lulululuyi/R1-Qwen-7B-R_HORIZON_Compose_2

Model Description

R1-Qwen-7B-R_HORIZON_Compose_2 is a reasoning model used in the paper:

“R-Horizon: How Far Can Your Large Reasoning Model Really Go in Breadth and Depth?”

This model corresponds to the Naive Training Data (n = 2) configuration reported in Table 1 of the paper.

The model is based on R1-distill-Qwen-7B and is trained on naively composed reasoning queries with composition depth n = 2. Each training sample is constructed by composing four reasoning queries into a single example, following the naive training data strategy introduced in the R-Horizon study.