reaperdoesntknow/Qwen3-1.7B-Coder-Distilled-SFT

TEXT GENERATIONConcurrency Cost:1Model Size:2BQuant:BF16Ctx Length:32kTool Calling:SupportedPublished:Mar 25, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

reaperdoesntknow/Qwen3-1.7B-Coder-Distilled-SFT is a 1.7 billion parameter Qwen3-based causal language model developed by Convergent Intelligence LLC. This model is uniquely trained through knowledge distillation from a 30B Coder teacher, followed by supervised fine-tuning on over 54,600 logical inference problems. It excels at structured reasoning, STEM derivation, and formal propositional logic, making it suitable for tasks requiring precise sequential logic and compositional decomposition.

Loading preview...

Model Overview

reaperdoesntknow/Qwen3-1.7B-Coder-Distilled-SFT is a 1.7 billion parameter model built on the Qwen3 architecture, developed by Convergent Intelligence LLC. Its unique training pipeline involves two stages: first, knowledge distillation from a 30B Coder teacher model to establish a strong STEM reasoning backbone, and second, supervised fine-tuning on a large dataset of logical inference problems. This approach aims to activate latent structural reasoning capabilities derived from the Coder teacher, making implicit logical structures explicit.

Key Capabilities

  • Structured Reasoning: Inherits precise sequential logic, state tracking, and compositional decomposition patterns from its Coder teacher.
  • STEM Derivation: Trained on over 6,000 STEM chain-of-thought samples across 12 domains, enabling rigorous derivations.
  • Logical Inference: Fine-tuned on ~54,600 instruction-response pairs covering propositional logic and formal inference, enhancing its ability to perform logical entailment.
  • Efficient Size: At 1.7B parameters, it offers advanced reasoning capabilities in a compact form factor.

Training Methodology

The model utilizes a proof-weighted knowledge distillation method (55% cross-entropy with decaying proof weights, 45% KL divergence at T=2.0) to transfer the teacher's full probability landscape, including its reasoning organization. The logical inference SFT stage further refines these capabilities, focusing on formal propositional logic.

Good for

  • Logical inference and propositional logic tasks.
  • Formal reasoning and structured argumentation.
  • STEM derivation and educational tutoring applications.
  • Component in verification pipelines.
  • Edge deployment via GGUF quantized versions.

Finetunes
2 models