109 lines
3.9 KiB
Markdown
109 lines
3.9 KiB
Markdown
---
|
|
license: apache-2.0
|
|
datasets:
|
|
- FreedomIntelligence/medical-o1-reasoning-SFT
|
|
- UCSC-VLAA/MedReason
|
|
base_model:
|
|
- Qwen/Qwen3-1.7B
|
|
language:
|
|
- en
|
|
pipeline_tag: text-generation
|
|
library_name: transformers
|
|
tags:
|
|
- text-generation-inference
|
|
- moe
|
|
- medical
|
|
- biology
|
|
- trl
|
|
---
|
|
|
|
|
|
|
|
# Sculptor-Qwen3\_Med-Reasoning
|
|
|
|
> **Sculptor-Qwen3\_Med-Reasoning** is a fine-tuned variant of the **Qwen3-4B** architecture, trained specifically on the **Med Reason Dataset** to maximize **accurate medical and clinical reasoning**. This model excels at structured diagnostic logic, symptom analysis, and treatment planning, while maintaining lightweight performance, making it ideal for healthcare, medical education, and clinical support applications.
|
|
|
|
> [!note]
|
|
[!GGUF] : https://huggingface.co/prithivMLmods/Sculptor-Qwen3_Med-Reasoning-Q4_K_M-GGUF
|
|
|
|
## Key Features
|
|
|
|
1. **Precision Medical Reasoning with Med Reason Dataset**
|
|
Tailored for clinical reasoning, medical question answering, and evidence-based analysis, powered by the specialized Med Reason fine-tuning.
|
|
|
|
2. **Lightweight Clinical Code Understanding**
|
|
Capable of interpreting and generating medical-related code (e.g., for health data analysis in Python or R), optimized for concise, logic-oriented scripts.
|
|
|
|
3. **Structured Output Formatting**
|
|
Produces well-organized responses in Markdown, JSON, LaTeX, and tabular formats suitable for electronic health records, research documentation, and structured reporting.
|
|
|
|
4. **Instruction-Following Accuracy**
|
|
Tuned for consistent multi-step instruction adherence in clinical cases and decision-making workflows, enhancing reliability for educational and medical use.
|
|
|
|
5. **Multilingual Medical Capabilities**
|
|
Supports clinical reasoning and documentation in over 20 languages, enabling accessibility for global healthcare professionals.
|
|
|
|
6. **Efficient 4B Architecture**
|
|
Based on Qwen3-4B, offering a balanced tradeoff between inference speed and domain-specific accuracy—suitable for deployment on mid-tier GPUs or cloud-based systems.
|
|
|
|
## Quickstart with Transformers
|
|
|
|
```python
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer
|
|
|
|
model_name = "prithivMLmods/Sculptor-Qwen3_Med-Reasoning"
|
|
|
|
model = AutoModelForCausalLM.from_pretrained(
|
|
model_name,
|
|
torch_dtype="auto",
|
|
device_map="auto"
|
|
)
|
|
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
|
|
|
prompt = "A 45-year-old male presents with chest pain and shortness of breath. List possible diagnoses and explain the reasoning."
|
|
|
|
messages = [
|
|
{"role": "system", "content": "You are a clinical reasoning assistant trained on the Med Reason Dataset."},
|
|
{"role": "user", "content": prompt}
|
|
]
|
|
|
|
text = tokenizer.apply_chat_template(
|
|
messages,
|
|
tokenize=False,
|
|
add_generation_prompt=True
|
|
)
|
|
|
|
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
|
|
|
|
generated_ids = model.generate(
|
|
**model_inputs,
|
|
max_new_tokens=512
|
|
)
|
|
generated_ids = [
|
|
output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
|
|
]
|
|
|
|
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
|
print(response)
|
|
```
|
|
|
|
## Intended Use
|
|
|
|
* Clinical reasoning and diagnosis support
|
|
* Medical question answering and tutoring
|
|
* Structured documentation and case analysis
|
|
* JSON/Markdown/tabular medical summaries
|
|
* Education tools for healthcare professionals
|
|
* Multilingual medical documentation and Q\&A
|
|
|
|
## Limitations
|
|
|
|
* Not designed for open-domain creative generation
|
|
* Limited context length compared to larger LLMs
|
|
* Sensitive to ambiguous or poorly formatted inputs
|
|
* May produce errors in complex or adversarial medical prompts
|
|
|
|
## References
|
|
|
|
1. [Qwen2.5 Technical Report](https://arxiv.org/pdf/2412.15115)
|
|
2. [YaRN: Context Window Extension for LLMs](https://arxiv.org/pdf/2309.00071) |