Feature/sg 000 proparage fixes from master (#1577)

* [Improvement] max_batches support to training log and tqdm progress bar. (#1554) * Added max_batches support to training log and tqdm progress bar. * Added changing string in accordance which parameter is used (len(loader) of max_batches) * Replaced stopping condition for the epoch with a smaller one (cherry picked from commit 749a9c7) * fix (#1558) Co-authored-by: Eugene Khvedchenya <ekhvedchenya@gmail.com> (cherry picked from commit 8a1d255) * fix (#1564) (cherry picked from commit 24798b0) * Bugfix of model.export() to work correct with bs>1 (#1551) (cherry picked from commit 0515496) * Fixed incorrect automatic variable used (#1565) $@ is the name of the target being generated, and $^ are the dependencies Co-authored-by: Louis-Dupont <35190946+Louis-Dupont@users.noreply.github.com> (cherry picked from commit 43f8bea) * fix typo in class documentation (#1548) Co-authored-by: Eugene Khvedchenya <ekhvedchenya@gmail.com> Co-authored-by: Louis-Dupont <35190946+Louis-Dupont@users.noreply.github.com> (cherry picked from commit ec21383) * Feature/sg 1198 mixed precision automatically changed with warning (#1567) * fix * work with tmpdir * minor change of comment * improve device_config (cherry picked from commit 34fda6c) * Fixed issue with torch 1.12 where _scale_fn_ref is missing in CyclicLR (#1575) (cherry picked from commit 23b4f7a) * Fixed issue with torch 1.12 issue with arange not supporting fp16 for CPU device. (#1574) (cherry picked from commit 1f15c76) --------- Co-authored-by: hakuryuu96 <marchenkophilip@gmail.com> Co-authored-by: Louis-Dupont <35190946+Louis-Dupont@users.noreply.github.com> Co-authored-by: Alessandro Ros <aler9.dev@gmail.com>
Deci-AI · Oct 26, 2023 · e8a51d3 · e8a51d3
1 parent c9c368d
commit e8a51d3
Show file tree

Hide file tree

Showing 23 changed files with 668 additions and 286 deletions.
diff --git a/Makefile b/Makefile
@@ -36,8 +36,8 @@ NOTEBOOKS = src/super_gradients/examples/model_export/models_export.ipynb
 
 # This Makefile target runs notebooks listed below and converts them to markdown files in documentation/source/
 run_and_convert_notebooks_to_docs: $(NOTEBOOKS)
-	jupyter nbconvert --to markdown --output-dir="documentation/source/" --execute $@
+	jupyter nbconvert --to markdown --output-dir="documentation/source/" --execute $^
 
 # This Makefile target runs notebooks listed below and converts them to markdown files in documentation/source/
 check_notebooks_version_match: $(NOTEBOOKS)
-	python tests/verify_notebook_version.py $@
+	python tests/verify_notebook_version.py $^
diff --git a/documentation/source/models_export.md b/documentation/source/models_export.md
@@ -28,6 +28,15 @@ A new export API is introduced in SG 3.2.0. It is aimed to simplify the export p
 - Customising NMS parameters and number of detections per image
 - Customising output format (flat or batched)
 
+
+```python
+!pip install super_gradients==3.3.1
+```
+
+    ERROR: Could not find a version that satisfies the requirement super_gradients==3.3.1 (from versions: 1.3.0, 1.3.1, 1.4.0, 1.5.0, 1.5.1, 1.5.2, 1.6.0, 1.7.1, 1.7.2, 1.7.3, 1.7.4, 1.7.5, 2.0.0, 2.0.1, 2.1.0, 2.2.0, 2.5.0, 2.6.0, 3.0.0, 3.0.1, 3.0.2, 3.0.3, 3.0.4, 3.0.5, 3.0.6, 3.0.7, 3.0.8, 3.0.9, 3.1.0, 3.1.1, 3.1.2, 3.1.3, 3.2.0, 3.2.1, 3.3.0)
+    ERROR: No matching distribution found for super_gradients==3.3.1
+
+
 ### Minimalistic export example
 
