shuoxing/qwen-3.5-27b-absorb

VISIONPricing:Input $1.6 / Cached $0.15 / Output $12Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 28, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

shuoxing/qwen-3.5-27b-absorb is a 27 billion parameter language model developed by shuoxing, based on Qwen/Qwen3.5-27B. It is fine-tuned using the Absorb self-distillation method, which leverages mathematical primitives as privileged information during training to enhance mathematical reasoning. This model is specifically optimized for solving complex mathematical proof problems, making it a specialized tool for advanced mathematical tasks. It operates as a drop-in replacement for the base model at inference time, requiring no primitive input.

Loading preview...

Overview

shuoxing/qwen-3.5-27b-absorb is a 27 billion parameter language model built upon Qwen/Qwen3.5-27B. Its key differentiator is the application of Absorb, a primitive-privileged self-distillation method. This technique involves training the model where a "teacher" (the base model) is given a mathematical primitive (a core conceptual observation) alongside a problem, while the "student" (the fine-tuned model) learns to generate its own solution. This process aims to improve the model's ability to diagnose and repair mathematical reasoning.

Key Capabilities

  • Enhanced Mathematical Reasoning: Specifically trained on 709 mathematics Ph.D. qualifying-exam proof problems from the shuoxing/math-phd-qual-709 dataset.
  • Self-Distillation with Primitives: Utilizes a unique self-distillation approach where the teacher and student are the same model, with the teacher receiving privileged primitive information during training.
  • Drop-in Replacement: At inference time, the model operates without requiring any primitive input, functioning as a direct replacement for the base Qwen3.5-27B model.
  • Text-Only Training: While the base model has a vision tower, training and evaluation for qwen-3.5-27b-absorb are text-only.

Good for

  • Advanced Mathematical Problem Solving: Ideal for applications requiring robust mathematical proof generation and complex reasoning.
  • Research in LLM Reasoning: Useful for researchers exploring methods to improve mathematical and logical reasoning in large language models.
  • Specialized AI Assistants: Can be integrated into systems that need to tackle challenging mathematical or scientific problems.

Finetunes
1 models