inputs size and nodes all dynamic in torchscript, and leverage on `aten::sym_size` to catch dynamic size between each Op. 1. Add `dynamic_axes` switch to avoid affecting what we have on static export. 2. Add test_fx_dynamic_onnruntime.py to test on some addhoc we have from old exporter. 3. Since `aten::sym_size` are operated with built-in function, a built-in function mapping is added. 4. sym_size output value is also fx.Node, and can be found in `fx_name_to_onnxscipt_value`, so it's operation stays the same as other ONNX ops in ONNX graph. [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for 35f192f - Browse repository at this point
Copy the full SHA 35f192fView commit details
Commits on Mar 15, 2023
-
Update on "[ONNX] Support dynamic axes"
Need microsoft/onnxscript#484 Support dynamic export on fx-ONNX exporter. Essentially, we set inputs size and nodes all dynamic in torchscript, and leverage on `aten::sym_size` to catch dynamic size between each Op. 1. Add `dynamic_axes` switch to avoid affecting what we have on static export. 2. Add test_fx_dynamic_onnruntime.py to test on some addhoc we have from old exporter. 3. Since `aten::sym_size` are operated with built-in function, a built-in function mapping is added. 4. sym_size output value is also fx.Node, and can be found in `fx_name_to_onnxscipt_value`, so it's operation stays the same as other ONNX ops in ONNX graph. [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for 5e116ac - Browse repository at this point
Copy the full SHA 5e116acView commit details -
Update on "[ONNX] Support dynamic axes"
Need microsoft/onnxscript#484 Support dynamic export on fx-ONNX exporter. Essentially, we set inputs size and nodes all dynamic in torchscript, and leverage on `aten::sym_size` to catch dynamic size between each Op. 1. Add `dynamic_axes` switch to avoid affecting what we have on static export. 2. Add test_fx_dynamic_onnruntime.py to test on some addhoc we have from old exporter. 3. Since `aten::sym_size` are operated with built-in function, a built-in function mapping is added. 4. sym_size output value is also fx.Node, and can be found in `fx_name_to_onnxscipt_value`, so it's operation stays the same as other ONNX ops in ONNX graph. [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for bef15c0 - Browse repository at this point
Copy the full SHA bef15c0View commit details -
Update on "[ONNX] Support dynamic axes"
Need microsoft/onnxscript#484 Support dynamic export on fx-ONNX exporter. Essentially, we set inputs size and nodes all dynamic in torchscript, and leverage on `aten::sym_size` to catch dynamic size between each Op. 1. Add `dynamic_axes` switch to avoid affecting what we have on static export. 2. Add test_fx_dynamic_onnruntime.py to test on some addhoc we have from old exporter. 3. Since `aten::sym_size` are operated with built-in function, a built-in function mapping is added. 4. sym_size output value is also fx.Node, and can be found in `fx_name_to_onnxscipt_value`, so it's operation stays the same as other ONNX ops in ONNX graph. [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for 49195a4 - Browse repository at this point
Copy the full SHA 49195a4View commit details
Commits on Mar 19, 2023
-
Update on "[ONNX] Support dynamic axes"
Need microsoft/onnxscript#484 Support dynamic export on fx-ONNX exporter. Essentially, we set inputs size and nodes all dynamic in torchscript, and leverage on `aten::sym_size` to catch dynamic size between each Op. 1. Add `dynamic_axes` switch to avoid affecting what we have on static export. 2. Add test_fx_dynamic_onnruntime.py to test on some addhoc we have from old exporter. 3. Since `aten::sym_size` are operated with built-in function, a built-in function mapping is added. 4. sym_size output value is also fx.Node, and can be found in `fx_name_to_onnxscipt_value`, so it's operation stays the same as other ONNX ops in ONNX graph. [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for 471a0dc - Browse repository at this point
Copy the full SHA 471a0dcView commit details
Commits on Mar 21, 2023
-
Update on "[ONNX] Support dynamic axes"
Need microsoft/onnxscript#484 Support dynamic export on fx-ONNX exporter. Essentially, we set inputs size and nodes all dynamic in torchscript, and leverage on `aten::sym_size` to catch dynamic size between each Op. 1. Add `dynamic_axes` switch to avoid affecting what we have on static export. 2. Add test_fx_dynamic_onnruntime.py to test on some addhoc we have from old exporter. 3. Since `aten::sym_size` are operated with built-in function, a built-in function mapping is added. 4. sym_size output value is also fx.Node, and can be found in `fx_name_to_onnxscipt_value`, so it's operation stays the same as other ONNX ops in ONNX graph. [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for d60f5b3 - Browse repository at this point
Copy the full SHA d60f5b3View commit details -
Update on "[ONNX] Support dynamic axes"
Need microsoft/onnxscript#484 Support dynamic export on fx-ONNX exporter. Essentially, we set inputs size and nodes all dynamic in torchscript, and leverage on `aten::sym_size` to catch dynamic size between each Op. 1. Add `dynamic_axes` switch to avoid affecting what we have on static export. 2. Add test_fx_dynamic_onnruntime.py to test on some addhoc we have from old exporter. 3. Since `aten::sym_size` are operated with built-in function, a built-in function mapping is added. 4. sym_size output value is also fx.Node, and can be found in `fx_name_to_onnxscipt_value`, so it's operation stays the same as other ONNX ops in ONNX graph. [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for fe4542b - Browse repository at this point
Copy the full SHA fe4542bView commit details -
Update on "[ONNX] Support dynamic axes"
