add x-token alignment foundation and arrow dataset wiring#2253
Open
avenkateshha wants to merge 1 commit intoNVIDIA-NeMo:mainfrom
Open
add x-token alignment foundation and arrow dataset wiring#2253avenkateshha wants to merge 1 commit intoNVIDIA-NeMo:mainfrom
avenkateshha wants to merge 1 commit intoNVIDIA-NeMo:mainfrom
Conversation
Port the baseline x-token alignment utilities and hook arrow_text dataset/config support needed for cross-tokenizer off-policy distillation on current main. Made-with: Cursor
RayenTian
reviewed
Apr 15, 2026
| # --teacher-model Qwen/Qwen3-8B-Base \ | ||
| # --initial-projection-path cross_tokenizer_data/transformation_counts_via_multitoken.pt | ||
|
|
||
| token_aligner: |
Contributor
There was a problem hiding this comment.
What if we move the token_aligner section below the distillation part? So that we know this is a distillation config at first glance?
RayenTian
reviewed
Apr 15, 2026
|
|
||
| data: | ||
| dataset_name: "arrow_text" | ||
| arrow_files: "/lustre/fsw/portfolios/llmservice/users/sdiao/data/climb_nm5.5_phase3_400b_shuffled_text_only_global_shuffle/data-00[0-4][0-9][0-9]-of-02476.arrow" |
Contributor
There was a problem hiding this comment.
If I understand it correctly, the .arrow file serves as the local cache for Hugging Face datasets when we download them using the load_dataset function. Is it possible for us to directly use the official Hugging Face dataset instead of relying on these local arrow cache files? By doing so, users would be able to conveniently reproduce the exact configuration and ensure consistency across different datasets.
WDYT about this? @yuki-97
RayenTian
reviewed
Apr 15, 2026
chtruong814
added
waiting-on-customer
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
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.
Port the baseline x-token alignment utilities and hook arrow_text dataset/config support needed for cross-tokenizer off-policy distillation on current main.
What does this PR do ?
Add a one line overview of what this PR aims to accomplish.
Issues
List issues that this PR closes (syntax):
Usage
# Add a code snippet demonstrating how to use thisBefore your PR is "Ready for review"
Pre checks:
Additional Information