Electrocom VLM V2 (450M)
Electrocom VLM V2 is a specialized Vision-Language Model optimized for electronic component detection and recognition. It is a fine-tuned version of Liquid AI's LFM2.5-VL-450M.
This "Phase 2" version was trained on a combined dataset of ElectroCom61 and Electronic Detection, significantly improving its visual robustness and detection accuracy.
📊 Performance
| Metric | Base LFM2.5-VL-450M | Electrocom VLM V2 (Ours) |
|---|---|---|
| Grounding Recall (IoU > 0.5) | 0.0% | 31.9% |
| Inference Speed (tok/s) | 81.6 | 85.2 |
Key Improvements:
- 7.5x Better Grounding: Compared to our V1 model, the V2 model shows a massive leap in its ability to correctly localize components.
- Structured JSON Output: The model has been trained to output detections in a precise JSON format, suitable for programmatic integration.
- Edge-Ready: At only 450M parameters, this model runs at high speeds even on modest hardware.
🚀 Usage with llama.cpp
The GGUF version of this model is included in this repository for use with llama.cpp.
# Download the model and projector
hf download erikku-sama/lfm2-electronics-vlm-v2 electrocom-vlm-v2-f16.gguf --local-dir .
hf download LiquidAI/LFM2.5-VL-450M-GGUF mmproj-LFM2.5-VL-450m-F16.gguf --local-dir .
# Run inference
llama-mtmd-cli \
--model electrocom-vlm-v2-f16.gguf \
--mmproj mmproj-LFM2.5-VL-450m-F16.gguf \
--image your_image.jpg \
-p "Inspect the image and detect all electronic components. Provide result as a valid JSON: [{\"label\": str, \"bbox\": [x1,y1,x2,y2]}, ...]. Coordinates must be normalized to 0-1." \
--temp 0.1 \
--jinja
🎯 Intended Use
- Automated inventory of electronic components
- Visual assistance for circuit board inspection
- Educational tools for electronics identification
📝 License
This model is based on LFM2.5-VL-450M and is subject to the Liquid AI model license.