ModelVerifiedExternal

Triage Hospital Agent

A GRPO fine-tuned version of Qwen3.5-4B specialized for hospital crisis management and clinical triage decision-making, trained as part of the TRIAGE multi-agent system.

Sourcehuggingface.co/balarajr/triage-hospital-agent/tree/86fdf1a21c69d71ab27c57c055a0e645a62eb65a

KindLoRABase modelQwen/Qwen3.5-4BVersionRevision 86fdf1aPublisher@balarajrCgrade

Model source

Kind: LoRA
File size: 465 MB–1.86 GB
Base model: Qwen/Qwen3.5-4B
Version: vRevision 86fdf1a
Updated: Oct 2, 2026
Publisher: @balarajr
Downloads: 113
Parameters: 4B
Tasks: Text generation
Source: Hugging Face

Source description

TRIAGE — Hospital Crisis Agent (Qwen3.5-4B GRPO)

A GRPO fine-tuned version of Qwen3.5-4B specialized for hospital crisis management and clinical triage decision-making, trained as part of the TRIAGE multi-agent system.

Model Description

This model serves as the backbone for a 6-agent hospital crisis simulation that coordinates:

🚑 ER Triage Agent — Patient severity classification (START protocol)
🏥 ICU Management Agent — Bed allocation and overflow protocols
💊 Pharmacy Agent — Drug order validation and contraindication detection
👩‍⚕️ HR Rostering Agent — Emergency staff deployment
💻 IT Systems Agent — EHR integrity and system failure response
🎯 CMO Oversight Agent — Override decisions and crisis governance

Benchmark Results (TRIAGE Multi-Agent Benchmark)

Scenario	Survival Rate	Violation Detection	Reward
Mass Casualty	100%	100%	10.0/10.0
Disease Outbreak	100%	100%	10.0/10.0
Equipment Failure	100%	100%	10.0/10.0
Staff Shortage	100%	100%	10.0/10.0
Combined Surge	100%	100%	10.0/10.0

Composite Score: 87.33/100 [A]
(Conservative — 20-step episodes; 50-step runs expected to yield 92+)

Comparison to Existing Work

System	Model Size	Hospital Ops	RL Environment	Score
TRIAGE (this model)	4B	✅ Full 6-agent	✅ OpenEnv	87.3+
MedAgents (ACL 2024)	GPT-4 (1T+)	❌ QA only	❌ No env	N/A
Gemini 2.5 Flash	Undisclosed	❌ Single-agent	❌ No env	73.8% ESI

Training Details

Parameter	Value
Base model	Qwen/Qwen3.5-4B
Training method	GRPO (Generative Reward Policy Optimization)
LoRA rank	16
LoRA alpha	16
Quantization	4-bit NF4 (bitsandbytes)
Training hardware	NVIDIA T4 / P100 (16GB VRAM)
Dataset	300 highly curated prompts
Reward Verifiers	8 custom medical verifiers
Epochs	1
Optimizer	paged_adamw_8bit

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained(
    "user/triage-qwen-4b-grpo",
    trust_remote_code=True,
)
tokenizer = AutoTokenizer.from_pretrained("user/triage-qwen-4b-grpo")

prompt = """Hospital Crisis Management System — Step 15
Crisis: mass_casualty | ICU: 45/60 beds | Critical patients: 8
Patients — Critical: 8, Untreated Critical: 3

What is the correct triage action?"""

inputs = tokenizer(prompt, return_tensors="pt")
output = model.generate(**inputs, max_new_tokens=150, temperature=0.1)
print(tokenizer.decode(output[0], skip_special_tokens=True))

Limitations

For research and simulation purposes only
Not validated for real clinical deployment
Accuracy depends on prompt quality and crisis scenario complexity
Should not replace professional medical judgment

Citation

@software{triage2025,
  title={TRIAGE: Multi-Agent Hospital Crisis Simulation with DPO Fine-tuning},
  year={2025},
  note={Meta PyTorch OpenEnv Hackathon submission},
  url={https://github.com/YOUR_USERNAME/triage}
}

License

Apache 2.0 — see LICENSE file.

