Cherry-pick 2.1 release branch into XRT branch through 9/14 #5574
will-cromar
merged
117 commits into
xrt
from
wcromar/xrt-cherry-picks-9-14
Sharding should be per output of IR Node, instead of per IR Node (#5330)
d638f6e4
Update Python device API for SPMD (#5129)
ee2e6cf1
Check out the release branch instead of origin/master in ansible (#5344)
6ef52a87
Also dump output sharding on HLO file (#5339)
9476d08b
Make all-reduce a no-op when world size is 1 (#5342)
b23a3337
add fs linker flag (#5347)
f203e938
Add py3.10 whl path to doc, refactor whl table (#5354)
10cea648
fix amp dtype setting for GPU (#5337)
5257b5fd
Add python test for SPMD+Runtime Python API (#5349)
b07770d5
Check the actual device instead of query env var for virtual device (…
ab01165e
[BE] use self.assertEquals instead of str equality in test_zero1.py (…
bde253d0
Revert "[BE] use self.assertEquals instead of str equality in test_ze…
f172dcd5
[Dynamo|TPU] Tweak `atol` and `rtol` for `test_dynamo.py` (#5363)
55491630
[Dynamo|TPU] Skip`DynamoTrainingBasicTest.test_resnet18` on TPU (#5362)
c49daaf5
Add a script for running stablehlo tests. (#5360)
e7ae1d46
Don't rewrite index hints in global save planning (#5348)
f993041e
[Dynamo|TPU] Skip `DynamoInferenceBasicTest.test_resnet18` on TPU (#5…
c7d7b234
[BE] use self.assertEquals instead of str equality in test_zero1.py (…
3fec0494
Fix ReplicateShardedData for int type (#5374)
7df484cf
Update dynamo.md (#5378)
52ab1c69
Revert "Fix ReplicateShardedData for int type (#5374)" (#5380)
9ae2efe5
Remove the mention of XRT_TPU_CONFIG in the CONTRIBUTING.md (#5379)
84a6635d
[Dynamo|TPU] Tweak `atol` and `rtol` for `test_simple_model_with_diff…
70b09d55
Rectify test_zero1.py once optim.load_state_dict doesn't guarantee im…
d2f82215
Add gpu doc for how to build PyTorch/XLA from source with GPU support…
3ff13bf2
clear pending ir should also clear the cc op tokens (#5385)
6c2d7af5
Port resnet data loading optimizations to SPMD test script (#5386)
87d397d0
Add support for in-place ops with self tensors in dynamo bridge (#5309)
59124a8d
Add dynamo test in TPU CI (#5381)
ab682142
Add manual seed in multihost checkpoint (#5392)
b13d1f2f
Fix change_id type in coverage uploading (#5394)
48f7f551
Update dynamo cpu fallback op to aten::_foobar (#5393)
7a80658a
Run single host multi GPU tests in the CI. (#5387)
a4a742de
[PJRT] Separate collective ops test from TPU runtime test. (#5396)
1d99226e
Fix ReplicateShardedData for int type (#5404)
33500a50
Update the dynamo backend name to `openxla` (#5402)
009f6747
[SPMD] Multi-host batch sharded data loading (#5331)
f786ddfa
Refactor to share code between export_torch_model and save_as_stableh…
2b2251f9
Fix TPU collective ops test for multi-host TPUs (#5408)
26391c17
Partially replicate lower-rank tensors (#5409)
88a68651
Revert "Partially replicate lower-rank tensors (#5409)" (#5412)
239119ff
SPMD cross slice-replication using partial_replication sharding (#5411)
fa13eb35
Fix the incorect clone arg condition in dynamo bridge (#5414)
3ffa11e9
[SPMD] named partition spec support (#5415)
ca3c0c45
[PJRT|TPU] Update `test_xla_devices_single_process_all_chips` for exp…
7c831ee6
Add repo for libcudnn8=8.7.0.84 and CUDA 11.8 (#5425)
16bc6283
Update fix_includes.sh (#5441)
7bfff692
[PJRT] Support `torchrun` with `pjrt://` `init_method` (#5438)
85150e53
Bugfix + add more test for llama (#5439)
b7bb55b3
Move the C++ test build to CI build job instead of test job (#5442)
6cf0446d
Update gcc to 10. (#5445)
60f5b0a9
Update the random seed for every dynamo execution (#5444)
55adfd10
Revert "Update gcc to 10. (#5445)" (#5449)
c664b9fa
Install gcc-10 (#5450)
143dd079
Revert "Install gcc-10 (#5450)" (#5452)
d1377979
parallelize SPMD inputhandler and GetDataShards (#5447)
f4890f22
Remove base image override from TPU CI build (#5453)
