[pull] master from tensorflow:master #233
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
![pull[bot]](https://avatars.githubusercontent.com/in/12910?s=60&v=4){kind=small}
PiperOrigin-RevId: 623932847
…orks PiperOrigin-RevId: 623935359
PiperOrigin-RevId: 623935433
PiperOrigin-RevId: 623935805
PiperOrigin-RevId: 623939572
…eSelector. PiperOrigin-RevId: 623946563
…ation layout shape. PiperOrigin-RevId: 623947612
PiperOrigin-RevId: 623955329
The `GFile` documentation does not mention what exceptions it raises: https://github.com/tensorflow/tensorflow/blob/38b17d708344a91234dad879794735e79f9af42a/tensorflow/python/platform/gfile.py#L37. It would be safer to explicitly check for file existence and return the appropriate result. PiperOrigin-RevId: 623961044
Breaks jax tests PiperOrigin-RevId: 623961307
Updates LLVM usage to match [ffb5bea2be9f](llvm/llvm-project@ffb5bea2be9f) PiperOrigin-RevId: 623964285
PiperOrigin-RevId: 623975602
PiperOrigin-RevId: 623975807
PiperOrigin-RevId: 624001836
This is required for `index_flat_map` where running the captured function may return a `Status`. PiperOrigin-RevId: 624026405
More changes will follow to demonstrate selective quantization. PiperOrigin-RevId: 624026901
Bumps [idna](https://github.com/kjd/idna) from 3.6 to 3.7. - [Release notes](https://github.com/kjd/idna/releases) - [Changelog](https://github.com/kjd/idna/blob/master/HISTORY.rst) - [Commits](kjd/idna@v3.6...v3.7) --- updated-dependencies: - dependency-name: idna dependency-type: direct:production ... Signed-off-by: dependabot[bot] <[email protected]>
PiperOrigin-RevId: 624045436
PiperOrigin-RevId: 624065316
Imported from GitHub PR openxla/xla#11444 1. cuDNN attention mask is not doing masking with -inf but multiply which is not correct. Hence disable patterns with mask. 2. Follow up PR to clean up the remaining mask related logic. Copybara import of the project: -- acf95b6cc7e1084026eaf87c0119ba3801ba8f8c by cjkkkk <[email protected]>: disable mask Merging this change closes #11444 PiperOrigin-RevId: 624068883
PiperOrigin-RevId: 624073633
When looking for ptxas and other NVIDIA binary tools we go through a list of potential locations until we find a suitable version of ptxas. Changing this list or its order has been deemed difficult in the past because there are many users out there that rely on the fact that this list happens to choose the right binary for them. So this is making a careful change that is only affecting Google environments and makes no change to any OSS build of TF, JAX or XLA. PiperOrigin-RevId: 624078399
PiperOrigin-RevId: 624088621
PiperOrigin-RevId: 624088926
PiperOrigin-RevId: 624091492
Reverts 5b63270 PiperOrigin-RevId: 624092399
Imported from GitHub PR openxla/xla#11053 This PR supports more patterns from TF->XLA GELU Approximate. It was found that the variations were resulting from associativity of multiply instructions like in the expression a * b * c. An utility function is added in this PR that captures patterns from two consecutive multiplies. Copybara import of the project: -- 74a287c7784138da3905d618d65a8301c821d7f6 by mdfaijul <[email protected]>: Patterns from tf.nn.gelu(approximate=True) -- 51b4962709de6cee8ec92c9fc225b5fc3371c276 by mdfaijul <[email protected]>: Fix typos Merging this change closes #11053 PiperOrigin-RevId: 624093341
PiperOrigin-RevId: 624095085
PiperOrigin-RevId: 624095561
…test Imported from GitHub PR openxla/xla#11352 Modified the test case to pass on rocm and cuda. There will be at least one more XLA related PR - to switch on Triton usage for ROCm, but only after PR for adding Triton build files for ROCm on openxla/triton. Copybara import of the project: -- dd80a6f44295642efd3ae6af6ffd3e2a3302d36e by Zoran Jovanovic <[email protected]>: [ROCm] Triton in XLA for ROCm - changed ir_emitter_triton_test to pass on rocm and cuda Merging this change closes #11352 PiperOrigin-RevId: 625361694