Sources

1 source

Hugging FacePrimaryVersion 86fdf1a21c69d71ab27c57c055a0e645a62eb65a

huggingface.co/balarajr/triage-hospital-agent/tree/86fdf1a21c69d71ab27c57c055a0e645a62eb65a

Verified Oct 2

Model artifacts

5 artifacts

model-00001-of-00005.safetensors

safetensors · 1.86 GB · SHA-256 ef6502cafb7d…8f96 · Hugging Face

Download

model-00002-of-00005.safetensors

safetensors · 1.86 GB · SHA-256 4f1186026085…893d · Hugging Face

Download

model-00003-of-00005.safetensors

safetensors · 1.83 GB · SHA-256 8e0eb864a05b…d036 · Hugging Face

Download

model-00004-of-00005.safetensors

safetensors · 1.83 GB · SHA-256 16853fbe6f4d…21aa · Hugging Face

Download

model-00005-of-00005.safetensors

safetensors · 465 MB · SHA-256 b72ea67aa577…d85d · Hugging Face

Download

Source excerpts

3 excerpts

ReadmeJul 11huggingface.co/balarajr/triage-hospital-agent

--- language: - en license: apache-2.0 base_model: Qwen/Qwen3.5-4B tags: - medical - triage - hospital - multi-agent - grpo - lora - qwen - clinical-ai - crisis-management datasets: - openlifescienceai/medmcqa - bigbio/med_qa pipeline_tag: text-generation --- # TRIAGE — Hospital Crisis Agent (Qwen3.5-4B GRPO) A **GRPO fine-tuned** version of `Qwen3.5-4B` specialized for **hospital crisis management** and **clinical triage decision-making**, trained as part of the TRIAGE multi-agent system. ## Model Description This model serves as the backbone for a **6-agent hospital crisis simulation** that coordinates: - 🚑 **ER Triage Agent** — Patient severity classification (START protocol) - 🏥 **ICU Management Agent** — Bed allocation and overflow protocols - 💊 **Pharmacy Agent** — Drug order validation and contraindication detection - 👩‍⚕️ **HR Rostering Agent** — Emergency staff deployment - 💻 **IT Systems Agent** — EHR integrity and system failure response - 🎯 **CMO Oversight Agent** — Override decisions and crisis governance ## Benchmark Results (TRIAGE Multi-Agent Benchmark) | Scenario | Survival Rate | Violation Detection | Reward | |---|---|---|---| | Mass Casualty | 100% | 100% | 10.0/10.0 | | Disease Outbreak | 100% | 100% | 10.0/10.0 | | Equipment Failure | 100% | 100% | 10.0/10.0 | | Staff Shortage | 100% | 100% | 10.0/10.0 | | Combined Surge | 100% | 100% | 10.0/10.0 | **Composite Score: 87.33/100 [A]** *(Conservative — 20-step episodes; 50-step runs expected to yield 92+)* ### Comparison to Existing Work | System | Model Size | Hospital Ops | RL Environment | Score | |---|---|---|---|---| | **TRIAGE (this model)** | **4B** | **✅ Full 6-agent** | **✅ OpenEnv** | **87.3+** | | MedAgents (ACL 2024) | GPT-4 (1T+) | ❌ QA only | ❌ No env | N/A | | Gemini 2.5 Flash | Undisclosed | ❌ Single-agent | ❌ No env | 73.8% ESI | ## Training Details | Parameter | Value | |---|---| | Base model | Qwen/Qwen3.5-4B | | Training method | GRPO (Generative Reward Policy Optimization) | | LoRA rank | 16 | | LoRA alpha | 16 | | Quantization | 4-bit NF4 (bitsandbytes) | | Training hardware | NVIDIA T4 / P100 (16GB VRAM) | | Dataset | 300 highly curated prompts | | Reward Verifiers | 8 custom medical verifiers | | Epochs | 1 | | Optimizer | paged_adamw_8bit | ## Usage ```python from transformers import AutoModelForCausalLM, AutoTokenizer model = AutoModelForCau...

NoteOct 2huggingface.co/balarajr/triage-hospital-agent

Source context: 113 downloads · 1 likes · Pipeline text-generation · Repo balarajr/triage-hospital-agent

NoteJul 16huggingface.co/balarajr/triage-hospital-agent

Source context: 18 downloads · 0 likes · Pipeline text-generation · Repo balarajr/triage-hospital-agent