suryadv/IncorrectTraceSFT-Qwen2.5-7B-MATH-random-mask-90

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:7.6BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 15, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

suryadv/IncorrectTraceSFT-Qwen2.5-7B-MATH-random-mask-90 is a 7.6 billion parameter Qwen2.5-7B model fine-tuned for mathematical reasoning. Developed by suryadv, this specific checkpoint, 'random_mask_90', is part of a study on how incorrect traces influence supervised fine-tuning. It is optimized for research into mathematical problem-solving and SFT data selection strategies.

Loading preview...

Model Overview

This model, IncorrectTraceSFT-Qwen2.5-7B-MATH-random-mask-90, is a 7.6 billion parameter variant of the Qwen2.5-7B architecture. It was developed by suryadv as part of research into the impact of incorrect traces during supervised fine-tuning (SFT), specifically focusing on mathematical tasks. This particular checkpoint represents the 'random_mask_90' condition from the 'math masking' experiment detailed in the associated research paper.

Key Characteristics

  • Base Model: Fine-tuned from Qwen/Qwen2.5-7B.
  • Training Method: Utilizes SLIME with Megatron-LM, BF16, and AdamW optimizer.
  • Context Length: Supports a context length of 32768 tokens.
  • Research Focus: Designed for studying mathematical reasoning and the effects of SFT data selection, particularly concerning incorrect intermediate steps.

Intended Use

This model is primarily released for research purposes related to mathematical reasoning and supervised fine-tuning data selection. It allows researchers to investigate how different masking strategies for incorrect traces influence model performance in mathematical problem-solving. Evaluation can be performed using the provided code repository to assess pass@1 metrics and Monte Carlo standard error.