Skip to content

fix(yaml): add cce and liger to nemotron-h example#3573

Merged
winglian merged 1 commit into
mainfrom
fix/nemo-h-improv
Apr 6, 2026
Merged

fix(yaml): add cce and liger to nemotron-h example#3573
winglian merged 1 commit into
mainfrom
fix/nemo-h-improv

Conversation

@NanoCode012
Copy link
Copy Markdown
Collaborator

@NanoCode012 NanoCode012 commented Apr 2, 2026

Description

Motivation and Context

How has this been tested?

AI Usage Disclaimer

Screenshots (if appropriate)

Types of changes

Social Handles (Optional)

Summary by CodeRabbit

  • Chores
    • Updated Nemotron model configuration files with plugin-based optimization system
    • Enabled multiple Liger optimization features including layer normalization, rotary position embeddings, RMS norm, GLU activation, and gated RMS normalization
    • Migrated from legacy settings to plugin-based approach for improved performance optimization
    • Cleaned up configuration files for better readability and maintainability

@coderabbitai
Copy link
Copy Markdown
Contributor

coderabbitai Bot commented Apr 2, 2026

📝 Walkthrough

Walkthrough

Two Nemotron LoRA YAML configuration files were updated to enable Axolotl plugins (CutCrossEntropyPlugin and LigerPlugin) with corresponding feature toggles, remove the legacy use_cut_cross_entropy setting, and refactor commented documentation for optional MoE expert training configuration.

Changes

Cohort / File(s) Summary
Nemotron LoRA Configuration Updates
examples/nemotron-h/120b-a12b-qlora.yaml, examples/nemotron-h/nano-30b-a3b-qlora.yaml
Added plugins list with CutCrossEntropyPlugin and LigerPlugin; enabled Liger feature flags (layer norm, RoPE, RMS norm, GLU activation, gated RMS norm); removed use_cut_cross_entropy: true in favor of plugin-based approach; reorganized commented guidance for optional lora_target_parameters MoE expert training.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Possibly related PRs

Suggested reviewers

  • winglian
🚥 Pre-merge checks | ✅ 3
✅ Passed checks (3 passed)
Check name Status Explanation
Description Check ✅ Passed Check skipped - CodeRabbit’s high-level summary is enabled.
Title check ✅ Passed The title accurately describes the main changes: adding Axolotl plugins (CCE/CutCrossEntropyPlugin and Liger/LigerPlugin) to the nemotron-h example YAML configuration files.
Docstring Coverage ✅ Passed No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

✏️ Tip: You can configure your own custom pre-merge checks in the settings.

✨ Finishing Touches
🧪 Generate unit tests (beta)
  • Create PR with unit tests
  • Commit unit tests in branch fix/nemo-h-improv

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

Comment @coderabbitai help to get the list of available commands and usage tips.

Copy link
Copy Markdown
Contributor

@coderabbitai coderabbitai Bot left a comment

Choose a reason for hiding this comment

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

Actionable comments posted: 2

🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@examples/nemotron-h/120b-a12b-qlora.yaml`:
- Around line 3-5: The YAML registers CutCrossEntropyPlugin but never sets the
toggle it relies on, so pre_model_load's check (if cfg.cut_cross_entropy)
prevents the plugin from activating; add the top-level config key
cut_cross_entropy: true in this YAML (alongside the existing plugins list) so
CutCrossEntropyPlugin will detect cfg.cut_cross_entropy and apply its patches as
intended.

In `@examples/nemotron-h/nano-30b-a3b-qlora.yaml`:
- Around line 4-6: The config registers CutCrossEntropyPlugin but doesn't enable
it; add the feature toggle cut_cross_entropy: true at the top-level of the YAML
(next to the plugins list) so
axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin is actually
activated and CCE patching is applied; ensure the key is present and set to true
in this and the other example config files that register the plugin.
🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

  • Push a commit to this branch (recommended)
  • Create a new PR with the fixes

ℹ️ Review info
⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: db64f520-029f-468c-8a55-e6acc895fd37

📥 Commits

Reviewing files that changed from the base of the PR and between 573726c and 4ce63a4.

📒 Files selected for processing (2)
  • examples/nemotron-h/120b-a12b-qlora.yaml
  • examples/nemotron-h/nano-30b-a3b-qlora.yaml

Comment thread examples/nemotron-h/120b-a12b-qlora.yaml
Comment thread examples/nemotron-h/nano-30b-a3b-qlora.yaml
@winglian winglian merged commit dc638e7 into main Apr 6, 2026
2 checks passed
@winglian winglian deleted the fix/nemo-h-improv branch April 6, 2026 17:10
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants