- 22 10月, 2022 1 次提交
-
-
由 Antonio Sanchez 提交于
If resulting dimensions overflow an int32, we were seeing an overflow and crash due to size mismatch during broadcast assignment. The cause is a simple dimension type mismatch. Note that actual tests for this are currently impractical, since successful operations require more than 2^32 elements and OOM on most machines. PiperOrigin-RevId: 479336566
-
- 21 10月, 2022 4 次提交
-
-
由 Mihai Maruseac 提交于
r2.10 cherry-pick: 1be74370 "Resolve a sanitizer issue with invalid char -> bool conversion."
-
由 Mihai Maruseac 提交于
r2.10 cherry-pick: a65411a1 "Fix OOB write in grappler."
-
由 Mihai Maruseac 提交于
Discovered via internal fuzzing. PiperOrigin-RevId: 482097391
-
由 Mihai Maruseac 提交于
When printing a tensor, we get it's data as a `const char*` array (since that's the underlying storage) and then we typecast it to the element type. However, conversions from `char` to `bool` are undefined if the `char` is not `0` or `1`, so sanitizers/fuzzers will crash. To fix, we're creating a mutable `char` array to convert the chars to 0/1, according to bool rules. Discovered via internal fuzzing. PiperOrigin-RevId: 482297563
-
- 13 10月, 2022 3 次提交
-
-
由 learning-to-play 提交于
r2.10 cherry-pick: 717ca98d "Adding missing requirement on inputs for MirrorPadGrad op and updating arithmetic to account for int32 padding values."
-
由 Jeremy Meredith 提交于
Adding missing requirement on inputs for MirrorPadGrad op and updating arithmetic to account for int32 padding values. PiperOrigin-RevId: 480691952
-
由 Mihai Maruseac 提交于
Cherry-pick Tensor.numpy() segfault fixes on 2.10
-
- 12 10月, 2022 4 次提交
-
-
由 learning-to-play 提交于
Merge pull request #58056 from vinila21/cherrypick-f5381e0e-on-r2.10 Fix OOB error when op input sizes do not match.
-
由 learning-to-play 提交于
Merge pull request #58060 from vinila21/cherrypick-b389f5c9-on-r2.10 Add true_classes input validation for candidate sampler ops.
-
由 Antonio Sanchez 提交于
The values must be within the valid range of the sampler. Added a check for this. PiperOrigin-RevId: 479441496
-
由 Antonio Sanchez 提交于
In cases where op input sizes are specified as in ``` REGISTER_OP("DynamicStitch") .Input("indices: N * int32") .Input("data: N * T") .Output("merged: T") .Attr("N : int >= 1") .Attr("T : type") .SetShapeFn(DynamicStitchShapeFunction); ``` if differing number of inputs are provided (e.g. 3 for `indices` and 4 for `data`) we can get a crash in the executor when parsing the inputs, even before the kernel called. Here we avoid this by checking the return code for the argument id and exit early. PiperOrigin-RevId: 478068540
-
- 08 10月, 2022 2 次提交
-
-
由 learning-to-play 提交于
Merge pull request #58015 from tensorflow/cherrypick-c676a2d7-on-r2.10 Add a check that strided slice op strides argument has reasonable size
-
由 Eugene Zhulenev 提交于
PiperOrigin-RevId: 478036251
-
- 07 10月, 2022 5 次提交
-
-
由 learning-to-play 提交于
Merge pull request #57952 from tensorflow/cherrypick-1ecf591f-on-r2.10 Fix std::bad_alloc exception (followed by crash) when loading a SavedModel containing a tensor of unknown rank.
