|
|
--- |
|
|
license: apache-2.0 |
|
|
language: en |
|
|
tags: |
|
|
- qwen3 |
|
|
- medical |
|
|
- chat |
|
|
- fine-tuned |
|
|
- gguf |
|
|
- healthcare |
|
|
datasets: |
|
|
- custom-medical-qa |
|
|
model_type: qwen3 |
|
|
base_model: Qwen/Qwen3-0.6B |
|
|
--- |
|
|
|
|
|
# Qwen3-0.6B-Medical-Finetuned-v1 |
|
|
|
|
|
This model is a fine-tuned version of `Qwen/Qwen3-0.6B` specialized for medical question-answering. It's designed to provide helpful, accurate medical information while emphasizing the importance of professional medical consultation. |
|
|
|
|
|
## 🏥 Model Description |
|
|
|
|
|
- **Base Model**: `Qwen/Qwen3-0.6B` |
|
|
- **Fine-tuning Method**: LoRA (Low-Rank Adaptation) |
|
|
- **Dataset**: Custom medical Q&A dataset covering common health topics. |
|
|
- **Training**: Optimized for conversational medical assistance. |
|
|
|
|
|
## ⚠️ Important Disclaimer |
|
|
|
|
|
**This model is NOT a substitute for professional medical advice, diagnosis, or treatment. Always consult qualified healthcare providers for medical concerns. Do not use this model for emergency situations - call emergency services immediately.** |
|
|
|
|
|
## 🚀 Usage |
|
|
|
|
|
### With `transformers` |
|
|
|
|
|
```python |
|
|
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline |
|
|
import torch |
|
|
|
|
|
model_id = "rohitnagareddy/Qwen3-0.6B-Medical-Finetuned-v1" |
|
|
|
|
|
# Load model and tokenizer |
|
|
tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True) |
|
|
model = AutoModelForCausalLM.from_pretrained( |
|
|
model_id, |
|
|
torch_dtype=torch.float16, |
|
|
device_map="auto", |
|
|
trust_remote_code=True |
|
|
) |
|
|
|
|
|
# Create a conversation pipeline |
|
|
pipe = pipeline( |
|
|
"text-generation", |
|
|
model=model, |
|
|
tokenizer=tokenizer |
|
|
) |
|
|
|
|
|
# Create conversation |
|
|
prompt = "<|im_start|>system\nYou are a helpful medical assistant providing accurate, evidence-based information.<|im_end|>\n<|im_start|>user\nWhat are the symptoms of hypertension?<|im_end|>\n<|im_start|>assistant\n" |
|
|
|
|
|
# Generate response |
|
|
response = pipe(prompt, max_new_tokens=300, temperature=0.7, top_p=0.9, do_sample=True) |
|
|
print(response[0]["generated_text"]) |
|
|
``` |
|
|
|
|
|
## 🔧 GGUF Versions |
|
|
|
|
|
This repository includes quantized GGUF versions for use with `llama.cpp` and compatible tools: |
|
|
|
|
|
- `Qwen3-0.6B-Medical-Finetuned-v1.fp16.gguf` - Full precision (largest, best quality) |
|
|
- `Qwen3-0.6B-Medical-Finetuned-v1.Q8_0.gguf` - 8-bit quantization (good balance) |
|
|
- `Qwen3-0.6B-Medical-Finetuned-v1.Q5_K_M.gguf` - 5-bit quantization (smaller, fast) |
|
|
- `Qwen3-0.6B-Medical-Finetuned-v1.Q4_K_M.gguf` - 4-bit quantization (smallest, fastest) |
|
|
|
|
|
### Using with Ollama |
|
|
|
|
|
```bash |
|
|
# Pull the model (once available on the Hub) |
|
|
ollama pull rohitnagareddy/Qwen3-0.6B-Medical-Finetuned-v1 |
|
|
|
|
|
# Run the model |
|
|
ollama run rohitnagareddy/Qwen3-0.6B-Medical-Finetuned-v1 "What are the early signs of diabetes?" |
|
|
``` |
|
|
|
|
|
## 📊 Training Details |
|
|
|
|
|
- **Training Epochs**: 2 |
|
|
- **Batch Size**: 2 (with 4 steps of gradient accumulation) |
|
|
- **Learning Rate**: 2e-4 |
|
|
- **Optimizer**: Paged AdamW 32-bit |
|
|
- **LoRA Rank**: 16 |
|
|
- **LoRA Alpha**: 32 |
|
|
- **Target Modules**: Auto-detected linear layers |
|
|
|
|
|
--- |
|
|
|
|
|
|
