reaperdoesntknow/DistilQwen3-1.7B-uncensored

Model Overview

reaperdoesntknow/DistilQwen3-1.7B-uncensored is a 1.7 billion parameter model developed by Convergent Intelligence LLC: Research Division, forming part of their DistilQwen3 series. This model is a product of a sophisticated distillation process rooted in Discrepancy Calculus (DISC), a measure-theoretic framework. DISC aims to decompose a teacher's output distribution to quantify local structural mismatches, which standard KL divergence might overlook. The underlying theory, "On the Formal Analysis of Discrepancy Calculus" (Colca, 2026), emphasizes structural understanding over surface-level pattern matching.

Key Capabilities & Methodology

• Proof-Weighted Distillation: The model utilizes a unique proof-weighted knowledge distillation method, combining 55% cross-entropy with decaying proof weights (2.5x to 1.5x) and 45% KL divergence at T=2.0. This approach amplifies loss on reasoning-critical tokens, compelling the student model to prioritize structural understanding.
• Teacher Model: Distilled from a powerful 30B-parameter Qwen3-30B-A3B (Instruct) teacher, ensuring high-quality knowledge transfer.
• Hardware & Precision: Unlike other models in the broader Convergent Intelligence catalog, the DistilQwen series was trained on H100 GPUs at BF16 precision, indicating a focus on leveraging premium compute for enhanced performance.

Use Cases

This model is particularly well-suited for tasks requiring:

• Instruction Following: Excelling at adhering to complex instructions.
• Structured Output: Generating responses in specific, predefined formats.
• Legal Reasoning: Demonstrating capabilities in tasks involving legal analysis and inference.