Fix a bug for large models in onnx importer. #3875

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Merged

zjgarvey merged 1 commit into llvm:main from zjgarvey:importer_data_dir

Nov 15, 2024

Collaborator

zjgarvey commented Nov 15, 2024

The method onnx.load_external_data_for_model function does not admit pathlib.Path as an input.


          fix a bug with the importer using Path instead of str

1f85936

zjgarvey requested a review from vivekkhandelwal1

November 15, 2024 04:01

vivekkhandelwal1 approved these changes

View reviewed changes

zjgarvey merged commit c26ca8b into llvm:main

3 checks passed

zjgarvey mentioned this pull request

Update opset version without weights for large nlp models nod-ai/SHARK-TestSuite#396

Merged

pdhirajkumarprasad pushed a commit to nod-ai/SHARK-TestSuite that referenced this pull request


          Update opset version without weights for large nlp models (#396)

c7894d0

We also need the importer change from
<llvm/torch-mlir#3875> for these to work. The
primary issue is that updating the opset version requires the model to
be smaller than 2GB. I've found a hacky way to update the opset version
without loading external data, and this is the result.

## Passing Summary

**TOTAL TESTS = 49**
|Stage|# Passing|% of Total|% of Attempted|
|--|--|--|--|
| Setup | 49 | 100.0% | 100.0% |
| IREE Compilation | 41 | 83.7% | 83.7% |
| Gold Inference | 41 | 83.7% | 100.0% |
| IREE Inference Invocation | 41 | 83.7% | 100.0% |
| Inference Comparison (PASS) | 31 | 63.3% | 75.6% |
## Fail Summary

**TOTAL TESTS = 49**
|Stage|# Failed at Stage|% of Total|
|--|--|--|
| Setup | 0 | 0.0% |
| IREE Compilation | 8 | 16.3% |
| Gold Inference | 0 | 0.0% |
| IREE Inference Invocation | 0 | 0.0% |
| Inference Comparison | 10 | 20.4% |
## Test Run Detail
Test was run with the following arguments:
Namespace(device='local-task', backend='llvm-cpu', target_chip='gfx942',
iree_compile_args=None, mode='cl-onnx-iree', torchtolinalg=False,
stages=None, skip_stages=None, benchmark=False, load_inputs=False,
groups='all', test_filter=None, testsfile='import.txt', tolerance=None,
verbose=True, rundirectory='test-run', no_artifacts=False, cleanup='3',
report=True, report_file='import.md', get_metadata=True)

| Test | Exit Status | Mean Benchmark Time (ms) | Notes |
|--|--|--|--|
| dm_nfnet_f3.dm_in1k | PASS | None | |
| dm_nfnet_f4.dm_in1k | PASS | None | |
| migraphx_sd__unet__model | compilation | None | |
| migraphx_sdxl__unet__model | compilation | None | |
| model--financial-summarization-pegasus--human-centered-summarization |
PASS | None | |
|
model--finetuned_gpt2-large_sst2_negation0.0001_pretrainedTrue_epochs1--jhaochenz
| PASS | None | |
|
model--finetuned_gpt2-large_sst2_negation0.001_pretrainedTrue_epochs1--jhaochenz
| PASS | None | |
|
model--finetuned_gpt2-large_sst2_negation0.001_pretrainedTrue_epochs2--jhaochenz
| PASS | None | |
|
model--finetuned_gpt2-large_sst2_negation0.001_pretrainedTrue_epochs3--jhaochenz
| PASS | None | |
| model--finetuned_gpt2-large_sst2_negation0.01--yuhuizhang | PASS |
None | |
|
model--finetuned_gpt2-large_sst2_negation0.01_pretrainedFalse_epochs10--jhaochenz
| PASS | None | |
|
model--finetuned_gpt2-large_sst2_negation0.01_pretrainedTrue_epochs1--jhaochenz
| PASS | None | |
| model--finetuned_gpt2-large_sst2_negation0.05--yuhuizhang | PASS |
None | |
|
model--finetuned_gpt2-large_sst2_negation0.0_pretrainedFalse--yuhuizhang
| PASS | None | |
|
model--finetuned_gpt2-large_sst2_negation0.1_pretrainedFalse_epochs10--jhaochenz
| PASS | None | |
|
model--finetuned_gpt2-large_sst2_negation0.1_pretrainedTrue_epochs1--jhaochenz
| PASS | None | |
| model--finetuned_gpt2-large_sst2_negation0.2--yuhuizhang | PASS | None
| |
| model--finetuned_gpt2-large_sst2_negation0.5--yuhuizhang | PASS | None
| |
|
model--finetuned_gpt2-large_sst2_negation0.5_pretrainedFalse--yuhuizhang
| PASS | None | |
| model--finetuned_gpt2-large_sst2_negation0.8--yuhuizhang | PASS | None
| |
|
model--finetuned_gpt2-large_sst2_negation0.8_pretrainedFalse--yuhuizhang
| PASS | None | |
| model--flan-t5-large-samsum--oguuzhansahin | Numerics | None | |
| model--long-t5-tglobal-large-pubmed-3k-booksum-16384-WIP--pszemraj |
compilation | None | |
| model--long-t5-tglobal-large-pubmed-3k-booksum-16384-WIP13--pszemraj |
compilation | None | |
| model--long-t5-tglobal-large-pubmed-3k-booksum-16384-WIP14--pszemraj |
compilation | None | |
| model--long-t5-tglobal-large-pubmed-3k-booksum-16384-WIP15--pszemraj |
compilation | None | |
| model--long-t5-tglobal-large-pubmed-3k-booksum-16384-WIP17--pszemraj |
compilation | None | |
| model--m2m100_418M-finetuned-kde4-en-to-pt_BR--danhsf | PASS | None |
|
| model--m2m100_418M-fr--Jour | PASS | None | |
| model--m2m100_418M-fr--NDugar | PASS | None | |
| model--m2m100_418M-ja--vivek-307306 | Numerics | None | |
|
model--manifestoberta-xlm-roberta-56policy-topics-sentence-2023-1-1--manifesto-project
| Numerics | None | |
| model--mT5-base-HunSum-1--SZTAKI-HLT | Numerics | None | |
| model--mT5_multilingual_XLSum--csebuetnlp | Numerics | None | |
| model--my_xlm-roberta-large-finetuned-conll03--BahAdoR0101 | Numerics
| None | |
| model--pegasus-cnn_dailymail--google | PASS | None | |
| model--pegasus-large-book-summary--pszemraj | PASS | None | |
| model--pegasus-large-booksum--cnicu | PASS | None | |
| model--pegasus-large-summary-explain--pszemraj | PASS | None | |
| model--pegasus-xsum--google | PASS | None | |
| model--pegasus_summarizer--tuner007 | PASS | None | |
| model--roberta-ner-multilingual--julian-schelb | Numerics | None | |
| model--t5-large-finetuned-xsum-cnn--sysresearch101 | Numerics | None |
|
| model--tglobal-large-booksum-WIP4-r1--pszemraj | compilation | None |
|
| model--xlm-roberta-large-squad2--deepset | Numerics | None | |
| model--xlmr-large-qa-fa--m3hrdadfi | Numerics | None | |
| model--YuisekinAI-mistral-0.7B--yuiseki | PASS | None | |
| vit_large_r50_s32_224.augreg_in21k_ft_in1k | PASS | None | |
| vit_large_r50_s32_384.augreg_in21k_ft_in1k | PASS | None | |

rahuls-cerebras added a commit that referenced this pull request


          Revert "Fix a bug for large models in onnx importer. (#3875)"

1b3bfd3

This reverts commit c26ca8b.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet