OSSForks/pytorch
1
0
Fork 0
mirror of https://github.com/zebrajr/pytorch.git synced 2025-12-07 12:21:27 +01:00
Code Issues Actions 37 Packages Projects Releases Wiki Activity
60,673 Commits 8,304 Branches 150 Tags 1.5 GiB
33a49eeae7
Commit Graph

230 Commits

Author SHA1 Message Date
Animesh Jain
33a49eeae7 [benchmark] Flag to switch on activation checkpointing for HF models (#102557) 
Pull Request resolved: https://github.com/pytorch/pytorch/pull/102557
Approved by: https://github.com/ngimel, https://github.com/Chillee
 2023-05-30 23:46:14 +00:00
Horace He
e71ab21422 update triton pin (#101919) 
Pull Request resolved: https://github.com/pytorch/pytorch/pull/101919
Approved by: https://github.com/ngimel
 2023-05-30 17:16:05 +00:00
Animesh Jain
040d2cc969 [dynamo] Some torchrec_dlrm related fixes (#101953) 
Issue 1 of https://github.com/pytorch/pytorch/issues/101918

Pull Request resolved: https://github.com/pytorch/pytorch/pull/101953
Approved by: https://github.com/jansel
 2023-05-28 17:56:08 +00:00
Bin Bao
ee33bae5c7 Fix an issue where checking sameness throw an exception (#102279) 
Summary: currently the exception is caught by outside and marked as
infra_error

Pull Request resolved: https://github.com/pytorch/pytorch/pull/102279
Approved by: https://github.com/anijain2305
 2023-05-25 19:49:23 +00:00
Jason Ansel
5ba16011d7 Suppress profiler spam in dynamo benchmarks (#101942) 
Makes this stuff go away:
```
STAGE:2023-05-20 20:49:34 63580:63580 ActivityProfilerController.cpp:311] Completed Stage: Warm Up
STAGE:2023-05-20 20:49:34 63580:63580 ActivityProfilerController.cpp:317] Completed Stage: Collection
STAGE:2023-05-20 20:49:34 63580:63580 ActivityProfilerController.cpp:321] Completed Stage: Post Processing
```

Pull Request resolved: https://github.com/pytorch/pytorch/pull/101942
Approved by: https://github.com/shunting314, https://github.com/desertfire
 2023-05-22 18:32:31 +00:00
Edward Z. Yang
22ca1a1124 Partially fix shape mismatch in vision_maskrcnn (#101477) 
The bulk of the heavy lifting is happening in
https://github.com/pytorch/vision/pull/7592

Signed-off-by: Edward Z. Yang <ezyang@meta.com>

Pull Request resolved: https://github.com/pytorch/pytorch/pull/101477
Approved by: https://github.com/voznesenskym
 2023-05-21 05:20:08 +00:00
drisspg
6f13d6892a Add meta support for multinomial (#101324) 
# Summary
Found this when trying to compile the text gen loop of nanogpt here: b33289942b/torchbenchmark/models/nanogpt_generate/model.py (L322)

Pull Request resolved: https://github.com/pytorch/pytorch/pull/101324
Approved by: https://github.com/ngimel
 2023-05-19 00:04:26 +00:00
Animesh Jain
794cc3952e adding moco to CI (#101098) 
Fixes #ISSUE_NUMBER

Pull Request resolved: https://github.com/pytorch/pytorch/pull/101098
Approved by: https://github.com/desertfire
 2023-05-18 10:01:49 +00:00
chuanqiw
b315c9b5ab [CI] Enlarge memory for OOM models in inductor cpu HF accuracy test (#101395) 
Change the Inductor CPU HF accuracy test node from `linux.4xlarge` (32GB) to `linux.24xlarge` (192GB) to enlarge the node memory. Also add 3 HF models back to CI test.

