activation-level disillation

[RaymondLi0]

✨ Description

Closes #385

Sanity checks:

• loading student and teacher from the same pretrained model gives 0 loss ✔️. But loss then increases to a small value instead of staying at 0.
• Distilling from scratch with the same architecture doesn't lead to 0 loss (orange)
• Distilling the pretrained model, but with a sliding-window leads to low loss, lower with a larger window (green and purple)

	lm-loss	Logit distillation	Logit + Activation distillation
Tokens/s/gpu	3500	2900	2800
max_reserved (GB)	44	77	78

With the caveat that distillation seems to experience memory spikes at specific points in training. The actual usage was lower most of the time:

🔍 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)

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:

[tscholak]

great progress! did you freeze everything except the randomly initialized mixers?