PeterPaker123/Qwen2.5-7B-Vietnamese-Medical-NER-GRPO

TEXT GENERATIONConcurrency Cost:1Model Size:7.6BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:Jun 4, 2026Architecture:Transformer Cold

PeterPaker123/Qwen2.5-7B-Vietnamese-Medical-NER-GRPO is a 7.6 billion parameter Qwen2.5-7B-Instruct based causal language model, strictly fine-tuned for Vietnamese Medical Named Entity Recognition (NER). It excels at extracting medical entities like Symptoms, Diseases, and Drugs from Vietnamese texts, outputting them in a structured JSON format. This model uniquely supports dynamic, prompt-defined entity extraction, acting as an intelligent agent for specific medical information retrieval.

Loading preview...

Overview

This model, developed by PeterPaker123, is a specialized generative Large Language Model built upon the Qwen2.5-7B architecture. Unlike traditional token-classification models, it functions as an intelligent extraction agent for Vietnamese Medical Named Entity Recognition (NER), outputting structured JSON lists from medical texts.

Key Capabilities

  • Dynamic Entity Extraction: Supports prompt-defined entity types, allowing users to specify any medical or clinical entity for extraction, such as SYMPTOM_AND_DISEASE, MEDICAL_PROCEDURE, and DRUG.
  • Structured Output: Delivers extracted entities in a strict JSON format, making it ideal for downstream processing.
  • Advanced Fine-tuning: Utilizes a two-stage alignment pipeline including Supervised Fine-Tuning (SFT) and Group Relative Policy Optimization (GRPO) with Reinforcement Learning (RL) to achieve high precision, recall, and type accuracy, significantly reducing hallucinations.
  • Vietnamese Language Focus: Exclusively trained and optimized for the Vietnamese language.

Good For

  • Medical Chatbot Enrichment: Enhancing RAG pipelines for healthcare applications.
  • Automated Summarization: Summarizing telehealth transcripts by extracting key medical information.
  • Clinical Knowledge Graph Construction: Building structured knowledge bases from unstructured medical text.

Limitations

  • Not for Diagnosis: Strictly an extraction tool; it does not provide medical advice or diagnoses.
  • No General Conversation: Aggressively aligned for JSON output, performing poorly on general conversational tasks.
  • Specific Entity Focus: Trained to ignore demographic data (Names, Ages, Locations) and will likely yield empty lists for such requests.
  • Hallucination Residuals: While reduced, generative LLMs may still occasionally hallucinate or alter extracted text.

Finetunes
1 models