Fill in typical peak LRs for full FT, LoRA, and DPO

Attempt it

Sensible peak LRs (with cosine schedule + warmup): full fine-tuning of a 7B model sits around to 5e-5. LoRA r=16 tolerates roughly × higher, so peak around 1e-4 to . DPO from an SFT checkpoint needs a much LOWER LR: around to 5e-6.

Quick reference

TL;DR

Full FT a 7B model around 1e-5 to 5e-5; LoRA roughly 10x higher near 1e-4 to 5e-4; DPO much lower near 5e-7 to 5e-6. Always cosine schedule with linear warmup.

Memory aid

Sign in for the mnemonic that makes this stick. See Pro pricing.

Easy to grasp

Imagine fixing up something valuable, and the size of your step is how bold each change is. If you repaint a whole finished masterpiece, you dab gently, because one big swipe ruins years of work. If you only add a few stick-on tabs that started blank, you can press hard and move fast, since nothing precious is underneath. The trickiest job is nudging someone's taste toward what they should like best: the hint is so faint that any big shove smears it, so you barely tap. And no matter the job, you start slow to find your footing, speed up, then slow down again as you near the finish, like easing into a parking spot.

Key concepts

Learning rate LoRA adapters Direct Preference Optimization Cosine schedule with warmup

Quick scan · 2–4 min

Everything important, quickly.

5 min: why LR scales with parameters moved and signal fragility + full FT band + LoRA headroom + DPO fragility + warmup and cosine schedule roles + when defaults break.

Side-by-side

Method	Peak LR (7B)	Why	Schedule
Full fine-tuning	1e-5 to 5e-5	All weights move from a delicate converged minimum	Cosine, 3 to 10 percent warmup
LoRA (rank 16)	1e-4 to 5e-4	Only zero-init adapters train, so more headroom	Cosine, short warmup
DPO from SFT	5e-7 to 5e-6	Fragile preference gradient near the reference policy	Cosine or constant, short warmup

Real-world examples

Real products, models, and research that use this idea.

Axolotl and Hugging Face TRL ship cosine schedule with linear warmup as the default for SFT and DPO runs on Llama 4 and similar open-weight bases.
Unsloth's LoRA notebooks for Llama 4 and Qwen3 default to a peak around 2e-4 with a short warmup ratio, matching the 1e-4 to 5e-4 LoRA band.
The original DPO reference implementation and TRL's DPOTrainer recommend learning rates near 5e-7, an order of magnitude below the SFT checkpoint they start from.
Together.ai and Modal Labs fine-tuning recipes for DeepSeek V4 and Llama 4 expose warmup ratio and cosine decay as first-class knobs because the defaults rarely need changing.

Interviewer follow-ups

What an interviewer would ask next. Try answering before peeking at the approach.

QWhy can LoRA tolerate a higher learning rate than full fine-tuning of the same base?

Sign in or go Pro to see the approach for this follow-up.

QWhat specifically goes wrong when you run DPO at an SFT-scale learning rate?

Sign in or go Pro to see the approach for this follow-up.

QHow do warmup ratio and cosine decay each change the training trajectory?

Sign in or go Pro to see the approach for this follow-up.

60-second cheat sheet

The night-before-the-interview bullets. Scan these on the way to the call.

Sign in for 5 unlocks/day or go Pro for unlimited revision sheets. See Pro pricing.

References

Primary sources. Skim if you want the original framing.