Pipeline-parallel support for Knowledge Distillation (NeMo 2)

[AAnoosheh]

What does this PR do ?

Enable Pipeline-Parallelism in conjunction with student-teacher distillation in NeMo 2

Collection: [LLM]

Changelog

• Create new script to enable KD in NeMo 2
• Add folder nemo/collections/llm/distillation
• Modify the MegatronParallel forward pass to run teacher in addition to student

Usage

        python scripts/llm/gpt_distillation.py \
          --name experiment_name \
          --teacher_path /path/to/nemo2/teacher \
          --student_path /path/to/nemo2/student \
          --tp_size 4 \
          --cp_size 1 \
          --pp_size 2 \
          --devices 8 \
          --log_dir /tmp/nemo2_llama_distill \
          --max_steps 100 \
          --gbs 64 \
          --mbs 4 \
          --data_paths weight1 /path/to/data/1 weight2 /path/to/data/2 \
          --index_mapping_dir /path/to/data/cache/ \
          --seq_length 8192 \
          --warmup_steps 5 \
          --val_check_interval 50 \
          --log_interval 5

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The Jenkins CI system has been replaced by GitHub Actions self-hosted runners.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:

• Make sure you read and followed Contributor guidelines
• Did you write any new necessary tests?
• Did you add or update any necessary documentation?
• Does the PR affect components that are optional to install? (Ex: Numba, Pynini, Apex etc)
  • Reviewer: Does the PR have correct import guards for all optional libraries?

PR Type:

• New Feature
• Bugfix
• Documentation

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

• Related to # (issue)

Megatron-LM vs NeMo 2.0 Llama3.1 8b->4b distillation comparison (Validation LM-Loss)

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 30.43%. Comparing base (09186c3) to head (b8dda2c).
Report is 16 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #11766      +/-   ##
==========================================
+ Coverage   30.30%   30.43%   +0.13%     
==========================================
  Files        1387     1399      +12     
  Lines      176283   177330    +1047     
  Branches    27091    27185      +94     
==========================================
+ Hits        53423    53972     +549     
- Misses     118776   119241     +465     
- Partials     4084     4117      +33     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.