Skip to content

[QEff. Finetuning] Adding finetune_experiemental.py and related files#731

Closed
quic-swatia wants to merge 5 commits intoquic:ft_experimentalfrom
quic-swatia:HF-Trainer-main
Closed

[QEff. Finetuning] Adding finetune_experiemental.py and related files#731
quic-swatia wants to merge 5 commits intoquic:ft_experimentalfrom
quic-swatia:HF-Trainer-main

Conversation

@quic-swatia
Copy link
Copy Markdown
Contributor

@quic-swatia quic-swatia commented Jan 16, 2026 •
edited
Loading

  1. Added FinetuningPipeline (finetune_experiemental.py) which integrates all the components added for HF-trainer and enable running fine tuning through it.
  2. Added files to handle PEFT and training config.
  3. Made changes in the config_manager and callbacks files.
  4. Added unit tests for the FinetuningPipeline (test_finetune.py)
  5. Updated tests in test_callback and test_config_manager based on above changes.

@quic-swatia
Adding finetune_experiemental.py and related files
b260357 
Signed-off-by: Swati Allabadi <sallabad@qti.qualcomm.com>
@quic-swatia quic-swatia changed the title Adding finetune_experiemental.py and related files [QEff. Finetuning] Adding finetune_experiemental.py and related files Jan 16, 2026
quic-akuruvil
quic-akuruvil reviewed Jan 16, 2026
View reviewed changes
Comment thread QEfficient/finetune/experimental/core/utils/training_config_utils.py
Comment thread QEfficient/finetune/experimental/core/utils/training_config_utils.py
quic-akuruvil
quic-akuruvil reviewed Jan 16, 2026
View reviewed changes
Copy link
Copy Markdown
Contributor

@quic-akuruvil quic-akuruvil left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Check and verify the functionality of python -m Qefficient.finetune_experimental.py with the new stack.

Comment thread QEfficient/cloud/finetune_experimental.py Outdated
@quic-swatia
Added unit test cases and addresses the review comments
d0d3251 
Signed-off-by: Swati Allabadi <sallabad@qti.qualcomm.com>
@quic-swatia quic-swatia force-pushed the HF-Trainer-main branch 2 times, most recently from 0043eaa to d0d3251 Compare January 28, 2026 21:29
@quic-swatia
Syncing it as per the changes in dataset and config manager file
a7278cb 
Signed-off-by: Swati Allabadi <sallabad@qti.qualcomm.com>
@quic-swatia
Fixing linting errors
a30e5a1 
Signed-off-by: Swati Allabadi <sallabad@qti.qualcomm.com>
quic-akuruvil
quic-akuruvil reviewed Jan 29, 2026
View reviewed changes
Comment thread QEfficient/cloud/finetune_experimental.py
quic-akuruvil
quic-akuruvil reviewed Jan 30, 2026
View reviewed changes
Comment thread QEfficient/cloud/finetune_experimental.py Outdated
quic-akuruvil
quic-akuruvil reviewed Jan 30, 2026
View reviewed changes
Comment thread QEfficient/cloud/finetune_experimental.py Outdated
quic-akuruvil
quic-akuruvil reviewed Jan 30, 2026
View reviewed changes
Copy link
Copy Markdown
Contributor

@quic-akuruvil quic-akuruvil left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

--enble_pp, pipeline parallelism support is also missing. Please add that too.

@quic-meetkuma quic-meetkuma mentioned this pull request Jan 30, 2026
Merged
quic-meetkuma
quic-meetkuma reviewed Jan 30, 2026
View reviewed changes
Copy link
Copy Markdown
Contributor

@quic-meetkuma quic-meetkuma left a comment •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looks cleaner. Tests might need proper refactoring. Let us try to close this at the earliest.

PS: add description to the PR.

