[Flava] Add ckpt loading and accuracy metric to finetuning #119
Conversation
This file contains hidden or 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
[ghstack-poisoned]
facebook-github-bot
added
the
CLA Signed
[ghstack-poisoned]
Codecov Report
@@ Coverage Diff @@
## gh/ankitade/7/base #119 +/- ##
=====================================================
Coverage ? 88.59%
=====================================================
Files ? 40
Lines ? 2087
Branches ? 0
=====================================================
Hits ? 1849
Misses ? 238
Partials ? 0 Continue to review full report at Codecov.
|
|
@ankitade has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator. |
- Accuracy metric for finetuning - Add checkpoint saving and best ckpt loading based on val accuracy - Load pretrained ckpt by default in classification model - make num gpus 1 in qnli.yaml Test plan python -m flava.finetune config=flava/configs/finetuning/qnli.yaml (val acc : 0.8651) Loaded model weights from checkpoint at /data/home/deankita/torchmultimodal/examples/flava-epoch=03-step=10000.ckpt /data/home/deankita/miniconda/envs/flava/lib/python3.8/site-packages/pytorch_lightning/trainer/connectors/data_connector.py:330: PossibleUserWarning: Using `DistributedSampler` with the dataloaders. During `trainer.validate()`, it is recommended to use `Trainer(devices=1)` to ensure each sample/batch gets evaluated exactly once. Otherwise, multi-device settings use `DistributedSampler` that replicates some samples to make sure all devices have same batch size in case of uneven inputs. rank_zero_warn( Validation DataLoader 0: 100%|████████████████████████████████████████████████████████████████████████████████████████████████| 171/171 [00:54<00:00, 3.15it/s] ┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓ ┃ Validate metric ┃ DataLoader 0 ┃ ┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┩ │ validation/accuracy/classification │ 0.8651315569877625 │ │ validation/losses/classification │ 0.4168359339237213 │ Differential Revision: [D37444938](https://our.internmc.facebook.com/intern/diff/D37444938) [ghstack-poisoned]
|
@ankitade has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator. |
1 similar comment
|
@ankitade has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator. |
- Accuracy metric for finetuning - Add checkpoint saving and best ckpt loading based on val accuracy - Load pretrained ckpt by default in classification model - make num gpus 1 in qnli.yaml Test plan python -m flava.finetune config=flava/configs/finetuning/qnli.yaml (val acc : 0.8651) Loaded model weights from checkpoint at /data/home/deankita/torchmultimodal/examples/flava-epoch=03-step=10000.ckpt /data/home/deankita/miniconda/envs/flava/lib/python3.8/site-packages/pytorch_lightning/trainer/connectors/data_connector.py:330: PossibleUserWarning: Using `DistributedSampler` with the dataloaders. During `trainer.validate()`, it is recommended to use `Trainer(devices=1)` to ensure each sample/batch gets evaluated exactly once. Otherwise, multi-device settings use `DistributedSampler` that replicates some samples to make sure all devices have same batch size in case of uneven inputs. rank_zero_warn( Validation DataLoader 0: 100%|████████████████████████████████████████████████████████████████████████████████████████████████| 171/171 [00:54<00:00, 3.15it/s] ┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓ ┃ Validate metric ┃ DataLoader 0 ┃ ┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┩ │ validation/accuracy/classification │ 0.8651315569877625 │ │ validation/losses/classification │ 0.4168359339237213 │ Differential Revision: [D37444938](https://our.internmc.facebook.com/intern/diff/D37444938) [ghstack-poisoned]
- Accuracy metric for finetuning - Add checkpoint saving and best ckpt loading based on val accuracy - Load pretrained ckpt by default in classification model - make num gpus 1 in qnli.yaml Test plan python -m flava.finetune config=flava/configs/finetuning/qnli.yaml (val acc : 0.8651) Loaded model weights from checkpoint at /data/home/deankita/torchmultimodal/examples/flava-epoch=03-step=10000.ckpt /data/home/deankita/miniconda/envs/flava/lib/python3.8/site-packages/pytorch_lightning/trainer/connectors/data_connector.py:330: PossibleUserWarning: Using `DistributedSampler` with the dataloaders. During `trainer.validate()`, it is recommended to use `Trainer(devices=1)` to ensure each sample/batch gets evaluated exactly once. Otherwise, multi-device settings use `DistributedSampler` that replicates some samples to make sure all devices have same batch size in case of uneven inputs. rank_zero_warn( Validation DataLoader 0: 100%|████████████████████████████████████████████████████████████████████████████████████████████████| 171/171 [00:54<00:00, 3.15it/s] ┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓ ┃ Validate metric ┃ DataLoader 0 ┃ ┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┩ │ validation/accuracy/classification │ 0.8651315569877625 │ │ validation/losses/classification │ 0.4168359339237213 │ Differential Revision: [D37444938](https://our.internmc.facebook.com/intern/diff/D37444938) [ghstack-poisoned]
ebsmothers
reviewed
Jun 30, 2022
| classifier_activation: Callable[..., nn.Module] = nn.ReLU, | ||
| classifier_normalization: Optional[Callable[..., nn.Module]] = None, | ||
| loss_fn: Optional[Callable[..., Tensor]] = None, | ||
| pretrained_model_key: Optional[str] = "flava_full", |
There was a problem hiding this comment.
Just wondering, why do we want to add this as a param to flava_model_for_classification? Feels to me like this class should not care about the pretrained weights. I like TorchVision's approach for handling this, maybe we can do something similar?
There was a problem hiding this comment.
I am just doing this for "short term uniformity" with the flava_for_pretraining. But yeah i agree, tv's approach is nicer and something we can potentially follow
This was referenced Jul 4, 2022
|
@apsdehal I am going to merge this today. please feel free to leave comments either ways and ping me. I can address them in a separate PR if I end up merging before u review |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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.
Test plan
python -m flava.finetune config=flava/configs/finetuning/qnli.yaml
(val acc : 0.8651)
Loaded model weights from checkpoint at /data/home/deankita/torchmultimodal/examples/flava-epoch=03-step=10000.ckpt
/data/home/deankita/miniconda/envs/flava/lib/python3.8/site-packages/pytorch_lightning/trainer/connectors/data_connector.py:330: PossibleUserWarning: Using
DistributedSamplerwith the dataloaders. Duringtrainer.validate(), it is recommended to useTrainer(devices=1)to ensure each sample/batch gets evaluated exactly once. Otherwise, multi-device settings useDistributedSamplerthat replicates some samples to make sure all devices have same batch size in case of uneven inputs.rank_zero_warn(
Validation DataLoader 0: 100%|████████████████████████████████████████████████████████████████████████████████████████████████| 171/171 [00:54<00:00, 3.15it/s]
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓
┃ Validate metric ┃ DataLoader 0 ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┩
│ validation/accuracy/classification │ 0.8651315569877625 │
│ validation/losses/classification │ 0.4168359339237213 │
Stack from ghstack (oldest at bottom):
Differential Revision: D37444938