 Let start with the most simple example of exporting a model to ONNX format.
@@ -203,9 +212,9 @@ pred_boxes, pred_boxes.shape
              [ 35.71795, 249.40926, 176.62216, 544.69794],
              [182.39618, 249.49301, 301.44122, 529.3324 ],
              ...,
-             [  0.     ,   0.     ,   0.     ,   0.     ],
-             [  0.     ,   0.     ,   0.     ,   0.     ],
-             [  0.     ,   0.     ,   0.     ,   0.     ]]], dtype=float32),
+             [ -1.     ,  -1.     ,  -1.     ,  -1.     ],
+             [ -1.     ,  -1.     ,  -1.     ,  -1.     ],
+             [ -1.     ,  -1.     ,  -1.     ,  -1.     ]]], dtype=float32),
      (1, 1000, 4))
 
 
@@ -219,8 +228,8 @@ pred_scores, pred_scores.shape
 
 
 
-    (array([[0.9694027, 0.9693378, 0.9665707, 0.9619047, 0.7538769, ...,
-             0.       , 0.       , 0.       , 0.       , 0.       ]],
+    (array([[ 0.9694027,  0.9693378,  0.9665707,  0.9619047,  0.7538769, ...,
+             -1.       , -1.       , -1.       , -1.       , -1.       ]],
            dtype=float32),
      (1, 1000))
 
@@ -235,8 +244,8 @@ pred_classes, pred_classes.shape
 
 
 
-    (array([[0, 0, 0, 0, 0, 0, 0, 0, 2, 2, ..., 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]],
-           dtype=int64),
+    (array([[ 0,  0,  0,  0,  0,  0,  0,  0,  2,  2, ..., -1, -1, -1, -1, -1,
+             -1, -1, -1, -1, -1]], dtype=int64),
      (1, 1000))
 
 
@@ -295,7 +304,7 @@ show_predictions_from_batch_format(image, result)
 
 
 
-![png](models_export_files/models_export_18_0.png)
+![png](models_export_files/models_export_19_0.png)
 
 
 
@@ -411,7 +420,7 @@ show_predictions_from_flat_format(image, result)
 
 
 
-![png](models_export_files/models_export_24_0.png)
+![png](models_export_files/models_export_25_0.png)
 
 
 
@@ -447,7 +456,7 @@ show_predictions_from_flat_format(image, result)
 
 
 
-![png](models_export_files/models_export_26_0.png)
+![png](models_export_files/models_export_27_0.png)
 
 
 
@@ -481,7 +490,7 @@ show_predictions_from_flat_format(image, result)
 
 
 
-![png](models_export_files/models_export_28_0.png)
+![png](models_export_files/models_export_29_0.png)
 
 
 
@@ -522,12 +531,12 @@ result = session.run(outputs, {inputs[0]: image_bchw})
 show_predictions_from_flat_format(image, result)
 ```
 
-     25%|█████████████████████████████████████████████████                                                                                                                                                   | 4/16 [00:11<00:34,  2.90s/it]
+     25%|██████████████████████████████▊                                                                                            | 4/16 [00:11<00:33,  2.79s/it]
 
 
 
 
-![png](models_export_files/models_export_30_1.png)
+![png](models_export_files/models_export_31_1.png)
 
 
 

diff --git a/src/super_gradients/common/environment/device_utils.py b/src/super_gradients/common/environment/device_utils.py
@@ -19,10 +19,24 @@ def _get_assigned_rank() -> int:
 
 @dataclasses.dataclass
 class DeviceConfig:
-    device: str = "cuda" if torch.cuda.is_available() else "cpu"
+    _device: str = "cuda" if torch.cuda.is_available() else "cpu"
     multi_gpu: str = None
     assigned_rank: int = dataclasses.field(default=_get_assigned_rank(), init=False)
 
+    @property
+    def device(self) -> str:
+        return self._device
+
+    @device.setter
+    def device(self, value: str):
+        if "cuda" in value and not torch.cuda.is_available():
+            raise ValueError("CUDA is not available, cannot set device to cuda")
+        self._device = value
+
+    @property
+    def is_cuda(self):
+        return "cuda" in self._device
+
 
 # Singleton holding the device information
 device_config = DeviceConfig()
diff --git a/src/super_gradients/conversion/conversion_utils.py b/src/super_gradients/conversion/conversion_utils.py
@@ -12,7 +12,7 @@
     (torch.int16, np.int16),
     (torch.int8, np.int8),
     (torch.uint8, np.uint8),
-    (torch.bool, np.bool),
+    (torch.bool, bool),
 ]