Odark007/

qwen2.5-1.5b-blindspots

	Blind Spots of Qwen2.5-1.5B (Backbone of VibeVoice-1.5B)




	
	Model Evaluated

This dataset evaluates Qwen/Qwen2.5-1.5B, a 1.5B-parameter base language model. Qwen2.5-1.5B serves as the backbone language model for VibeVoice-1.5B, a frontier open-source text-to-speech system that integrates an LLM with acoustic tokenizers and a diffusion decoding head. Because VibeVoice relies on Qwen2.5-1.5B for textual reasoning and dialogue modeling, blind spots in the base LLM can propagate… See the full description on the dataset page: https://huggingface.co/datasets/Odark007/qwen2.5-1.5b-blindspots.