[QEff Finetune] : Made fixes to training script #439
+155
−71
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
quic-mamta
requested review from
quic-rishinr,
ochougul,
quic-hemagnih and
quic-amitraj
as code owners
quic-mamta
commented
Jun 10, 2025
|
|
||
|
|
||
| def get_preprocessed_samsum(dataset_config, tokenizer, split, context_length=None): | ||
| dataset = datasets.load_dataset("Samsung/samsum", split=split, trust_remote_code=True) | ||
| dataset = datasets.load_dataset("knkarthick/samsum", split=split, trust_remote_code=True) |
There was a problem hiding this comment.
Please check if this dataset can be used.
| def get_dataloader_kwargs(train_config, dataset, dataset_processer, mode): | ||
| kwargs = {} | ||
| batch_size = train_config.batch_size_training if mode == "train" else train_config.val_batch_size | ||
| if train_config.enable_ddp: | ||
| print("Length of dataset before: ", len(dataset)) | ||
| dataset = pad_dataset(dataset, batch_size, 2) |
There was a problem hiding this comment.
instead of 2 use world_size here
| @@ -115,10 +115,26 @@ def generate_dataset_config(dataset_name: str) -> Any: | |||
| return dataset_config | |||
|
|
|||
|
|
|||
| def pad_dataset(dataset, batch_size, num_replicas): | |||
| reminder = len(dataset) % (batch_size * num_replicas) | |||
There was a problem hiding this comment.
Please use remainder as variable name here.
quic-mamta
force-pushed
the
jitender_fixes
branch
from
375701d to
6d833cf
Compare
Signed-off-by: Mamta Singh <mamtsing@qti.qualcomm.com>
quic-mamta
force-pushed
the
jitender_fixes
branch
from
6d833cf to
d269d0c
Compare
quic-mamta
changed the title
quic-mamta
assigned quic-mamta and quic-meetkuma and unassigned quic-mamta and quic-meetkuma
quic-swatia
reviewed
Jun 16, 2025
| @@ -235,11 +241,23 @@ def train( | |||
| train_step_metric.append(step_metric_val) | |||
|
|
|||
| if train_config.grad_scaler: | |||
| scaler.scale(loss).backward() # backward pass | |||
| if train_config.enable_ddp: | |||
| with model.no_sync(): | |||
There was a problem hiding this comment.
This will result in no syncing of gradients at any step.
| if train_config.enable_ddp: | ||
| # FIXME: We can not stop transfer of gradient across devices every time. | ||
| # In grad accumulation last step should transfer gradients across devices. | ||
| with model.no_sync(): |
There was a problem hiding this comment.
This will result in no syncing of gradients at any step here as well.
added 3 commits
June 17, 2025 09:38
…ght parameter to make the loss for padded samples as zero.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Made fixes to training script.