…l_fix PiperOrigin-RevId: 625362770
Proto messages for IFRT types may be used for various goals other than transferring them over wire. This change reflects it on the proto comments. Also set up a linter directive for `DType::Kind` and `DTypeProto::Kind` whose mirroring is easy to miss. PiperOrigin-RevId: 625365783
…utures to communicate events completion PiperOrigin-RevId: 625375855
…e sharding on a tuple op, by using the same sharding for all get-tuple-elements. PiperOrigin-RevId: 625384767
PiperOrigin-RevId: 625391182
PiperOrigin-RevId: 625394661
PiperOrigin-RevId: 625409989
…ng two conditions are true. 1. converting `mhlo::ReturnOp` or `mlir::func::ReturnOp` 2. `num_return_values == 1` 3. `!return_tuple || !is_entry_function` PiperOrigin-RevId: 625411731
PiperOrigin-RevId: 625413358
PiperOrigin-RevId: 625419138
…>= 8.9.4 Imported from GitHub PR openxla/xla#11478 * Build XLA with cuDNN 8.6 will cause compilation error because flash attention graph will use `CudnnfMHAUid` which is only defined with cuDNN > 8.8. * Add a guard for flash attention graph with at least cuDNN 8.9.4. So the current logic would be only compile FMHA with cuDNN > 8.8 and only compile flash attention with cuDNN > 8.9.4. Copybara import of the project: -- a1aa585f4e6ce42c7486336549447151cd5f7690 by cjkkkk <[email protected]>: add guards for flash attention graph with at least 8.9.4 Merging this change closes #11478 PiperOrigin-RevId: 625425491
PiperOrigin-RevId: 625426863
Imported from GitHub PR openxla/xla#11353 This implements async dynamic-slice and dynamic-update-slice for host memory offloading on GPU. Since the emitter does not understand dynamic slicing instructions in async computation, we wrap them in a fusion node and mark them for execution on a different stream. This is all we need to execute the offloading of slices asynchronously. Copybara import of the project: -- 905db6df8130300014728bf4eceb2ee2fc23d6e1 by Jaroslav Sevcik <[email protected]>: Wrap async dynamic slicing into fusion Merging this change closes #11353 PiperOrigin-RevId: 625428675
Otherwise, I'm getting this error:
```
Tensor("Const:0", shape=(4, 3), dtype=int64) must be from the same graph
as Tensor("TensorSliceDataset:0", shape=(), dtype=variant)
```
PiperOrigin-RevId: 625429037
This workflow is no longer relevant. PiperOrigin-RevId: 625430005
…es are not divisible by the mesh dimensions. Also get rid of a duplicate debug print statement PiperOrigin-RevId: 625454917
PiperOrigin-RevId: 625455577
Updates LLVM usage to match [694c444b5bbb](llvm/llvm-project@694c444b5bbb) PiperOrigin-RevId: 625464884
Reverts 94350b8 PiperOrigin-RevId: 625469451
This will allow tf_runtime to depend on XLA in a followup change PiperOrigin-RevId: 625474344
Without this check, it could crash if `index_map_fn` returns an out-of-bound index due to: https://github.com/tensorflow/tensorflow/blob/69a908420c5c5b90027f23905cd842c76ca3955c/tensorflow/core/framework/tensor.cc#L1104 PiperOrigin-RevId: 625476301
This commit adds folders for the ops `tfl.maximum` and `tfl.minimum` for the case where one of the arguments is `-FLT_MAX` and `FLT_MAX`, respectively. PiperOrigin-RevId: 625494071
PiperOrigin-RevId: 625502609
…k with async operations PiperOrigin-RevId: 625518793
…tore the `IfrtRestoreTensorRegistry` in IfrtServingExecutable for looking up the dtype and shape. PiperOrigin-RevId: 625527802
PiperOrigin-RevId: 625548761
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
See Commits and Changes for more details.
Created by
pull[bot]
Can you help keep this open source service alive? 💖 Please sponsor : )