OhhMoo/qwen05b-gsm8k-sft-instruct

TEXT GENERATIONConcurrency Cost:1Model Size:0.5BQuant:BF16Ctx Length:32kTool Calling:SupportedPublished:Jun 25, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

OhhMoo/qwen05b-gsm8k-sft-instruct is a 0.5 billion parameter Qwen2.5-0.5B-Instruct model, specifically fine-tuned by OhhMoo for mathematical reasoning tasks. This model excels at solving grade-school math problems, achieving 36.2% accuracy on the GSM8k test set. Its primary strength lies in step-by-step mathematical problem-solving, making it suitable for applications requiring numerical reasoning.

Loading preview...

OhhMoo/qwen05b-gsm8k-sft-instruct Overview

This model is a specialized 0.5 billion parameter variant of the Qwen2.5-0.5B-Instruct architecture, developed by OhhMoo. It has undergone a full supervised fine-tuning (SFT) process on the GSM8k dataset, which comprises 7,473 worked solutions for grade-school math problems.

Key Capabilities

  • Mathematical Reasoning: Optimized for solving arithmetic and word problems, demonstrating a strong ability to follow step-by-step reasoning.
  • GSM8k Performance: Achieves a test accuracy of 36.2% (477 out of 1319 examples) on the GSM8k benchmark using greedy decoding.
  • Instruction Following: Trained to adhere to a specific prompt format, requiring responses to end with #### <number> for the final numerical answer.

Training Details

The model was fine-tuned for 3 epochs (348 steps) using bf16 mixed precision, AdamW optimizer, and a maximum sequence length of 1024 tokens. The training data utilized a chat format, with the assistant's target being the full worked solution.

Good For

  • Applications requiring a compact model for mathematical problem-solving.
  • Educational tools or systems that need to generate step-by-step math solutions.
  • Scenarios where a small, specialized model for numerical reasoning is preferred over larger, general-purpose LLMs.

Finetunes
1 models