GDN mixer

[oleksost]

✨ Description

Adding support for gated DeltaNet mixer. This tries to follow the reference code from, with some differences https://github.com/huggingface/transformers/blob/a5c903f877fda21e739027eed133e03162eb7712/src/transformers/models/qwen3_next/modeling_qwen3_next.py#L593

TODOs:

• make sure tensor parallel training works
• add model exports
• make sure varlen works
• add tests
• update dependencies: new dependency is fast-linear-attention
• run training runs with it, compare loss and grad norms with and without various TP settings <- not needed since we have tests?

Note:

• compare_factor had to be raised to 10 to pass bf16 and fp16 tests, otherwise it fails in the norm
• tp2 should not be used with these modules (kda, gdn) because of norm inside of them

🔍 Type of change

Select all that apply:

• 🐛 Bug fix (non-breaking change that addresses a specific issue)
• 🚀 New feature (non-breaking change that adds functionality)
• ⚠️ Breaking change (a change that could affect existing functionality)
• 📈 Performance improvement/optimization (improves speed, memory usage, or efficiency)
• 🛠️ Code refactor (non-functional changes that improve code readability, structure, etc.)
• 📦 Dependency bump (updates dependencies, including Dockerfile or package changes)
• 📝 Documentation change (updates documentation, including new content or typo fixes)
• 🔧 Infrastructure/Build change (affects build process, CI/CD, or dependencies)

📝 Changes

• Added new mixer -- gated Delta Net, reference implementation is https://github.com/huggingface/transformers/blob/a5c903f877fda21e739027eed133e03162eb7712/src/transformers/models/qwen3_next/modeling_qwen3_next.py#L593
• added a test to check that packed (varlen) implementation results into the same outputs and grads as non-packed

✅ Checklist

Make sure the following tasks are completed before submitting the PR:

General

• 📜 I have read and followed the contributing guidelines.
• 🏷️ I am using a clear and descriptive PR title that summarizes the key change or feature introduced.
• 🎉 The functionality is complete, and I have tested the changes.
• 📝 I have updated the documentation if needed.
• ⚠️ The change does not introduce any new issues (e.g., runtime warnings, type checker errors, linting problems, unhandled edge cases).
• 🧩 I have commented my code, especially in hard-to-understand areas.

Dependencies and Configuration

• 🐋 I have updated the Docker configuration or dependencies, if applicable.
• 🔄 I have ensured compatibility with the existing setup after dependency changes.

Testing

• 🧪 I have added or updated tests to cover my changes.
• ✔️ New and existing tests pass locally with my changes.
• 🚦 I have tested these changes on GPUs and verified training stability.
• 🏋️ I have tested the changes on realistic training workloads, if applicable.

Performance Impact

• 📊 I have run benchmarks where applicable to evaluate the performance impact.
• ✅ The benchmarks show no performance regression.
• 🚀 The benchmarks indicate a potential performance improvement.
• ⚠️ The benchmarks indicate a potential performance degradation.
• 📈 I have provided benchmark results and detailed any performance impact below, if applicable.

📊 Performance Impact Details

If there is any impact on performance, describe it and provide benchmark results, if applicable:

---

🗒️ Additional Notes

Include any additional context, information, or considerations here, such as known issues, follow-up tasks, or backward compatibility concerns.

[tscholak]

good progress, let's land GDN first please.
can you split this up into two PRs?

[tscholak]

LGTM, great work!

[tscholak]

can you keep both the apriel2 and the hybrid_gdn fixtures here please?