AxonMed-7B-Flash
a clinically trained medical language model
2 min readupdated
Built on the AxonMed architecture and developed by SaathiMed, this model is designed to process complex medical data. It is optimized for analyzing electronic health records (EHR), clinical decision support, entity extraction, and standardized ICD-10 coding. With 7.62B parameters, the model surpasses the passing threshold for the US Medical Licensing Examination (USMLE), outperforming several similarly sized open-source alternatives. It is built for medical informatics researchers, clinical software developers, and educational use cases.
Key features
- Clinical specialization built on the
AxonMedarchitecture with support for Harvard BRIDGE protocols. - Scores 60.02% on the MedQA benchmark (USMLE Step 1/2/3) in zero-shot mode (without task-specific examples).
- Supports 7 languages: English, French, German, Spanish, Chinese, Hindi, and Russian.
- Open weights released under the Apache 2.0 license for research and applied projects.
What you can do
- Integrate it into clinical decision support systems to assist doctors with diagnostic triage and hospitalization risk assessment.
- Apply it to oncology and pathology analysis, including TNM staging and tumor coding.
- Automate adverse drug event (ADE) extraction and medication dosing recommendations.
- Set up automated ICD-10 coding for electronic health records.
Example
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM
model_id = "Saathimed/AxonMed-7B-Flash"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
model_id,
torch_dtype=torch.float16,
device_map="auto"
)
prompt = """Solve it in a step-by-step fashion, return your answer in the following format, PROVIDE DETAILED ANALYSIS BEFORE THE RESULT:
Analysis:
...
Result:
Task: Cantemist-Coding
Input: Histopathologic examination of lung biopsy reveals invasive adenocarcinoma."""
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=256, temperature=0.2)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))Read more
DetailsBenchmarks: 60.02% on MedQA (USMLE), 74.40% on PubMedQA.
Parameters: 7.62 billion (7.62B), compatible with the
transformers library.Architecture: based on
Saathimed/AxonMed-7B-Thinking.Limitations: intended for research and decision support only; does not replace professional medical diagnosis.










Comments