---
library_name: peft
base_model: LiquidAI/LFM2.5-VL-450M
tags:
  - lora
  - vision-language-model
  - remote-sensing
  - nwpu-resisc45
license: mit
---

# LFM2.5-VL-450M — NWPU-RESISC45 LoRA

LoRA adapter fine-tuned on [NWPU-RESISC45](https://huggingface.co/datasets/tarek199147/nwpu-resisc45)
for satellite scene classification (45 classes).

## Base model
[LiquidAI/LFM2.5-VL-450M](https://huggingface.co/LiquidAI/LFM2.5-VL-450M)

## Training
- Task: scene classification (classify prompt → class label)
- Dataset: NWPU-RESISC45, ~25 200 train / ~4 725 val samples
- LoRA: rank 8, alpha 16, no dropout
- Epochs: 3, LR: 5e-6 cosine w/ min 5e-7, warmup 100 steps
- Batch: 16 × 2 grad-accum (effective 32), bfloat16