-
Notifications
You must be signed in to change notification settings - Fork 2.6k
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.
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
add support for new mcore ds features #9388
Conversation
Signed-off-by: dimapihtar <[email protected]>
Signed-off-by: dimapihtar <[email protected]>
Signed-off-by: dimapihtar <[email protected]>
nemo/collections/nlp/data/language_modeling/megatron/data_samplers.py
Dismissed
Show dismissed
Hide dismissed
Signed-off-by: dimapihtar <[email protected]>
Signed-off-by: dimapihtar <[email protected]>
Signed-off-by: dimapihtar <[email protected]>
@dimapihtar - can you add |
@ShriyaPalsamudram we have these values as True by default so I think all the tests are running in this way. |
Signed-off-by: dimapihtar <[email protected]>
Signed-off-by: dimapihtar <[email protected]>
Signed-off-by: dimapihtar <[email protected]>
Signed-off-by: dimapihtar <[email protected]>
One necessary consistency test will be to measure validation eval between legacy and mcore code paths, with all options turned on to eval on the entire validation dataset You'll have to make sure that the loss and ppl computation is able to handle partial batches and partial sequences. NeMo already does this at least for the legacy code path, but you'll want to make sure there aren't any breaking changes. |
nemo/collections/nlp/models/language_modeling/megatron_gpt_model.py
Outdated
Show resolved
Hide resolved
nemo/collections/nlp/data/language_modeling/megatron/data_samplers.py
Outdated
Show resolved
Hide resolved
nemo/collections/nlp/models/language_modeling/megatron_gpt_model.py
Outdated
Show resolved
Hide resolved
…lers.py Co-authored-by: jbaczek <[email protected]> Signed-off-by: Dmytro Pykhtar <[email protected]>
…el.py Co-authored-by: jbaczek <[email protected]> Signed-off-by: Dmytro Pykhtar <[email protected]>
Signed-off-by: Dmytro Pykhtar <[email protected]>
* add validation_drop_last and add_extra_token params support for mcore ds Signed-off-by: dimapihtar <[email protected]> * pad samples with dummy tokens only Signed-off-by: dimapihtar <[email protected]> * Apply isort and black reformatting Signed-off-by: dimapihtar <[email protected]> * use no_seqlen_plus_one_input_tokens as mcore's add_extra_token Signed-off-by: dimapihtar <[email protected]> * revert config Signed-off-by: dimapihtar <[email protected]> * revert config Signed-off-by: dimapihtar <[email protected]> * set train_valid_test_num_samples[1] to None Signed-off-by: dimapihtar <[email protected]> * add test case when validation_drop_last is False Signed-off-by: dimapihtar <[email protected]> * revert config Signed-off-by: dimapihtar <[email protected]> * set validation_drop_last as True by default Signed-off-by: dimapihtar <[email protected]> * Update nemo/collections/nlp/data/language_modeling/megatron/data_samplers.py Co-authored-by: jbaczek <[email protected]> Signed-off-by: Dmytro Pykhtar <[email protected]> * Update nemo/collections/nlp/models/language_modeling/megatron_gpt_model.py Co-authored-by: jbaczek <[email protected]> Signed-off-by: Dmytro Pykhtar <[email protected]> --------- Signed-off-by: dimapihtar <[email protected]> Signed-off-by: dimapihtar <[email protected]> Signed-off-by: Dmytro Pykhtar <[email protected]> Co-authored-by: dimapihtar <[email protected]> Co-authored-by: jbaczek <[email protected]> Signed-off-by: Jan Lasek <[email protected]>
* add validation_drop_last and add_extra_token params support for mcore ds Signed-off-by: dimapihtar <[email protected]> * pad samples with dummy tokens only Signed-off-by: dimapihtar <[email protected]> * Apply isort and black reformatting Signed-off-by: dimapihtar <[email protected]> * use no_seqlen_plus_one_input_tokens as mcore's add_extra_token Signed-off-by: dimapihtar <[email protected]> * revert config Signed-off-by: dimapihtar <[email protected]> * revert config Signed-off-by: dimapihtar <[email protected]> * set train_valid_test_num_samples[1] to None Signed-off-by: dimapihtar <[email protected]> * add test case when validation_drop_last is False Signed-off-by: dimapihtar <[email protected]> * revert config Signed-off-by: dimapihtar <[email protected]> * set validation_drop_last as True by default Signed-off-by: dimapihtar <[email protected]> * Update nemo/collections/nlp/data/language_modeling/megatron/data_samplers.py Co-authored-by: jbaczek <[email protected]> Signed-off-by: Dmytro Pykhtar <[email protected]> * Update nemo/collections/nlp/models/language_modeling/megatron_gpt_model.py Co-authored-by: jbaczek <[email protected]> Signed-off-by: Dmytro Pykhtar <[email protected]> --------- Signed-off-by: dimapihtar <[email protected]> Signed-off-by: dimapihtar <[email protected]> Signed-off-by: Dmytro Pykhtar <[email protected]> Co-authored-by: dimapihtar <[email protected]> Co-authored-by: jbaczek <[email protected]>