Fixes #101390

Pull Request resolved: https://github.com/pytorch/pytorch/pull/101395
Approved by: https://github.com/EikanWang, https://github.com/desertfire, https://github.com/huydhn
 2023-05-18 09:23:30 +00:00
Jason Ansel
403ce1a1c9 Fix benchmark model names printouts with tqdm (#101627) 
With the TQDM changes in #100969 -- the models names ended up getting hidden from the benchmark printouts.  We would print the model name with no newline, then tqdm would print a `\r` and overwrite the name of the running model.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/101627
Approved by: https://github.com/ezyang
 2023-05-17 15:31:11 +00:00
PaliC
e0fc24cdc5 add retries to inductor benchmark suite (#101019) 
This pr accomplishes
1) Enables retries for downloading torchbenchmark and huggingface models in a similar method to how we do it for timm models right now.
2) creates a `_download_model` function for the hugging face and TIMM runners whose output I plan to use to preload the models somewhere if possible (please double check I'll be saving the right thing). Instead of retries, we plan to just add torchbench to a docker image as it is relatively small.

<!--
copilot:poem
-->
### <samp>🤖 Generated by Copilot at 3361a4c</samp>

> _We're the brave and bold coders of the `common.py` module_
> _We've made a handy function for downloading models_
> _We've shared it with our mates in the other runners_
> _So pull and push and try again, we'll get them all in time_

Pull Request resolved: https://github.com/pytorch/pytorch/pull/101019
Approved by: https://github.com/huydhn, https://github.com/desertfire
 2023-05-16 21:41:50 +00:00
Edward Z. Yang
41468833fb vision_maskrcnn is now deterministic (#101116) 
Signed-off-by: Edward Z. Yang <ezyang@meta.com>

Pull Request resolved: https://github.com/pytorch/pytorch/pull/101116
Approved by: https://github.com/ngimel
 2023-05-16 21:32:17 +00:00
Yanbo Liang
e4eaf33346 Re-enable detectron2_maskrcnn on CI (#100791) 
#99665 has been fixed, we can re-enable these models on CI.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/100791
Approved by: https://github.com/huydhn
 2023-05-16 04:25:58 +00:00
Edward Z. Yang
f48718f749 Update torchbench pin (#101365) 
Signed-off-by: Edward Z. Yang <ezyang@meta.com>

Pull Request resolved: https://github.com/pytorch/pytorch/pull/101365
Approved by: https://github.com/albanD, https://github.com/awgu
 2023-05-15 16:52:31 +00:00
Natalia Gimelshein
49578913fb update timm commit (#100931) 
Fixes #100903

Pull Request resolved: https://github.com/pytorch/pytorch/pull/100931
Approved by: https://github.com/ezyang, https://github.com/malfet
 2023-05-12 04:22:08 +00:00
Edward Z. Yang
41a4e22015 Update torchbench pin (#101071) 
Signed-off-by: Edward Z. Yang <ezyang@meta.com>

Pull Request resolved: https://github.com/pytorch/pytorch/pull/101071
Approved by: https://github.com/malfet
 2023-05-11 18:09:40 +00:00
Jason Ansel
036a8d6b4a Remove NullContext() from benchmark runners (#100309) 
Pull Request resolved: https://github.com/pytorch/pytorch/pull/100309
Approved by: https://github.com/Skylion007, https://github.com/anijain2305
 2023-05-11 06:42:27 +00:00
XiaobingSuper
c84627c2ee benchmarks: make --amp works for cpu path (#101057) 
Pull Request resolved: https://github.com/pytorch/pytorch/pull/101057
Approved by: https://github.com/jgong5, https://github.com/desertfire, https://github.com/jansel
 2023-05-11 02:51:38 +00:00
Edward Z. Yang
c658732950 [RFC] Add tqdm to benchmarking script (#100969) 
Here's what it looks like, on a slower running benchmark:

https://github.com/pytorch/pytorch/assets/13564/47c4a5bd-e963-45de-a15c-2fd943de0fa4

There's actually quite a bit of dead time, it's possible there are more spots we should add tqdm to. Looking for opinions on utility of this.

