Swift Qwen3.8 27B
A 27-billion-parameter open-source large language model fine-tuned by UkisAI on the Qwen3 architecture. It focuses on addressing the over-thinking problem in reasoning models, reducing token consumption while maintaining reasoning accuracy through efficient thinking training.
Timeline (last 90 days)
Swift Qwen3.8 27B downloads have surpassed 100,000.
Swift Qwen3.8 27B has topped the HuggingFace Trending list, ranking first among fine-tuned models and ninth among all models overall.
Swift Qwen3.8 27B is UkisAI's first open-source model.
The approach adopted by UkisAI does not directly train the model to shorten its thinking, but instead trains it to think more efficiently, in order to address the overthinking problem in small LLMs.
This approach reduces token usage by 58.3% and improves inference speed by 1.95×, with no loss in accuracy.
Quantization contributor bartowski has released a corresponding GGUF version for Swift Qwen3.8.
The original model can be downloaded at huggingface.co/ukisai/Swift-Qwen3.8-27b.
All Facts (5)
The original model can be downloaded at huggingface.co/ukisai/Swift-Qwen3.8-27b.
60%UnverifiedSwift Qwen3.8 27B is UkisAI's first open-source model.
50%UnverifiedThe approach adopted by UkisAI does not directly train the model to shorten its thinking, but instead trains it to think more efficiently, in order to address the overthinking problem in small LLMs.
50%UnverifiedThis approach reduces token usage by 58.3% and improves inference speed by 1.95×, with no loss in accuracy.
50%UnverifiedQuantization contributor bartowski has released a corresponding GGUF version for Swift Qwen3.8.
50%