Llama · Meta
Fine-tuning Llama 3.2 3B
What does it take to fine-tune Llama 3.2 3B?
Llama 3.2 3B needs 10 GB for half-precision LoRA training and 6 GB in four-bit, and 7 GB to serve. It supports supervised fine-tuning, preference tuning (dpo), under the Llama 3.2 Community licence. The cheapest qualifying class is RTX 3080 at $0.09 per GPU-hour.
The middle of the small Llama range, and a common landing point for teams shrinking a task off a frontier API. It is large enough to keep formatting discipline under pressure and small enough that serving it continuously costs less than a coffee a day.
What to know before choosing it
This is where a lot of frontier-API replacements land when the task is narrow and the volume is high. It holds formatting discipline under pressure better than most 3B models, which is the specific property deciding whether an extraction pipeline needs a repair layer.
Serving it continuously is cheap enough that scale-to-zero is often not worth the cold-start risk. That is a genuinely different operational posture from the larger models, and it simplifies the deployment considerably.
Specification
| Parameters | 3B |
|---|---|
| Architecture | Dense |
| Size tier | Small |
| LoRA training memory | 10 GBHalf precision, frozen base |
| QLoRA training memory | 6 GBFour-bit base, higher-precision adapter |
| Serving memory | 7 GBHalf precision, before attention cache |
| Objectives | SFT, DPO |
| Licence | Llama 3.2 Community |
| Repository | meta-llama/Llama-3.2-3B |
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 3080 | 12 GB | $0.09 | $0.11 | LoRA and QLoRA |
| RTX 4000 Ada | 20 GB | $0.09 | $0.11 | LoRA and QLoRA |
| 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 |
Serving fits on RTX 3080 at $0.11 per hour — before the attention cache, which grows with context length and concurrency.
Good starting point for
- Replacing a frontier model on a narrow, high-volume task
- Always-on endpoints with tight cost ceilings
Preference tuning on this model
Preference tuning holds a frozen reference copy of the model alongside the one being trained, so budget roughly 20 GB rather than 10 GB. That is the thing that catches people out — supervised training on this model fits on a class that preference tuning will overflow.
Other sizes in this family
| Model | Params | QLoRA | Objectives |
|---|---|---|---|
| Llama 3.2 1B | 1B | 3 GB | SFT, DPO |
| Llama 3.1 8B | 8B | 14 GB | SFT, DPO, GRPO |
| Llama 3.3 70B | 70B | 48 GB | SFT |
| Llama 4 Scout 109B | 109B | 80 GB | SFT |
Comparable sizes elsewhere
Frequently asked questions
How much VRAM does it take to fine-tune Llama 3.2 3B?
10 GB for half-precision LoRA and 6 GB for four-bit QLoRA. Serving needs 7 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 Llama 3.2 3B?
Four-bit QLoRA on RTX 3080 at $0.09 per GPU-hour is the cheapest class that meets the 6 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 Llama 3.2 3B?
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 Llama 3.2 3B carry?
Llama 3.2 Community. 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 Llama 3.2 3B
Upload a dataset, forecast the run, and see the cost before any compute is leased.