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

TEXT GENERATIONPricing:Input $0.32 / Cached $0.064 / Output $1.6Concurrent Unit Cost:1Model Size:2BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Mar 25, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

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

Loading preview...

Overview

reaperdoesntknow/Qwen3-1.7B-Coder-Distilled-SFT is a 1.7 billion parameter model from Convergent Intelligence LLC, designed for advanced logical inference and STEM reasoning. It leverages a unique two-stage training process: first, knowledge distillation from a 30B Qwen3-Coder teacher to instill a structured reasoning backbone, and second, supervised fine-tuning on a large dataset of logical inference problems. This methodology aims to activate latent sequential logic, state tracking, and compositional reasoning capabilities derived from the Coder teacher.

Key Capabilities

  • Logical Inference: Proficient in propositional logic, formal reasoning, and logical entailment.
  • STEM Derivation: Capable of rigorous derivations in scientific, technical, engineering, and mathematics domains.
  • Structured Argumentation: Designed to produce structured and explicit reasoning steps.
  • Efficient Deployment: Available in GGUF format for edge deployment.

Good For

  • Educational Tutoring: Assisting with logical and mathematical problem-solving.
  • Verification Pipelines: Serving as a component for checking logical consistency.
  • Research in Formal Reasoning: Exploring the application of LLMs to structured thought processes.
  • Applications requiring explicit reasoning paths: Where not just the answer, but the derivation, is important.

This model is part of the DistilQwen collection, which utilizes proof-weighted knowledge distillation and was trained on premium hardware (H100 at BF16) to enhance structural understanding over surface-level pattern matching. Its training context length is 1024 tokens. While strong in logical inference, it is not intended for general code generation or complex multi-step inferences with many quantifiers beyond its training scope.

Finetunes
2 models