023d3adb
Update to GCC 10 (#5451)
22595dc1
Cache sharded placeholder for dynamo execution (#5446)
a1ab79c4
Remove Docker image override from dev image (#5456)
f8ebe054
hack: implement (unimplement?) GetDataShard for XRT
89a27e6e
skip flaky test (#5459)
8e60236d
Neuron import hook (#5429)
3bc6d427
Add missing includes (#5434)
c8650949
[GPU]Update README.md with wheel/docker for CUDA12.0 and deprecate CU…
bacc9de5
update remote cache key in ansible (#5463)
4ec5835a
Fix data type in Pow with Scalar base and Tensor exponent (#5467)
edc3d615
bump the timeout for CI (#5470)
0136410c
Fix the input sharding for dynamo (#5469)
455a6e17
Enabling sharding device data IR (#5475)
0d741e3a
Introduce `torch_xla.runtime.use_spmd()` (#5474)
a1ea65f8
Enable PJRT C API Client and other changes for Neuron (#5428)
e174a2a7
Don't move full tensor to device in deferred_init (#4819)
b7756b63
[SPMD] Fix HybridMesh ordering (#5478)
5b6c284e
[SPMD] Properly skip tests on TPU V2 (#5479)
b7e9cb88
Add @yeounoh to .github CODEOWNERS (#5482)
b42b47d9
Add Python API to execute StableHLO bytecode (#5476)
c3ff9de1
[SPMD] Fix TPU CI after #5478 (#5487)
8e4b5434
[SPMD] Fix XLA_DUMP_POST_OPTIMIZATIONS test (#5485)
2bb5ff2c
[Dist] Refactor ZeRO-1 (#5145)
9879617c
Update artifacts.auto.tfvars for 2.1 release (#5483)
bc098180
Add ShardingSpec to XLATensor when it is created with a PJRTShardedDa…
a8f2a266
Add topological sorting to dynamo partitions (#5472)
72f69d8d
[SPMD] Patch nn.Linear (#5491)
42cadd25
[original author: mrnikwaws] Neuron operator support (#5471)
6378fca6
[SPMD] Make IR sharding custom sharding op (#5433)
80975f55
Support input sharding changed after first dynamo tracing (#5477)
b559584e
Always use ExecuteReplicated with SPMD (#5494)
b9c69550
Skip a couple tests on TPU due to precision issue (#5496)
673fa1e4
Refactor stablehlo API and put them in official location. (#5493)
717f1d48
Support tuples in partition spec (#5488)
3aec1959
Add a API to explictly init runtime (#5500)
4543cb09
Add explict error message when tensor is on CPU for dynamo backend (#…
4964915d
remove torchvision in stablehlo.py (#5501)
25bda8db
Fix tupled partition spec test on v3 (#5503)
fd956753
Update dynamo doc (#5506)
0495761b
Update dynamo.md (#5509)
08a3bbc7
Support tuples in partition spec (#5488)
3aec1959
Add a API to explictly init runtime (#5500)
4543cb09
Add explict error message when tensor is on CPU for dynamo backend (#…
4964915d
remove torchvision in stablehlo.py (#5501)
25bda8db
Fix tupled partition spec test on v3 (#5503)
fd956753
Get original_traced_args as example_inputs. (#5511)
a85b860c
mark_sharding over a replicated tensor is allowed. (#5513)
ff94c942
Disable cxx abi in ansible when building pt/xla for branch r2.0 (#5332)
73873f1e
Update pytorch git tag for r2.1 (#5529)
9c623571
Enable megacore_dense by default (#5520) (#5531)
2d7e92f2
Add option to unbundle libtpu (#5534) (#5536)
ff30d64c
Revert 2.1 terraform changes (#5537)
19d74d4c
Fix FSDP for Models with Frozen Weights (#5484) (#5539)
7e225099
Update r2.1 wheel to be compatible with PyPI (#5550)
cdcc3d8a
Add resnet50-weight-quant colab notebook (#5407) (#5556)
565a915b
hack: add placeholders for `HasSharding` and `GetSharding` to XRT
60c9ad02
formatting
e324b7be
hack: always return false from `HasSharding`
5f1e8b38
Update torch pin to current RC for CI testing
41f8276c
Cherry pick `pjrt://` init method rename and doc updates (#5562)
515cf17a
will-cromar
changed the title
Use new cache silo and skip test build
8416c066
hack: disable missing test
da2dd820
hack: alter cache silo name
586c0d01
formatting
ebb3733a
JackCaoG
approved these changes
on 2023-09-15
Reviewers
Assignees
No one assigned
Labels
None yet
Milestone
No milestone
Login to write a write a comment.
Login via GitHub