Instructions to use microsoft/MediPhi-MedCode with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use microsoft/MediPhi-MedCode with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="microsoft/MediPhi-MedCode")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("microsoft/MediPhi-MedCode")
model = AutoModelForCausalLM.from_pretrained("microsoft/MediPhi-MedCode")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use microsoft/MediPhi-MedCode with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "microsoft/MediPhi-MedCode"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "microsoft/MediPhi-MedCode",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/microsoft/MediPhi-MedCode

SGLang

How to use microsoft/MediPhi-MedCode with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "microsoft/MediPhi-MedCode" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "microsoft/MediPhi-MedCode",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "microsoft/MediPhi-MedCode" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "microsoft/MediPhi-MedCode",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use microsoft/MediPhi-MedCode with Docker Model Runner:
```
docker model run hf.co/microsoft/MediPhi-MedCode
```

jpcorb20 commited on Jul 11, 2025

Commit

a9aeef0

verified ·

1 Parent(s): a802445

Update README.md

Browse files

Files changed (1) hide show

README.md +2 -14

README.md CHANGED Viewed

@@ -1,11 +1,5 @@
 ---
 license: mit
-datasets:
-- ncbi/pubmed
-- starmpcc/Asclepius-Synthetic-Clinical-Notes
-- akemiH/NoteChat
-- zhengyun21/PMC-Patients
-- jpcorb20/medical_wikipedia
 language:
 - en
 base_model:
@@ -26,7 +20,7 @@ The MediPhi Model Collection comprises 7 small language models of 3.8B parameter
 ## Model Details
 ### Model Description
-This model is `MediPhi` obtained by merging all 5 experts with the BreadCrumbs technique into this unified expert.
 - **Developed by:** Microsoft Healthcare \& Life Sciences
 - **Model type:** Phi3
@@ -90,7 +84,7 @@ Researchers should apply responsible AI best practices, including mapping, measu
     torch.random.manual_seed(0)
-    model_name = "microsoft/MediPhi"
     model = AutoModelForCausalLM.from_pretrained(
         model_name,
         device_map="cuda",
@@ -134,13 +128,7 @@ Check `microsoft/Phi-3.5-mini-instruct` for details about the tokenizer, require
 ### Training Data
 Continual Pre-training:
-- PubMed (commercial subset) and abstracts from `ncbi/pubmed`.
-- Medical Guideline `epfl-llm/guidelines`.
-- Medical Wikipedia `jpcorb20/medical_wikipedia`.
 - Medical Coding: ICD10CM, ICD10PROC, ICD9CM, ICD9PROC, and ATC.
-- Clinical documents:
-    - `zhengyun21/PMC-Patients`, `akemiH/NoteChat`, and `starmpcc/Asclepius-Synthetic-Clinical-Notes` (only commercial-friendly licenses across all three datasets)
-    - mtsamples
 See paper for details.

 ---
 license: mit
 language:
 - en
 base_model:
 ## Model Details
 ### Model Description
+This model is `MediPhi-MedCode` obtained by merging the fine-tuned MedCode expert with the SLERP technique into its base model at 50%.
 - **Developed by:** Microsoft Healthcare \& Life Sciences
 - **Model type:** Phi3
     torch.random.manual_seed(0)
+    model_name = "microsoft/MediPhi-MedCode"
     model = AutoModelForCausalLM.from_pretrained(
         model_name,
         device_map="cuda",
 ### Training Data
 Continual Pre-training:
 - Medical Coding: ICD10CM, ICD10PROC, ICD9CM, ICD9PROC, and ATC.
 See paper for details.