-
Notifications
You must be signed in to change notification settings - Fork 519
Insights: pytorch/xla
Overview
Could not load contribution data
Please try again later
18 Pull requests merged by 9 people
-
Add tooling and documentation for setting up clangd
#9137 merged
May 9, 2025 -
9080 expose mat mul precision
#9081 merged
May 9, 2025 -
Add instructions on making VSCode discover the pytorch/xla repo.
#9134 merged
May 9, 2025 -
Unify style across pytorch/xla
#9124 merged
May 9, 2025 -
[torchax] Support mark_sharding_with_gradients
#9122 merged
May 9, 2025 -
Remove trailing whitespace from the repo
#9065 merged
May 8, 2025 -
Upgrade sccache to v0.10.0 in upstream docker image
#9102 merged
May 8, 2025 -
Fix typo
executation->execution#9109 merged
May 8, 2025 -
Fix nightly installation instruction
#9106 merged
May 7, 2025 -
Add instructions on creating a PR and brining forks up-to-date.
#9105 merged
May 7, 2025 -
Enrich instructions for setting up dev environment.
#9104 merged
May 6, 2025 -
Revert "Add
clanglink toPATH."#9099 merged
May 6, 2025 -
change all_to_all check to allow for split sizes > 1
#9100 merged
May 6, 2025 -
Add missing tensor data types (unsigned int 16, 32, 64) to PopulateTensorBuffer
#9090 merged
May 6, 2025 -
Create copy of nightly wheel without version number
#9091 merged
May 5, 2025 -
Add
clanglink toPATH.#9053 merged
May 5, 2025 -
Add tengyifei, bhavya01 and qihqi to infra owners
#9093 merged
May 5, 2025 -
Try to re-enable previously disabled CPU test
#9085 merged
May 5, 2025
16 Pull requests opened by 12 people
-
[not for review] update pin
#9087 opened
May 4, 2025 -
Add features needed for vllm
#9092 opened
May 5, 2025 -
Move requires_jax to inner flash_attention functions
#9098 opened
May 5, 2025 -
9082 educate users on mat mul precision
#9103 opened
May 6, 2025 -
Collect coverage data.
#9107 opened
May 7, 2025 -
Update numerical verification for SPMD Linear checkpointing
#9113 opened
May 7, 2025 -
[benchmarks] Fix run single config command error
#9115 opened
May 8, 2025 -
get master ip address for neuron device
#9120 opened
May 8, 2025 -
Lower clamp
#9123 opened
May 8, 2025 -
Yifeit/torchax 2 7
#9126 opened
May 9, 2025 -
remove version number from nightly build references
#9130 opened
May 9, 2025 -
Document the difference between tracing time and execution time
#9133 opened
May 9, 2025 -
Add assume_pure_torch implementation for forward pass only
#9135 opened
May 9, 2025 -
Throw a Python exception if compilation fails.
#9138 opened
May 10, 2025 -
Test torchax on Python 3.10 - 3.12
#9139 opened
May 10, 2025 -
Silence distributed warning
#9140 opened
May 10, 2025
23 Issues closed by 7 people
-
Expose mat_mul_precision
#9080 closed
May 9, 2025 -
[torch-xla 2.6] Training performance regression in torch-xla 2.6 for medium/small models
#9037 closed
May 9, 2025 -
Remove trailing whitespace from the repo
#9101 closed
May 8, 2025 -
Fix typo `executation` -> `execution`
#9108 closed
May 8, 2025 -
Conform codebase to formatting as outlined in `CONTRIBUTING.md`
#9110 closed
May 7, 2025 -
Fix the contribution instructions for creating PRs
#9056 closed
May 7, 2025 -
CUDA momery：how can i control xla reserved in total by PyTorch with GPU
#4662 closed
May 7, 2025 -
Cloud installation error of .whl file
#4663 closed
May 7, 2025 -
Unsupported unsigned (16, 32, 64) integer tensor types
#9074 closed
May 7, 2025 -
distributed.all_to_all_single fails to compile when input split size is > 1
#9068 closed
May 6, 2025 -
Create a nightly torch_xla wheel without version name
#8877 closed
May 6, 2025 -
Got error when build xla from source
#4666 closed
May 6, 2025 -
Zero copy tensor conversion between xla:gpu and torch.cuda
#4692 closed
May 6, 2025 -
Run Pytorch 2.0 benchmarks with XLA backend
#4699 closed
May 6, 2025 -
Increasing rendezvous timeout patience?
#4831 closed
May 5, 2025 -
Error while trying to run on TPU from VM instance.