Comment thread QEfficient/cloud/finetune_experimental.py
try:
import torch_qaic # noqa: F401
except ImportError as e:
logger.log_rank_zero(
Copy link
Copy Markdown
Contributor

@quic-meetkuma quic-meetkuma Jan 30, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Here we are passing and not blocking. It will be fine here but if user has provided device="qaic" and torch_qaic is loaded then we need to break the execution. This has to happen either here or inside ConfigManager. I believe this kind of validation and all other validations about config should reside inside ConfigManager.

CC: @tchawada

Copy link
Copy Markdown
Contributor

@tchawada tchawada Feb 12, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Added in new PR

Comment thread QEfficient/cloud/finetune_experimental.py Outdated
Comment thread QEfficient/finetune/experimental/core/callbacks.py
Comment thread QEfficient/finetune/experimental/core/callbacks.py Outdated
Comment thread QEfficient/finetune/experimental/core/callbacks.py Outdated
Comment thread QEfficient/cloud/finetune_experimental.py Outdated
Comment thread QEfficient/cloud/finetune_experimental.py Outdated
Comment thread QEfficient/finetune/experimental/core/callbacks.py
Comment thread QEfficient/finetune/experimental/tests/test_finetune.py Outdated
Comment thread QEfficient/finetune/experimental/tests/test_finetune.py
@quic-akuruvil
Copy link
Copy Markdown
Contributor

quic-akuruvil commented Feb 4, 2026

Please check and need to add trl library also in requirements, if not already present.

@quic-swatia
Addressed review comments and made changes to make code compatible wi…
04279f2 
…th recetly merged PRs. Made changes in the test files as well accordinlgy.

Signed-off-by: Swati Allabadi <sallabad@qti.qualcomm.com>
@quic-swatia
Copy link
Copy Markdown
Contributor Author

quic-swatia commented Feb 11, 2026

Check and verify the functionality of python -m Qefficient.finetune_experimental.py with the new stack.

Its's working with this PR.

@quic-swatia
Copy link
Copy Markdown
Contributor Author

quic-swatia commented Feb 11, 2026

--enble_pp, pipeline parallelism support is also missing. Please add that too.

PP enablement will be done iteratively in the subsequent PR.

tchawada
tchawada reviewed Feb 13, 2026
View reviewed changes
Comment thread QEfficient/cloud/finetune_experimental.py Outdated
# callback_config.callbacks is a dictionary of callback configurations
for callback_name, callback_kwargs in callback_config["callbacks"].items():
try:
callback_instance = create_callbacks(callback_name, **callback_kwargs)
Copy link
Copy Markdown
Contributor

@tchawada tchawada Feb 2, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Currently it is not handling case in which callback_kwargs is None, can you see it once.

Comment thread QEfficient/cloud/finetune_experimental.py
Execute the complete fine-tuning pipeline.
"""
# Validate configuration
self.config_manager.validate_config()
Copy link
Copy Markdown
Contributor

@tchawada tchawada Feb 2, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Okay

Comment thread QEfficient/cloud/finetune_experimental.py
try:
import torch_qaic # noqa: F401
except ImportError as e:
logger.log_rank_zero(
Copy link
Copy Markdown
Contributor

@tchawada tchawada Feb 12, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Added in new PR

@quic-swatia
Copy link
Copy Markdown
Contributor Author

quic-swatia commented Mar 25, 2026

These changes have been added in PR #791 and it's merged.

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

Reviewers

@quic-akuruvil quic-akuruvil quic-akuruvil left review comments

@quic-meetkuma quic-meetkuma quic-meetkuma left review comments

@quic-rishinr quic-rishinr Awaiting requested review from quic-rishinr quic-rishinr is a code owner

@ochougul ochougul Awaiting requested review from ochougul ochougul is a code owner

@quic-hemagnih quic-hemagnih Awaiting requested review from quic-hemagnih quic-hemagnih is a code owner

@quic-amitraj quic-amitraj Awaiting requested review from quic-amitraj quic-amitraj is a code owner

+1 more reviewer

@tchawada tchawada tchawada left review comments

Reviewers whose approvals may not affect merge requirements

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

4 participants

@quic-swatia @quic-akuruvil @quic-meetkuma @tchawada