CoolFace
Modelpublic

Arty1001/aquaveritas-lfm-GGUF

sourceHugging Faceapache-2.0updated 5mo agoView on Hugging Face
0likes40downloads
Model Card

AquaVeritas-LFM — GGUF

AquaVeritas-LFM is a fine-tuned vision-language model for automated freshwater body monitoring from Sentinel-2 satellite imagery. It is a full fine-tune of LiquidAI/LFM2.5-VL-450M trained on 2,820 labeled observations across 20 global freshwater locations spanning 2018–2024.

Model Description

  • —Base model: LiquidAI/LFM2.5-VL-450M (Liquid Foundation Model 2.5 Vision-Language)
  • —Task: Satellite image analysis → structured JSON environmental assessment
  • —Fine-tuning: Full fine-tune (no LoRA/PEFT), 3 epochs on H100
  • —Training loss: 0.0113 | Eval loss: 0.01542
  • —Input: RGB + SWIR Sentinel-2 tiles (15 km × 15 km, 10 m/px)
  • —Output: Structured JSON assessment of water extent, flood risk, water clarity, shoreline encroachment, agriculture stress, and land-use change

Files

FileSizeDescription
aquaveritas-lfm-q8_0.gguf~451 MBFull model, Q8_0 quantized — use with llama-server
The base model's mmproj (vision encoder projector) is required for inference. Download it from LiquidAI/LFM2.5-VL-450M-GGUF: mmproj-LFM2.5-VL-450m-F16.gguf

Usage

bash
# Download mmproj from LiquidAI official repo
huggingface-cli download LiquidAI/LFM2.5-VL-450M-GGUF mmproj-LFM2.5-VL-450m-F16.gguf

# Start llama-server
llama-server \
  -m aquaveritas-lfm-q8_0.gguf \
  --mmproj mmproj-LFM2.5-VL-450m-F16.gguf \
  --port 8080 \
  --ctx-size 4096 \
  -ngl 99

Training Data

  • —20 global freshwater locations: Lake Chad, Aral Sea, Dead Sea, Lake Urmia, Lake Victoria, Lake Titicaca, Lake Turkana, Tonle Sap, Okavango, Niger Delta, Nile Delta, Mekong Delta, Amazon Delta, Congo Delta, Danube Delta, Mesopotamian Marshes, Omo River, Tana River, Po Valley, Salton Sea
  • —7-year temporal range: 2018–2024 (monthly observations)
  • —2,820 training examples (core zone + buffer zone analysis per observation)
  • —Labels generated by Claude Opus oracle with field-level accuracy ~99%

Evaluation (vs Claude oracle, 30 test observations)

FieldClaudeBase LFM**AquaVeritas-LFM**Δ
Water Extent Status86.7%0.0%100.0%▲ 100%
Flood Risk73.3%33.3%100.0%▲ 67%
Water Clarity93.3%0.0%100.0%▲ 100%
Shoreline Encroachment80.0%50.0%100.0%▲ 50%

Citation

@misc{aquaveritas2026,
  title={AquaVeritas: Automated Freshwater Monitoring via Fine-tuned Vision-Language Models},
  author={ML_LABS},
  year={2026},
  url={https://huggingface.co/Arty1001/aquaveritas-lfm-GGUF}
}