apertus-1.5-8b-biomedical-ade API & Inference Endpoint

Example Prompts & Expected Outputs

Example 1: Multi-Drug Toxicity Detection

Prompt:

text
You are an expert biomedical information extraction system. Extract all mentioned drugs and adverse drug events (ADEs) from the clinical narrative as JSON.

Clinical Narrative:
"A 64-year-old female patient with chronic atrial fibrillation presented with severe bradycardia, visual yellow halos, and nausea after increasing her digoxin dosage to 0.25 mg daily. Electrocardiogram confirmed digitalis toxicity. Concurrently, she had been prescribed amiodarone two weeks earlier."

Model Output:

json
{
  "drugs": ["digoxin", "amiodarone"],
  "adverse_drug_events": ["severe bradycardia", "visual yellow halos", "nausea"]
}

Example 2: Negative Control (No Adverse Effects)

Prompt:

text
You are an expert biomedical information extraction system. Extract all mentioned drugs and adverse drug events (ADEs) from the clinical narrative as JSON.

Clinical Narrative:
"The patient was initiated on metformin 500 mg twice daily for type 2 diabetes mellitus and reported good glycemic control without any adverse symptoms."

Model Output:

json
{
  "drugs": ["metformin"],
  "adverse_drug_events": []
}

Quickstart (vLLM / Transformers / Ollama)

Direct Ingestion with vLLM

bash
vllm serve andreasmartin/apertus-1.5-8b-biomedical-ade --tensor-parallel-size 1 --gpu-memory-utilization 0.9

Direct Python Ingestion

python
from transformers import AutoModelForCausalLM, AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained("andreasmartin/apertus-1.5-8b-biomedical-ade")
model = AutoModelForCausalLM.from_pretrained("andreasmartin/apertus-1.5-8b-biomedical-ade", device_map="auto")

Training Configuration & Metrics

Table with columns: Parameter, Value
Parameter	Value
Base Model	`andreasmartin/apertus-v1.5-8b-text`
Dataset	`mihirhirave/entity_extraction_ade_v2_with_validation`
Dataset Slice	`train[:600]` (Train) / `validation[:100]` (Validation)
Optimizer Steps	60

Example Prompts & Expected Outputs

Example 1: Multi-Drug Toxicity Detection

Prompt:

text
You are an expert biomedical information extraction system. Extract all mentioned drugs and adverse drug events (ADEs) from the clinical narrative as JSON.

Clinical Narrative:
"A 64-year-old female patient with chronic atrial fibrillation presented with severe bradycardia, visual yellow halos, and nausea after increasing her digoxin dosage to 0.25 mg daily. Electrocardiogram confirmed digitalis toxicity. Concurrently, she had been prescribed amiodarone two weeks earlier."

Model Output:

json
{
  "drugs": ["digoxin", "amiodarone"],
  "adverse_drug_events": ["severe bradycardia", "visual yellow halos", "nausea"]
}

Example 2: Negative Control (No Adverse Effects)

Prompt:

text
You are an expert biomedical information extraction system. Extract all mentioned drugs and adverse drug events (ADEs) from the clinical narrative as JSON.

Clinical Narrative:
"The patient was initiated on metformin 500 mg twice daily for type 2 diabetes mellitus and reported good glycemic control without any adverse symptoms."

Model Output:

json
{
  "drugs": ["metformin"],
  "adverse_drug_events": []
}

Quickstart (vLLM / Transformers / Ollama)

Direct Ingestion with vLLM

bash
vllm serve andreasmartin/apertus-1.5-8b-biomedical-ade --tensor-parallel-size 1 --gpu-memory-utilization 0.9

Direct Python Ingestion

python
from transformers import AutoModelForCausalLM, AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained("andreasmartin/apertus-1.5-8b-biomedical-ade")
model = AutoModelForCausalLM.from_pretrained("andreasmartin/apertus-1.5-8b-biomedical-ade", device_map="auto")

Training Configuration & Metrics

Table with columns: Parameter, Value
Parameter	Value
Base Model	`andreasmartin/apertus-v1.5-8b-text`
Dataset	`mihirhirave/entity_extraction_ade_v2_with_validation`
Dataset Slice	`train[:600]` (Train) / `validation[:100]` (Validation)
Optimizer Steps	60

apertus-1.5-8b-biomedical-ade

README

Example Prompts & Expected Outputs

Example 1: Multi-Drug Toxicity Detection

Example 2: Negative Control (No Adverse Effects)

Quickstart (vLLM / Transformers / Ollama)

Direct Ingestion with vLLM

Direct Python Ingestion

Training Configuration & Metrics

Explore FriendliAI today

README

Example Prompts & Expected Outputs

Example 1: Multi-Drug Toxicity Detection

Example 2: Negative Control (No Adverse Effects)

Quickstart (vLLM / Transformers / Ollama)

Direct Ingestion with vLLM

Direct Python Ingestion

Training Configuration & Metrics