HIKARI-Altair-8B-SkinDx

Healthcare-oriented Intelligent Knowledge Augmented Retrieval and Inference
Named after Altair — bright star in Aquila, fast and direct, like single-image fine-tuning

---

📦 Model Type: Merged Full Model

This is a fully merged model — the LoRA adapter weights have been merged directly into the base model weights.

✅ No adapter loading needed. Load and run directly with transformers, vLLM, or SGLang.

💾 Size: ~17 GB (4 safetensor shards)

🔌 Lightweight adapter version: E27085921/HIKARI-Altair-8B-SkinDx-LoRA (~1.1 GB)

---

Overview

HIKARI-Altair is the single-image fine-tuning baseline for 10-class skin disease diagnosis. It is fine-tuned from Qwen/Qwen3-VL-8B-Thinking with Fuzzy Top-K label sampling, without cascade pretraining or RAG. Use this model to understand the baseline performance before cascade and RAG improvements.

Property	Value
Task	10-class skin disease diagnosis (Stage 2)
Base model	Qwen/Qwen3-VL-8B-Thinking
Training	Unsloth + LoRA, Fuzzy Top-K sampling
Val accuracy	74.00% (99 samples, SkinCAP 3-stage split)
Model type	Merged full model

10 Disease Classes

acne_vulgaris · atopic_dermatitis · melanocytic_nevi · psoriasis · sccis · seborrheic_dermatitis · skin_tag · tinea_versicolor · urticaria · photodermatoses

---

📊 Comparison with Other HIKARI Disease Models

Model	Accuracy	Improvement over this model
HIKARI-Altair (this model)	74.00%	—
HIKARI-Deneb (Cascade FT)	79.80%	+5.80 pp
HIKARI-Sirius (RAG-in-Training) ⭐	85.86%	+11.86 pp

---

🔧 Quick Inference — transformers

from transformers import Qwen3VLForConditionalGeneration, AutoProcessor
import torch
from PIL import Image

model_id = "E27085921/HIKARI-Altair-8B-SkinDx"

processor = AutoProcessor.from_pretrained(model_id, trust_remote_code=True)
model = Qwen3VLForConditionalGeneration.from_pretrained(
    model_id, torch_dtype=torch.bfloat16, device_map="auto", trust_remote_code=True
)

image = Image.open("skin_lesion.jpg").convert("RGB")
group = "inflammatory"  # from Stage 1 (HIKARI-Subaru)

PROMPT = (
    "This skin lesion belongs to the group '{group}'. "
    "Examine the lesion morphology (papules, plaques, macules), "
    "color (red, violet, white, brown), scale/crust, border sharpness, "
    "and distribution pattern. Based on these visual features, "
    "what is the specific skin disease?"
)

messages = [{"role": "user", "content": [
    {"type": "image", "image": image},
    {"type": "text", "text": PROMPT.format(group=group)},
]}]
text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = processor(text=[text], images=[image], return_tensors="pt").to(model.device)

with torch.no_grad():
    out = model.generate(**inputs, max_new_tokens=64, temperature=0.0, do_sample=False)

print(processor.batch_decode(out[:, inputs["input_ids"].shape[1]:], skip_special_tokens=True)[0].strip())

---

🔌 LoRA Adapter Version

from peft import PeftModel
from transformers import Qwen3VLForConditionalGeneration
import torch

base = Qwen3VLForConditionalGeneration.from_pretrained(
    "Qwen/Qwen3-VL-8B-Thinking", torch_dtype=torch.bfloat16, device_map="auto"
)
model = PeftModel.from_pretrained(base, "E27085921/HIKARI-Altair-8B-SkinDx-LoRA")

→ E27085921/HIKARI-Altair-8B-SkinDx-LoRA

---

📄 Citation

@misc{hikari2026,
  title  = {HIKARI: RAG-in-Training for Skin Disease Diagnosis
            with Cascaded Vision-Language Models},
  author = {Watin Promfiy and Pawitra Boonprasart},
  year   = {2026},
  institution = {King Mongkut's Institute of Technology Ladkrabang,
                 Department of Information Technology, Bangkok, Thailand}
}

Made with ❤️ at King Mongkut's Institute of Technology Ladkrabang (KMITL)