Signed-off-by: Edward Z. Yang <ezyang@meta.com>

Pull Request resolved: https://github.com/pytorch/pytorch/pull/100969
Approved by: https://github.com/Skylion007
 2023-05-10 15:39:24 +00:00
Bin Bao
76cc3ab4f3 [CI] Delete skips from https://github.com/pytorch/pytorch/issues/93847 (#96049) 
Pull Request resolved: https://github.com/pytorch/pytorch/pull/96049
Approved by: https://github.com/jansel
 2023-05-10 01:27:27 +00:00
Edward Z. Yang
9eab13fc90 Reenable llama benchmark (#100877) 
Signed-off-by: Edward Z. Yang <ezyang@meta.com>

Pull Request resolved: https://github.com/pytorch/pytorch/pull/100877
Approved by: https://github.com/albanD
 2023-05-09 01:12:54 +00:00
Natalia Gimelshein
9790f9174a skip lcnet (#100726) 
Per title

Pull Request resolved: https://github.com/pytorch/pytorch/pull/100726
Approved by: https://github.com/voznesenskym
 2023-05-05 23:19:42 +00:00
Animesh Jain
3f025c607c summarize graph breaks (#100696) 
Pull Request resolved: https://github.com/pytorch/pytorch/pull/100696
Approved by: https://github.com/yanboliang
 2023-05-05 22:27:47 +00:00
Animesh Jain
8994d9e610 [dynamo] Hide guard_fail_hook behind a flag to improve cache lookup time (+10% DebertaV2) (#100590) 
For TorchDynamo eager backend, DebertaV2 speedup improves from 0.77x to 0.87x.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/100590
Approved by: https://github.com/voznesenskym, https://github.com/wconstab
 2023-05-04 18:52:21 +00:00
Yanbo Liang
896eb1db26 [Dynamo] Skip TB Background_Matting model eager accuracy check because of non deterministic (#100513) 
Fixes #ISSUE_NUMBER

Pull Request resolved: https://github.com/pytorch/pytorch/pull/100513
Approved by: https://github.com/anijain2305
 2023-05-03 07:06:50 +00:00
Jason Ansel
fdc853b14c Add --baseline option to benchmark runners (#100266) 
Pull Request resolved: https://github.com/pytorch/pytorch/pull/100266
Approved by: https://github.com/ngimel
 2023-05-02 02:35:11 +00:00
Edward Z. Yang
e918fd18e7 Disable densenet121 as it is flaky (#100371) 
Signed-off-by: Edward Z. Yang <ezyang@meta.com>

Pull Request resolved: https://github.com/pytorch/pytorch/pull/100371
Approved by: https://github.com/voznesenskym
 2023-05-02 01:49:11 +00:00
Edward Z. Yang
5d93265cce Report timeout/infra_error instead of 0.0000 on infra error (#100372) 
Signed-off-by: Edward Z. Yang <ezyang@meta.com>

Pull Request resolved: https://github.com/pytorch/pytorch/pull/100372
Approved by: https://github.com/Skylion007, https://github.com/albanD
 2023-05-01 14:56:01 +00:00
Huy Do
9a69634b28 Skip some failing dynamic shape models on periodic (#99895) 
After some recent changes, these tests are failing in periodic trunk.  So let's move them to unstable while waiting for the team to root cause the issue https://github.com/pytorch/pytorch/issues/99893.  Note that a forward fix can use `ciflow/unstable` to run those unstable jobs to confirm that they are fixed.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/99895
Approved by: https://github.com/malfet
 2023-04-25 07:05:08 +00:00
Edward Z. Yang
04e8df4dd7 Return full accuracy status for printing, not abbreviated version (#99894) 
Signed-off-by: Edward Z. Yang <ezyang@meta.com>

Pull Request resolved: https://github.com/pytorch/pytorch/pull/99894
Approved by: https://github.com/jansel
 2023-04-25 05:17:10 +00:00
Edward Z. Yang
cd61707167 yolov3 dynamic training accuracy is fixed (#99896) 
Signed-off-by: Edward Z. Yang <ezyang@meta.com>

Pull Request resolved: https://github.com/pytorch/pytorch/pull/99896
Approved by: https://github.com/albanD
 2023-04-25 01:15:24 +00:00
chuanqiw
e9e5ffe83e Re-enable dynamic shapes test in dynamo benchmark (#99816) 
Set `torch._dynamo.config.assume_static_by_default = False` for dynamic shapes flag enabled

Fixes #99815

Pull Request resolved: https://github.com/pytorch/pytorch/pull/99816
Approved by: https://github.com/jgong5, https://github.com/ezyang
 2023-04-24 20:34:52 +00:00
Edward Z. Yang
f602b3a6ae Preserve mark_dynamic when cloning inputs (#99617) 
Signed-off-by: Edward Z. Yang <ezyang@meta.com>

