L1nus/qwen3-4b-pubmedqa-thinking-no-ctx-default

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:4BQuant:BF16Ctx Length:32kPublished:May 27, 2026License:apache-2.0Architecture:Transformer Open Weights Warm

L1nus/qwen3-4b-pubmedqa-thinking-no-ctx-default is a 4 billion parameter Qwen3 model developed by L1nus, fine-tuned from unsloth/Qwen3-4B. This model was trained with Unsloth, enabling a 2x faster training process. It is designed for specific applications, likely related to PubMed QA or similar reasoning tasks, given its naming convention.

Loading preview...

Model Overview

L1nus/qwen3-4b-pubmedqa-thinking-no-ctx-default is a 4 billion parameter model based on the Qwen3 architecture, developed by L1nus. It was fine-tuned from the unsloth/Qwen3-4B base model.

Key Characteristics

  • Architecture: Qwen3
  • Parameter Count: 4 billion parameters
  • Training Efficiency: This model was trained 2x faster using the Unsloth framework, which specializes in accelerating large language model training.
  • License: The model is released under the Apache-2.0 license.

Potential Use Cases

While specific use cases are not detailed in the provided README, the model's name, pubmedqa-thinking-no-ctx-default, suggests it is likely optimized for:

  • Question Answering (QA) tasks, particularly within the biomedical domain (e.g., PubMed).
  • Reasoning capabilities, potentially without relying heavily on external context during inference.

Developers looking for a Qwen3-based model with efficient training and a focus on specialized QA or reasoning tasks in specific domains may find this model suitable.