Gemma · Google
Fine-tuning Gemma 3 12B
What does it take to fine-tune Gemma 3 12B?
Gemma 3 12B needs 24 GB for half-precision LoRA training and 18 GB in four-bit, and 24 GB to serve. It supports supervised fine-tuning, under the Gemma Terms of Use licence. The cheapest qualifying class is RTX 4000 Ada at $0.09 per GPU-hour.
Twenty-four gigabytes for half-precision LoRA puts this at the exact ceiling of the mainstream 24GB classes — it fits, with nothing to spare. If your sequences are long, budget for a 48GB class instead of discovering the limit at step 400.
What to know before choosing it
Twenty-four gigabytes for half-precision LoRA puts this exactly at the ceiling of the mainstream classes: it fits, with nothing to spare. If your sequences are long, budget for a 48GB class rather than discovering the limit at step four hundred with a partly-trained model.
At this size Gemma's multilingual behaviour becomes genuinely useful rather than nominal, which is the main reason to choose it over a Qwen3 or Phi of comparable size for non-English work.
Specification
| Parameters | 12B |
|---|---|
| Architecture | Dense |
| Size tier | Mid-large |
| LoRA training memory | 24 GBHalf precision, frozen base |
| QLoRA training memory | 18 GBFour-bit base, higher-precision adapter |
| Serving memory | 24 GBHalf precision, before attention cache |
| Objectives | SFT |
| Licence | Gemma Terms of Use |
| Repository | google/gemma-3-12b-it |
What it costs to train
Every class with enough memory for four-bit training, cheapest first. Total cost is the rate multiplied by wall time, so the cheapest rate is not always the cheapest run — a faster class that finishes sooner frequently wins.
| GPU class | VRAM | Training / hr | Serving / hr | Fits |
|---|---|---|---|---|
| RTX 4000 Ada | 20 GB | $0.09 | $0.11 | QLoRA only |
| L4 | 24 GB | $0.17 | $0.20 | LoRA and QLoRA |
| RTX 3090 | 24 GB | $0.22 | $0.26 | LoRA and QLoRA |
| RTX 4090 | 24 GB | $0.38 | $0.43 | LoRA and QLoRA |
| A40 | 48 GB | $0.42 | $0.48 | LoRA and QLoRA |
| RTX 6000 Ada | 48 GB | $0.61 | $0.71 | LoRA and QLoRA |
| A6000 | 48 GB | $0.65 | $0.75 | LoRA and QLoRA |
| L40S | 48 GB | $0.78 | $0.90 | LoRA and QLoRA |
| A100 40 GB | 40 GB | $1.17 | $1.35 | LoRA and QLoRA |
| A100 80 GB | 80 GB | $1.68 | $1.94 | LoRA and QLoRA |
| H100 80 GB | 80 GB | $2.59 | $2.99 | LoRA and QLoRA |
| H200 | 141 GB | $4.55 | $5.25 | LoRA and QLoRA |
Half-precision LoRA needs 24 GB, so the cheapest class for it is L4 at $0.17 per hour. Below that, training has to be quantised.
Serving fits on L4 at $0.20 per hour — before the attention cache, which grows with context length and concurrency.
Good starting point for
- Mid-large quality on mainstream hardware
- Multilingual work
Other sizes in this family
| Model | Params | QLoRA | Objectives |
|---|---|---|---|
| Gemma 3 1B | 1B | 3 GB | SFT |
| Gemma 3 4B | 4B | 8 GB | SFT |
| Gemma 3 27B | 27B | 36 GB | SFT |
Comparable sizes elsewhere
Frequently asked questions
How much VRAM does it take to fine-tune Gemma 3 12B?
24 GB for half-precision LoRA and 18 GB for four-bit QLoRA. Serving needs 24 GB. Preference tuning roughly doubles the training figure, because a frozen reference model is held alongside the one being trained.
What is the cheapest way to fine-tune Gemma 3 12B?
Four-bit QLoRA on RTX 4000 Ada at $0.09 per GPU-hour is the cheapest class that meets the 18 GB threshold. Note that quantised training is slower per step, so a faster class sometimes costs less over the whole run.
Can I download the weights after fine-tuning Gemma 3 12B?
Yes. Every finished run exposes its trained weights for download, and publishing to a model hub is a single call with a generated model card recording the base model and version the adapter applies to.
What licence does Gemma 3 12B carry?
Gemma Terms of Use. The licence follows the fine-tune — a derivative inherits the base model’s terms, and those terms pass to anyone you give the model to.
Last verified 6 August 2026. Memory thresholds are the platform's own admission limits.
Fine-tune Gemma 3 12B
Upload a dataset, forecast the run, and see the cost before any compute is leased.