-
由 Reed Wanderman-Milne 提交于
If a numpy array is created with a shape such that one element is zero and the others sum to a large number, an error will be raised. E.g. the following raises an error: np.ones((0, 2**31, 2**31)) TF previously would not check the return value of PyArray_SimpleNewFromData, which returns null on such shapes. Now the return value is checked. PiperOrigin-RevId: 477751929
-
由 Reed Wanderman-Milne 提交于
Numpy arrays can have at most 32 dimensions. When a tensor is converted to a Numpy array, PyArray_SimpleNewFromData is called, which returns nullptr if given more than 32 dimensions. Before, TF would dereference this nullptr. This change adds a check for the number of dimensions, and raises a proper error instead if there are more than 32 dimensions. PiperOrigin-RevId: 469580660
-
由 Alan Liu 提交于
Note: This fix will have to be cherry picked in r2.10, r2.9, and r2.8. PiperOrigin-RevId: 479125772
-
由 Alan Liu 提交于
Note: This fix will have to be cherry picked in r2.10, r2.9, and r2.8. PiperOrigin-RevId: 479141644
-
- 05 10月, 2022 2 次提交
-
-
由 learning-to-play 提交于
Merge pull request #57974 from nitins17/cherrypick-72210c84-on-r2.10 Use setup.py for building the installer wheels for collaborator builds
-
由 Nitin Srinivasan 提交于
PiperOrigin-RevId: 478593715
-
- 04 10月, 2022 1 次提交
-
-
由 Mathieu Guillame-Bert 提交于
Fix std::bad_alloc exception (followed by crash) when loading a SavedModel containing a tensor of unknown rank. PiperOrigin-RevId: 475317100
-
- 30 9月, 2022 1 次提交
-
-
由 Mihai Maruseac 提交于
Merge pull request #57811 from tensorflow/cherrypick-02400ea2-on-r2.10 Fix tf.math.sobol_sample vulnerability with non-scalar inputs.
-
- 29 9月, 2022 2 次提交
-
-
由 Alan Liu 提交于
Check that the element_shape input is valid. Add graph/eager unit tests. Note: This fix will have to be cherry picked in r2.10, r2.9, and r2.8. PiperOrigin-RevId: 477544091
-
由 Alan Liu 提交于
Check that the size input is valid. Add graph/eager unit tests. Graph mode was already ok but eager mode was not. Note: This fix will have to be cherry picked in r2.10, r2.9, and r2.8. PiperOrigin-RevId: 477002316
-
- 28 9月, 2022 3 次提交
-
-
由 Alan Liu 提交于
Check that the num_elements input is valid. Add graph/eager unit tests. Graph mode was already ok but eager mode was not. Note: This fix will have to be cherry picked in r2.10, r2.9, and r2.8. PiperOrigin-RevId: 477225661
-
由 TensorFlow Jenkins 提交于
Check that the size input is valid. Add graph/eager unit tests. Graph mode was already ok but eager mode was not. Note: This fix will have to be cherry picked in r2.10, r2.9, and r2.8. PiperOrigin-RevId: 477002316 Co-authored-by: NAlan Liu <liualan@google.com>
-
由 TensorFlow Jenkins 提交于
If a list of quantized tensors is assigned to an attribute, the pywrap code was failing to parse the tensor and returning a `nullptr`, which wasn't caught. Here we check the return value and set an appropriate error status. PiperOrigin-RevId: 476981029 Co-authored-by: NAntonio Sanchez <cantonios@google.com>
-
- 23 9月, 2022 1 次提交
-
-
由 Alan Liu 提交于
Check that given inputs are valid. Add graph/eager unit tests. Graph mode was already ok but eager mode was not. PiperOrigin-RevId: 476117492
-
- 03 9月, 2022 4 次提交
-
-
由 Mihai Maruseac 提交于
Update estimator and keras version in TF 2.10 branch for 2.10.0.
-
由 Vinila S 提交于
-
由 Mihai Maruseac 提交于
Update version numbers for TensorFlow 2.10.0
-
-
- 30 8月, 2022 2 次提交
-
-
由 Mihai Maruseac 提交于
Update release notes with security updates
-
由 Vinila S 提交于
Update release notes with security updates
-
- 26 8月, 2022 5 次提交
-
-
由 Mihai Maruseac 提交于
r2.10 cherry-pick: b5f6fbfb "Re-enable testTensorListReserveWithNonScalarNumElements to work with mlir as well."
-
由 Laura Pak 提交于
PiperOrigin-RevId: 466460987
-
由 Mihai Maruseac 提交于
r2.10 cherry-pick: b306fb31 "Apply `PyArray_Check` in `FillStringBufferWithPyArray` to avoid a hard crash when user pass in a TF eager tensor as input."
-