Pull Request resolved: https://github.com/pytorch/pytorch/pull/99617
Approved by: https://github.com/ngimel, https://github.com/voznesenskym, https://github.com/anijain2305
 2023-04-22 19:46:31 +00:00
Bin Bao
e09f785a72 [CI] Remove inductor skip list for Huggingface (#99375) 
Pull Request resolved: https://github.com/pytorch/pytorch/pull/99375
Approved by: https://github.com/anijain2305
 2023-04-21 18:13:22 +00:00
Edward Z. Yang
fc8fa6c356 Require at least one tensor to be marked dynamic with --dynamic-batch-only (#99620) 
Signed-off-by: Edward Z. Yang <ezyang@meta.com>

Pull Request resolved: https://github.com/pytorch/pytorch/pull/99620
Approved by: https://github.com/voznesenskym
 2023-04-21 00:17:08 +00:00
Huy Do
5315317b7b Skip some detectron2_maskrcnn models with KeyError _ignore_torch_cuda_oom (#99599) 
These tests are failing in trunk 233cc34d3b with `KeyError: '_ignore_torch_cuda_oom'`

Pull Request resolved: https://github.com/pytorch/pytorch/pull/99599
Approved by: https://github.com/malfet
 2023-04-20 18:11:35 +00:00
Jason Ansel
3233450d07 Add TorchXLA option to benchmark runner (#99505) 
Pull Request resolved: https://github.com/pytorch/pytorch/pull/99505
Approved by: https://github.com/voznesenskym
 2023-04-19 22:44:52 +00:00
Will Constable
9ac2b041c9 Make opacus xfail instead of skip (#99380) 
Pull Request resolved: https://github.com/pytorch/pytorch/pull/99380
Approved by: https://github.com/desertfire, https://github.com/anijain2305
 2023-04-19 21:09:06 +00:00
Michael Voznesensky
113bd11cf4 Skip levit (#99491) 
Pull Request resolved: https://github.com/pytorch/pytorch/pull/99491
Approved by: https://github.com/ezyang
 2023-04-19 07:41:42 +00:00
Edward Z. Yang
039faf0dbf Add invariant that all symbolic shapes must be bound in graph (#99089) 
Previously, we had a problem when partitioning forward-backward dynamic graphs, which is that we could end up with a backward graph that mentions a symbol in an input tensor (e.g., `f32[s0 + s1]`), but without this symbol being otherwise bound elsewhere. When this happens, we have no way of actually deriving the values of `s0` and `s1`. Our fix for this in https://github.com/pytorch/pytorch/pull/93059 was to just retrace the graph, so that s0 + s1 got allocated a new symbol s2 and everything was happy. However, this strategy had other problems, namely (1) we lost all information from the previous ShapeEnv, including guards and (2) we end up allocating a LOT of fresh new symbols in backwards.

With this change, we preserve the same ShapeEnv between forward and backwards. How do we do this? We simply require that every symbol which may be present inside tensors, ALSO be a plain SymInt input to the graph. This invariant is enforced by Dynamo. Once we have done this, we can straightforwardly modify the partitioner to preserve these SymInt as saved for backwards, if they are needed in the backwards graph to preserve the invariant as well.

This apparently breaks yolov3, but since everything else is OK I'm merging this as obviously good and investigating later.

