LFM2 · Liquid AI
Fine-tuning LFM2 350M
What does it take to fine-tune LFM2 350M?
LFM2 350M needs 3 GB for half-precision LoRA training and 2 GB in four-bit, and 2 GB to serve. It supports supervised fine-tuning, under the LFM Open licence. The cheapest qualifying class is RTX 3080 at $0.09 per GPU-hour.
The smallest model in the catalogue by an order of magnitude, using a hybrid of gated convolutions and grouped-query attention rather than pure attention. Two gigabytes to serve. Licence is Apache-2.0-derived but adds a commercial revenue threshold, so check it if you are past ten million in revenue.
What to know before choosing it
Two gigabytes to serve puts this in a different category from everything else in the catalogue — it is deployable where a GPU may not exist at all. The hybrid architecture, combining gated convolutions with attention, is what makes that footprint achievable at this quality.
The licence is Apache-2.0-derived but adds a commercial revenue threshold. That is fine for most teams and a blocker for some, and it is exactly the kind of clause that is cheaper to read now than to discover during a funding round.
Specification
| Parameters | 350M |
|---|---|
| Architecture | Hybrid |
| Size tier | Tiny |
| LoRA training memory | 3 GBHalf precision, frozen base |
| QLoRA training memory | 2 GBFour-bit base, higher-precision adapter |
| Serving memory | 2 GBHalf precision, before attention cache |
| Objectives | SFT |
| Licence | LFM Open |
| Repository | LiquidAI/LFM2-350M |
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
- Genuinely on-device inference
- Extremely constrained memory budgets
- Single-purpose extraction
Other sizes in this family
| Model | Params | QLoRA | Objectives |
|---|---|---|---|
| LFM2 700M | 700M | 3 GB | SFT |
| LFM2 1.2B | 1.2B | 4 GB | SFT |
| LFM2 2.6B | 2.6B | 6 GB | SFT |
| LFM2 8B-A1B | 8.3B | 12 GB | SFT |
| LFM2 24B-A2B | 24B | 28 GB | SFT |
Comparable sizes elsewhere
Frequently asked questions
How much VRAM does it take to fine-tune LFM2 350M?
3 GB for half-precision LoRA and 2 GB for four-bit QLoRA. Serving needs 2 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 LFM2 350M?
Four-bit QLoRA on RTX 3080 at $0.09 per GPU-hour is the cheapest class that meets the 2 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 LFM2 350M?
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 LFM2 350M carry?
LFM Open. 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 LFM2 350M
Upload a dataset, forecast the run, and see the cost before any compute is leased.