tsor13/spectrum-Qwen3-14B-v1

The tsor13/spectrum-Qwen3-14B-v1 is a 14 billion parameter language model based on the Qwen3 architecture, specifically designed for "Spectrum Tuning." This post-training technique optimizes the model for distributional coverage and in-context steerability, allowing it to accurately match and sample from specified output distributions. It excels at tasks requiring precise probability estimation and controlled generation based on descriptions and few-shot examples, making it suitable for complex reasoning and data generation scenarios.

Warm
Public
14B
FP8
32768
Hugging Face

No reviews yet. Be the first to review!