Signed-off-by: Edward Z. Yang <ezyang@meta.com>

Pull Request resolved: https://github.com/pytorch/pytorch/pull/99089
Approved by: https://github.com/voznesenskym
 2023-04-16 01:48:19 +00:00
Yanbo Liang
15fe5a0798 [Dynamo] Fix benchmark --verbose error (#99224) 
Dynamo benchmark --verbose is broken:
```
Traceback (most recent call last):
  File "/scratch/ybliang/work/repos/pytorch/benchmarks/dynamo/torchbench.py", line 400, in <module>
    torchbench_main()
  File "/scratch/ybliang/work/repos/pytorch/benchmarks/dynamo/torchbench.py", line 396, in torchbench_main
    main(TorchBenchmarkRunner(), original_dir)
  File "/scratch/ybliang/work/repos/pytorch/benchmarks/dynamo/common.py", line 1967, in main
    return maybe_fresh_cache(
  File "/scratch/ybliang/work/repos/pytorch/benchmarks/dynamo/common.py", line 993, in inner
    return fn(*args, **kwargs)
  File "/scratch/ybliang/work/repos/pytorch/benchmarks/dynamo/common.py", line 2135, in run
    torch._dynamo.config.log_level = logging.DEBUG
  File "/scratch/ybliang/work/repos/pytorch/torch/_dynamo/config_utils.py", line 67, in __setattr__
    raise AttributeError(f"{self.__name__}.{name} does not exist")
AttributeError: torch._dynamo.config.log_level does not exist
```

Pull Request resolved: https://github.com/pytorch/pytorch/pull/99224
Approved by: https://github.com/voznesenskym
 2023-04-15 20:18:50 +00:00
Bin Bao
34f681c13b [CI] Remove inductor skip list for timm_models (#98840) 
Summary: check against the expected csv file instead of skipping tests

Pull Request resolved: https://github.com/pytorch/pytorch/pull/98840
Approved by: https://github.com/ezyang
 2023-04-15 13:54:41 +00:00
Bin Bao
e5501a967e [inductor] Support IndexPutFallback in cpp_wrapper (#98972) 
Summary:
1) Make the fallback index_put generate the right cpp code in cpp_wapper
2) Add a --cpp-wrapper option to common.py

Pull Request resolved: https://github.com/pytorch/pytorch/pull/98972
Approved by: https://github.com/jgong5, https://github.com/jansel
 2023-04-13 15:41:03 +00:00
Edward Z. Yang
b8b840be3d Convert logging f-strings to use % format, part five (#98765) 
This does some annoying but simple cases by hand.

Signed-off-by: Edward Z. Yang <ezyang@meta.com>

Pull Request resolved: https://github.com/pytorch/pytorch/pull/98765
Approved by: https://github.com/wanchaol
 2023-04-11 13:17:59 +00:00
Edward Z. Yang
b09722f540 Convert logging f-strings to use % format, part two (#98700) 
This hits multi-line logging strings

Signed-off-by: Edward Z. Yang <ezyang@meta.com>

Pull Request resolved: https://github.com/pytorch/pytorch/pull/98700
Approved by: https://github.com/voznesenskym
 2023-04-10 12:19:31 +00:00
Edward Z. Yang
9a8f71f23e Convert logging f-strings to use % format (#98697) 
Codemod done with
https://gist.github.com/ezyang/2e8b0463cdc6be278478495b23ff0530 with
assistance from ChatGPT.

Signed-off-by: Edward Z. Yang <ezyang@meta.com>

Pull Request resolved: https://github.com/pytorch/pytorch/pull/98697
Approved by: https://github.com/voznesenskym
 2023-04-10 12:19:31 +00:00
Edward Z. Yang
bdb79a8f52 Turn off divisible_by_16 for dynamic shapes; support ablation (#98471) 
Signed-off-by: Edward Z. Yang <ezyang@meta.com>

Pull Request resolved: https://github.com/pytorch/pytorch/pull/98471
Approved by: https://github.com/ngimel, https://github.com/voznesenskym
 2023-04-06 12:57:07 +00:00
Edward Z. Yang
cf1bfca2ba Require batch dimensions to be compiled dynamically (#98334) 
Signed-off-by: Edward Z. Yang <ezyang@meta.com>

Pull Request resolved: https://github.com/pytorch/pytorch/pull/98334
Approved by: https://github.com/voznesenskym
 2023-04-05 19:40:22 +00:00
Edward Z. Yang
b923f84805 Switch accuracy CI to dynamic batch only (#98307) 
Signed-off-by: Edward Z. Yang <ezyang@meta.com>

Pull Request resolved: https://github.com/pytorch/pytorch/pull/98307
Approved by: https://github.com/wconstab
 2023-04-05 01:20:12 +00:00
Elias Ellison
a3365e1d0d Increment pending forwards after invocation (#98101) 
Forwards are only pending following invocation, not before.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/98101
Approved by: https://github.com/ngimel
 2023-04-05 00:04:39 +00:00