hawkeye-warehouse-safety-gguf

Warehouse Safety Hazard Detection, 2B VLM (Q4 GGUF)

A 1.8 GB vision-language model trained on synthetic warehouse data, quantized to Q4_K_M, that runs offline on consumer hardware via Ollama.

Research project at the University of Guelph supervised by Dr. John Akinyemi.

Quick start

ollama pull hf.co/Jsohal174/hawkeye-warehouse-safety-gguf

ollama run hf.co/Jsohal174/hawkeye-warehouse-safety-gguf \
  "Classify this warehouse image into exactly one of: spill, forklift_violation, improper_stacking, obstacle, safe. Respond with the category and a one-line description." \
  /path/to/image.jpg

For the faster configuration, prefix the prompt with /no_think.

Results

Held-out test set, n=186 images across five categories. Inference on a Mac Mini M4.

Training delta: +26.6 percentage points over the same base, same prompt, same hardware.

On a 12-image hard set of pedestrian-proximity forklift violations, this model scores 12/12, matching GPT-5.5 and Gemini 3.1 Pro through OpenRouter, against 8/12 for Qwen3-VL-235B and 4/12 for Llama-4-Maverick.

Categories

spill, forklift_violation, improper_stacking, obstacle, safe

Training

LoRA on a 2B vision-language base, 2,521 conversations from 727 images, 3.5 hours on a free Kaggle T4. Merged, then quantized to Q4_K_M via llama.cpp with the vision encoder kept at F16.

Related

• Dataset, 913 labeled images: https://huggingface.co/datasets/Jsohal174/warehouse-safety-hazard-dataset
• LoRA adapter: https://huggingface.co/Jsohal174/hawkeye-warehouse-safety-lora
• Full precision merged: https://huggingface.co/Jsohal174/hawkeye-warehouse-safety-merged
• Code and benchmarks: https://github.com/Jsohal174/safety-hazard-detection

Citation

@misc{sohal2026synthetic,
  author = {Sohal, Jaskirat Singh},
  title  = {Synthetic Specialization: Training and Compressing Foundation Models for Domain-Specific Tasks},
  year   = {2026},
  howpublished = {University of Guelph},
  note   = {Supervisor: Dr. John Akinyemi},
  url    = {https://github.com/Jsohal174/safety-hazard-detection}
}