#4896 closed
May 5, 2025 -
Undesired interaction between DeepSpeed and XLA
#4988 closed
May 5, 2025 -
Different Graph generations
#4994 closed
May 5, 2025 -
Collective operations not working when using multiple CPUs
#5005 closed
May 5, 2025 -
torch.distributed.reduce vs torch_xla.core.xla_model.all_reduce
#5022 closed
May 5, 2025 -
How to do multi-machine SPMD/FSDPv2 training with TPU？
#8492 closed
May 5, 2025 -
2 questions for the composite op feature
#8486 closed
May 5, 2025 -
Re-enable CPU test `test/test_python_ops.py -k TestPythonOps` for `uint8` dtype
#8799 closed
May 5, 2025
18 Issues opened by 11 people
-
Create automated runnable tutorial build system
#9136 opened
May 9, 2025 -
Implement `@assume_pure` with torch-fx and aotautograd in torch_xla
#9131 opened
May 9, 2025 -
set_mat_mul_precision is flakey
#9129 opened
May 9, 2025 -
SPMD Linear Model test failing with GA API refinement
#9128 opened
May 9, 2025 -
set up clangd for C++ development
#9127 opened
May 9, 2025 -
torchax/test/test_unbounded_dynamism.py fails under PyTorch 2.7
#9125 opened
May 9, 2025 -
The benchmarks does not work on CPU because of AMP
#9119 opened
May 8, 2025 -
Add installation instructions to `benchmarks/README.md`
#9118 opened
May 8, 2025 -
Get the number of graphs with dynamo for the models in benchmarks
#9117 opened
May 8, 2025 -
Rename and/or move test_utils.py
#9116 opened
May 8, 2025 -
install clangd in the dev container
#9114 opened
May 8, 2025 -
Enhance CI to only build & test affected targets for PRs
#9112 opened
May 7, 2025 -
Support Inplace (Sliced) Copy based on Runtime
#9097 opened
May 5, 2025 -
PyTorch/XLA shouldn't crash on XLA errors
#9096 opened
May 5, 2025 -
Support Dynamic Grid in Pallas Kernel
#9095 opened
May 5, 2025 -
tpu torch xla is not using xla_cache
#9094 opened
May 5, 2025 -
Error on collective operations when using XLA:CPU.
#9089 opened
May 5, 2025 -
Unnecessary & expensive copy ops prior to index_copy_ on SPMD sharded tensor
#9088 opened
May 5, 2025
34 Unresolved conversations
Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.
-
Replace implementation of `xla_torch.sync()` to `xm.mark_step()`
#9086 commented on
May 9, 2025 • 8 new comments -
Disable running slow CI for doc-only changes
#9072 commented on
May 10, 2025 • 2 new comments -
[Draft] Add Experimental limited sparse embedding bag
#8905 commented on
May 7, 2025 • 2 new comments -
Race condition on GPU device
#4541 commented on
May 7, 2025 • 0 new comments -
Support Python 3.12
#8703 commented on
May 8, 2025 • 0 new comments -
torch.linalg.lstsq issues on GPU/TPU
#8953 commented on
May 8, 2025 • 0 new comments -
TPU+GPU test flake: test_diagonal_write_transposed_r3 (__main__.TestAtenXlaTensor)
#8985 commented on
May 8, 2025 • 0 new comments -
Apr 28 pin update onwards: update how we download nightly JAX
#9064 commented on
May 9, 2025 • 0 new comments -
Support memory stat querying API with SPMD
#9022 commented on
May 9, 2025 • 0 new comments -
[scan] Avoid re-tracing the combine function on every call
#8632 commented on
May 10, 2025 • 0 new comments -
add torch_xla_graph_execution_check_level (default disabled) flag that emits warning(1) or throw error(2) during tensor sync and output the python frame
#9057 commented on
May 7, 2025 • 0 new comments -
Assert on empty PJRT buffers
#9062 commented on
May 6, 2025 • 0 new comments -
Replace `xm.mark_step` with `torch_xla.sync()` wherever possible
#9070 commented on
May 8, 2025 • 0 new comments -
Fix + Run `DynamicShapeDetector` tests on CI.
#9075 commented on
May 6, 2025 • 0 new comments -
[cherry-pick to r2.6_aws_neuron] add torch_xla_graph_execution_check_level (default disabled) flag that emits warning(1) or throw error(2) during tensor sync and output the python frame
#9077 commented on
May 6, 2025 • 0 new comments -
Refine the gradient accumulation API
#9078 commented on
May 9, 2025 • 0 new comments -
Check the scan op input for requires_grad
#9083 commented on
May 8, 2025 • 0 new comments -
Educate users on mat mul precision
#9082 commented on
May 5, 2025 • 0 new comments -
Libtpu pin update after 04/25 hangs
#9084 commented on
May 5, 2025 • 0 new comments -
Code coverage regreessed from 62% to 58% on May 3
#4981 commented on
May 5, 2025 • 0 new comments -
Support for torch.distributed.scatter in PyTorch XLA
#4940 commented on
May 5, 2025 • 0 new comments -
Lower Multinomial without Replacement
#4865 commented on
May 5, 2025 • 0 new comments -
Drop the use of tokens for ordering Collective-communication ops.
#4836 commented on
May 5, 2025 • 0 new comments -
Enable multiprocessing on pytorch XLA for TPU VM
#4893 commented on
May 5, 2025 • 0 new comments -
GSPMD + PyTorch Compile + TPU crash
#4824 commented on
May 5, 2025 • 0 new comments -
Document the difference between `device=` vs `.to(device)`
#8861 commented on
May 5, 2025 • 0 new comments -
Placeholder tracing segmentation faults
#9049 commented on
May 6, 2025 • 0 new comments -
Pytorch 2 compile + fsdp + transformers crash
#4823 commented on
May 6, 2025 • 0 new comments -
Lower `aten::repeat_interleave.Tensor`
#4821 commented on
May 6, 2025 • 0 new comments -
Colab TPU Wheel Pytorch XLA 1.13 Python 3.8
#4820 commented on
May 6, 2025 • 0 new comments -
Training Multiple Steps on TPU as in TF2.x
#4773 commented on
May 6, 2025 • 0 new comments -
Clean up IR Node
#4568 commented on
May 7, 2025 • 0 new comments -
Remove XlaNode::Clone(torch::lazy::OpList operands)
#4567 commented on
May 7, 2025 • 0 new comments -
CrossEntropyLoss fail to detect the negative index
#4552 commented on
May 7, 2025 • 0 new comments