Rayless sft val loss #5
Merged
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
There was a problem hiding this comment.
Pull request overview
This PR adds validation loss calculation capability to the SFT training script, enabling periodic evaluation on a separate validation dataset during training.
Key Changes:
- Introduces validation data source initialization and periodic validation loss calculation
- Adds
--val-prompt-data,--val-interval, and--val-stepscommand-line arguments - Implements
calculate_val_loss()and_val_step()methods for validation execution
Reviewed changes
Copilot reviewed 4 out of 4 changed files in this pull request and generated 11 comments.
| File | Description |
|---|---|
| train_sft.py | Adds validation loss calculation methods, updates data source initialization to support separate validation data, and integrates periodic validation into the training loop |
| miles/rollout/data_source.py | Modifies RolloutDataSource constructor to accept optional prompt_data parameter, enabling reuse for both training and validation datasets |
| miles/utils/arguments.py | Adds three new command-line arguments for configuring validation: data path, interval, and number of steps |
| scripts/run-sft-torchrun.sh | Provides example usage of the new validation arguments with validation data path and configuration |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
emergenz
reviewed
Dec 23, 2025
emergenz
reviewed
Dec 23, 2025
emergenz
reviewed
Dec 23, 2025
emergenz
reviewed
Dec 23, 2025
scripts/run-sft-torchrun.sh
Outdated
| SFT_ARGS=( | ||
| --rollout-function-path miles.rollout.sft_rollout.generate_rollout | ||
| --prompt-data /fast/project/HFMI_SynergyUnit/tab_model/huggingface/nemo_hf_part_jsonl_4k_tokens.parquet | ||
| --val-prompt-data /fast/project/HFMI_SynergyUnit/tab_model/huggingface/nemo_hf_part_jsonl_4k_tokens_validation.parquet |
emergenz
reviewed
Dec 23, 2025
emergenz
reviewed
Dec 23, 2025
emergenz
reviewed
Dec 23, 2025
emergenz
requested changes
Dec 23, 2025
Co-authored-by: Franz Srambical <79149449+emergenz@users.noreply.github.com>
Co-authored-by: Franz Srambical <79149449+emergenz@users.noreply.github.com>
emergenz
reviewed
Dec 23, 2025
miles/utils/arguments.py
Outdated
| type=str, | ||
| default=None, | ||
| help=( | ||
| "The path to the prompt data. " |
There was a problem hiding this comment.
Ah shit sorry, I didn't see this. In that case, let's add The path to the validation prompt data. below.
emergenz
reviewed
Dec 23, 2025
scripts/run-sft-torchrun.sh
Outdated
| --prompt-data /fast/project/HFMI_SynergyUnit/tab_model/huggingface/nemo_hf_part_jsonl_4k_tokens.parquet | ||
| --val-prompt-data /fast/project/HFMI_SynergyUnit/tab_model/huggingface/nemo_hf_part_jsonl_4k_tokens_validation.parquet | ||
| --val-interval 100 | ||
| --val-steps 50 |
There was a problem hiding this comment.
That sounds like a lot of steps for an interval of 100.
emergenz
approved these changes
Dec 23, 2025
emergenz
pushed a commit
that referenced
this pull request
Dec 31, 2025
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.
No description provided.