HealthDataHub/PARTAGES-Qwen3-4B-PDAPT-SLERP

Hugging Face
TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Dec 3, 2025License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Warm

HealthDataHub/PARTAGES-Qwen3-4B-PDAPT-SLERP is a 4 billion parameter causal language model developed by HealthDataHub, based on Qwen3-4B-Base. It was created through domain-adaptive pre-training (DAPT) on the French biomedical corpus PARCOMED, then fused using SLERP with the original base model to retain general capabilities. This model is specifically designed for biomedical applications in French healthcare settings, offering enhanced performance on medical tasks while maintaining general domain knowledge, and is licensed for commercial use.

Loading preview...

PARTAGES-Qwen3-4B-PDAPT-SLERP: French Biomedical Language Model

This model, developed by HealthDataHub, is a 4 billion parameter causal language model built upon Qwen3-4B-Base. It is specifically designed to serve as a foundational, pre-trained model for various tasks and use cases within healthcare institutions, particularly in French.

Key Capabilities

  • Domain Adaptation: Underwent Domain-Adaptive Pre-Training (DAPT) on PARCOMED, a comprehensive French biomedical corpus, enhancing its understanding of medical terminology and contexts.
  • Balanced Performance: Utilizes Spherical Linear Interpolation (SLERP) to merge the specialized DAPT checkpoint with the original Qwen3-4B-Base model. This fusion strategy mitigates the degradation of general language capabilities often seen in domain-specific models.
  • Commercial Use Compatible: Trained exclusively on commercially licensed data sources, making it suitable for both research and commercial applications.
  • French Biomedical Focus: Optimized for French medical tasks, showing improved accuracy on FR-MEDICAL benchmarks compared to the base model, across categories like Clinical Knowledge and Medical Genetics.

Good for

  • Base for Fine-tuning: Intended as a robust base model for further fine-tuning on specific downstream tasks within French healthcare.
  • Biomedical NLP in French: Applications requiring strong comprehension and generation of French medical text.
  • Commercial Healthcare Solutions: Development of AI tools for healthcare that require a commercially viable language model.
  • Maintaining General Knowledge: Use cases where domain specialization is needed without significant loss of broader language understanding.

Finetunes
1 models