* add validation_drop_last and add_extra_token params support for mcore ds Signed-off-by: dimapihtar <[email protected]> * pad samples with dummy tokens only Signed-off-by: dimapihtar <[email protected]> * Apply isort and black reformatting Signed-off-by: dimapihtar <[email protected]> * use no_seqlen_plus_one_input_tokens as mcore's add_extra_token Signed-off-by: dimapihtar <[email protected]> * revert config Signed-off-by: dimapihtar <[email protected]> * revert config Signed-off-by: dimapihtar <[email protected]> * set train_valid_test_num_samples[1] to None Signed-off-by: dimapihtar <[email protected]> * add test case when validation_drop_last is False Signed-off-by: dimapihtar <[email protected]> * revert config Signed-off-by: dimapihtar <[email protected]> * set validation_drop_last as True by default Signed-off-by: dimapihtar <[email protected]> * Update nemo/collections/nlp/data/language_modeling/megatron/data_samplers.py Co-authored-by: jbaczek <[email protected]> Signed-off-by: Dmytro Pykhtar <[email protected]> * Update nemo/collections/nlp/models/language_modeling/megatron_gpt_model.py Co-authored-by: jbaczek <[email protected]> Signed-off-by: Dmytro Pykhtar <[email protected]> --------- Signed-off-by: dimapihtar <[email protected]> Signed-off-by: dimapihtar <[email protected]> Signed-off-by: Dmytro Pykhtar <[email protected]> Co-authored-by: dimapihtar <[email protected]> Co-authored-by: jbaczek <[email protected]>
* add validation_drop_last and add_extra_token params support for mcore ds Signed-off-by: dimapihtar <[email protected]> * pad samples with dummy tokens only Signed-off-by: dimapihtar <[email protected]> * Apply isort and black reformatting Signed-off-by: dimapihtar <[email protected]> * use no_seqlen_plus_one_input_tokens as mcore's add_extra_token Signed-off-by: dimapihtar <[email protected]> * revert config Signed-off-by: dimapihtar <[email protected]> * revert config Signed-off-by: dimapihtar <[email protected]> * set train_valid_test_num_samples[1] to None Signed-off-by: dimapihtar <[email protected]> * add test case when validation_drop_last is False Signed-off-by: dimapihtar <[email protected]> * revert config Signed-off-by: dimapihtar <[email protected]> * set validation_drop_last as True by default Signed-off-by: dimapihtar <[email protected]> * Update nemo/collections/nlp/data/language_modeling/megatron/data_samplers.py Co-authored-by: jbaczek <[email protected]> Signed-off-by: Dmytro Pykhtar <[email protected]> * Update nemo/collections/nlp/models/language_modeling/megatron_gpt_model.py Co-authored-by: jbaczek <[email protected]> Signed-off-by: Dmytro Pykhtar <[email protected]> --------- Signed-off-by: dimapihtar <[email protected]> Signed-off-by: dimapihtar <[email protected]> Signed-off-by: Dmytro Pykhtar <[email protected]> Co-authored-by: dimapihtar <[email protected]> Co-authored-by: jbaczek <[email protected]>
What does this PR do ?
Adds new mcore dataset features to NeMo.
Collection: [Note which collection this PR will affect]
Changelog
Usage
# Add a code snippet demonstrating how to use this
GitHub Actions CI
The Jenkins CI system has been replaced by GitHub Actions self-hosted runners.
The GitHub Actions CI will run automatically when the "Run CICD" label is added to the PR.
To re-run CI remove and add the label again.
To run CI on an untrusted fork, a NeMo user with write access must first click "Approve and run".
Before your PR is "Ready for review"
Pre checks:
PR Type:
If you haven't finished some of the above items you can still open "Draft" PR.
Who can review?
Anyone in the community is free to review the PR once the checks have passed.
Contributor guidelines contains specific people who can review PRs to various areas.
Additional Information