Need microsoft/onnxscript#484 Support dynamic export on fx-ONNX exporter. Essentially, we set inputs size and nodes all dynamic in torchscript, and leverage on `aten::sym_size` to catch dynamic size between each Op. 1. Add `dynamic_axes` switch to avoid affecting what we have on static export. 2. Add test_fx_dynamic_onnruntime.py to test on some addhoc we have from old exporter. 3. Since `aten::sym_size` are operated with built-in function, a built-in function mapping is added. 4. sym_size output value is also fx.Node, and can be found in `fx_name_to_onnxscipt_value`, so it's operation stays the same as other ONNX ops in ONNX graph. [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for b8a67e3 - Browse repository at this point
Copy the full SHA b8a67e3View commit details
Commits on Mar 22, 2023
-
Update on "[ONNX] Support converting fx graph with symbolic shape to …
…ONNX" ~~Need microsoft/onnxscript#484 Support dynamic export on fx-ONNX exporter. Essentially, we set inputs size and nodes all dynamic in torchscript, and leverage on `aten::sym_size` to catch dynamic size between each Op. 1. Add `dynamic_axes` switch between symbolic tracing (dynamic sizes) and fake mode (static). Set it to default True, as most of our tests are happy with sumbolic tracing. Except GPT2 stays with fake mode with error: microsoft/onnxscript#523 2. Add test_fx_dynamic_onnruntime.py to test on some addhoc we have from old exporter. This can be removed once tests are integrated with #96479 3. Since `aten::sym_size` are operated with built-in function, a built-in function mapping is added to support SymFloat/SymInt. (FIXME: #97201). sym_size output value is also fx.Node, and can be found in `fx_name_to_onnxscipt_value`, so it's operation stays the same as other ONNX ops in ONNX graph. 4. Fully deprecated FakeTensorProp as make_fx() should provide all node meta info. 5. Put complicated fx.Node related ArgumentType into _type_utils.py [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for 6abe6bc - Browse repository at this point
Copy the full SHA 6abe6bcView commit details
Commits on Mar 24, 2023
-
Update on "[ONNX] Support converting fx graph with symbolic shape to …
…ONNX" ~~Need microsoft/onnxscript#484 Support dynamic export on fx-ONNX exporter. Essentially, we set inputs size and nodes all dynamic in torchscript, and leverage on `aten::sym_size` to catch dynamic size between each Op. 1. Add `dynamic_axes` switch between symbolic tracing (dynamic sizes) and fake mode (static). Set it to default True, as most of our tests are happy with sumbolic tracing. Except GPT2 stays with fake mode with error: microsoft/onnxscript#523 2. Add test_fx_dynamic_onnruntime.py to test on some addhoc we have from old exporter. This can be removed once tests are integrated with #96479 3. Since `aten::sym_size` are operated with built-in function, a built-in function mapping is added to support SymFloat/SymInt. (FIXME: #97201). sym_size output value is also fx.Node, and can be found in `fx_name_to_onnxscipt_value`, so it's operation stays the same as other ONNX ops in ONNX graph. 4. Fully deprecated FakeTensorProp as make_fx() should provide all node meta info. 5. Put complicated fx.Node related ArgumentType into _type_utils.py [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for 84f05a2 - Browse repository at this point
Copy the full SHA 84f05a2View commit details -
Update on "[ONNX] Support converting fx graph with symbolic shape to …
…ONNX" ~~Need microsoft/onnxscript#484 Support dynamic export on fx-ONNX exporter. Essentially, we set inputs size and nodes all dynamic in torchscript, and leverage on `aten::sym_size` to catch dynamic size between each Op. 1. Add `dynamic_axes` switch between symbolic tracing (dynamic sizes) and fake mode (static). Set it to default True, as most of our tests are happy with sumbolic tracing. Except GPT2 stays with fake mode with error: microsoft/onnxscript#523 2. Add test_fx_dynamic_onnruntime.py to test on some addhoc we have from old exporter. This can be removed once tests are integrated with #96479 3. Since `aten::sym_size` are operated with built-in function, a built-in function mapping is added to support SymFloat/SymInt. (FIXME: #97201). sym_size output value is also fx.Node, and can be found in `fx_name_to_onnxscipt_value`, so it's operation stays the same as other ONNX ops in ONNX graph. 4. Fully deprecated FakeTensorProp as make_fx() should provide all node meta info. 5. Put complicated fx.Node related ArgumentType into _type_utils.py [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for 87e3ea0 - Browse repository at this point
Copy the full SHA 87e3ea0View commit details -
Update on "[ONNX] Support converting fx graph with symbolic shape to …
…ONNX" ~~Need microsoft/onnxscript#484 Support dynamic export on fx-ONNX exporter. Essentially, we set inputs size and nodes all dynamic in torchscript, and leverage on `aten::sym_size` to catch dynamic size between each Op. 1. Add `dynamic_axes` switch between symbolic tracing (dynamic sizes) and fake mode (static). Set it to default True, as most of our tests are happy with sumbolic tracing. Except GPT2 stays with fake mode with error: microsoft/onnxscript#523 2. Add test_fx_dynamic_onnruntime.py to test on some addhoc we have from old exporter. This can be removed once tests are integrated with #96479 3. Since `aten::sym_size` are operated with built-in function, a built-in function mapping is added to support SymFloat/SymInt. (FIXME: #97201). sym_size output value is also fx.Node, and can be found in `fx_name_to_onnxscipt_value`, so it's operation stays the same as other ONNX ops in ONNX graph. 4. Fully deprecated FakeTensorProp as make_fx() should provide all node meta info. 5. Put complicated fx.Node related ArgumentType into _type_utils.py [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for 030c219 - Browse repository at this point
Copy the full SHA 030c219View commit details