bbanany/qwen25-3b-korean-pii-merged

TEXT GENERATIONPricing:Input $0.32 / Cached $0.064 / Output $1.6Concurrent Unit Cost:1Model Size:3.1BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Aug 8, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The bbanany/qwen25-3b-korean-pii-merged model is a 3.1 billion parameter language model based on the Qwen2.5-3B-Instruct architecture, developed by bbanany. This merged checkpoint integrates a LoRA adapter for Korean PII detection directly into the base model, allowing for direct loading without PEFT or bitsandbytes. It is specifically designed to identify Personal Identifiable Information (PII) in Korean text, outputting 'PII' or 'NOT_PII' classifications. The model is optimized for efficient deployment in environments like vLLM, CPU, and Apple Silicon.

Loading preview...

Model Overview

The bbanany/qwen25-3b-korean-pii-merged is a specialized 3.1 billion parameter language model built upon the Qwen/Qwen2.5-3B-Instruct base. This model is a standalone Hugging Face checkpoint, created by merging a LoRA adapter (bbanany/qwen25-3b-korean-pii-qlora2) directly into the FP16 base model using merge_and_unload(). This integration means it can be loaded and used directly with standard Transformers and vLLM without requiring PEFT adapters, bitsandbytes, NF4, or NVIDIA GPUs.

Key Capabilities

  • Korean PII Detection: The primary function of this model is to identify Personal Identifiable Information within Korean text. It processes JSON-formatted user messages containing text and potential PII candidates.
  • Direct Deployment: Designed for ease of use, it can be loaded directly into inference frameworks like vLLM, making it suitable for production environments.
  • CPU/Apple Silicon Compatibility: For lightweight execution on CPU or Apple Silicon devices, a GGUF distribution is available at bbanany/qwen25-3b-korean-pii-gguf.

Usage and Output

Users should apply the Qwen chat template from the original model card and embed a specific JSON string as the user message. The model's output will be either PII or NOT_PII, indicating whether personal information was detected. It's important to note that for final PII processing, independent evaluation and human review are recommended.

Licensing

The model operates under the Apache-2.0 license, with the terms of the base model also applying.

Finetunes
1 models