reaperdoesntknow/DistilQwen3-1.7B-uncensored

DistilQwen3-1.7B-uncensored Overview

This model, developed by Convergent Intelligence LLC: Research Division, is a 1.7 billion parameter language model from the DistilQwen3 Series. It is a product of a sophisticated distillation chain based on Discrepancy Calculus (DISC), a measure-theoretic framework that decomposes a teacher's output distribution to quantify local structural mismatch. Unlike standard KL divergence, DISC focuses on capturing nuanced structural differences.

Key Capabilities

• Proof-Weighted Knowledge Distillation: Utilizes a unique methodology with 55% cross-entropy (decaying proof weights from 2.5x to 1.5x) and 45% KL divergence (T=2.0). This amplifies loss on reasoning-critical tokens, compelling the student model to prioritize structural understanding.
• Optimized for Reasoning: Designed to allocate capacity towards logical inference, structured decomposition, and STEM derivation, rather than just surface-level pattern recognition.
• BF16 Training: Part of the only BF16 collection in the Convergent Intelligence portfolio, trained on H100 GPUs with a 30B-parameter teacher model, indicating a focus on high-fidelity training.

Good For

• Applications requiring strong logical inference and structured output.
• Tasks benefiting from extended deliberation and higher-entropy distributions.
• Use cases where proof derivation and STEM problem-solving are critical.
• Developers seeking a compact model with enhanced structural understanding derived from a larger, powerful teacher model.