VibrationRobotics/bygheart-vision-v2

Bygheart Vision V2 👁️💙

Visual mental health support model with image understanding capabilities.

Model Description

Bygheart Vision V2 is a fine-tuned version of Qwen2.5-VL-7B-Instruct, specialized for mental health support with visual understanding. It can analyze images (facial expressions, environments) to provide more contextual emotional support.

Benchmark Results

Metric	Vision V1	Vision V2	Change
Empathy Score	53.3%	86.7%	+33.4%
Helpfulness	73.3%	86.7%	+13.4%
Crisis Safety	83.3%	66.7%	-16.6%
Overall MH Score	71.2%	78.6%	+7.4%

Model Variants

Variant	Size	Use Case
lora/	323MB	LoRA adapter for fine-tuning
merged/	16.5GB	Full merged model (FP16)
int4/	5.5GB	INT4 quantized for edge deployment

Usage

from transformers import Qwen2_5_VLForConditionalGeneration, AutoProcessor
from PIL import Image

# Load INT4 model for efficient inference
model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
    "VibrationRobotics/bygheart-vision-v2",
    subfolder="int4",
    device_map="auto"
)
processor = AutoProcessor.from_pretrained(
    "VibrationRobotics/bygheart-vision-v2",
    subfolder="int4"
)

# Process image and text
image = Image.open("user_photo.jpg")
messages = [
    {"role": "user", "content": [
        {"type": "image", "image": image},
        {"type": "text", "text": "I'm feeling really down today. This is how I look."}
    ]}
]

# Generate empathetic response
inputs = processor(messages, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=256)
response = processor.decode(outputs[0], skip_special_tokens=True)
print(response)

Training Details

• Base Model: Qwen2.5-VL-7B-Instruct
• Training Time: 2h 20m on DGX Spark (NVIDIA GB10)
• Epochs: 3
• Token Accuracy: 98.3%
• Dataset: 2000 mental health vision scenarios

Intended Use

• Mental health support applications
• Emotional wellness chatbots
• Crisis intervention systems with visual context
• Therapeutic companion apps

Limitations

• Not a replacement for professional mental health care
• May not detect all visual signs of distress
• Should be used alongside human oversight

Citation

@misc{bygheart-vision-v2,
  author = {IAMVC Holdings LLC},
  title = {Bygheart Vision V2: Visual Mental Health Support Model},
  year = {2026},
  publisher = {HuggingFace},
  url = {https://huggingface.co/VibrationRobotics/bygheart-vision-v2}
}

License

Apache 2.0