| base_model: | |
| - Qwen-Qwen2.5-1.5B-Instruct | |
| library_name: transformers | |
| tags: | |
| - mergekit | |
| - merge | |
| pipeline_tag: text-generation | |
| # qwen25-instruct-with-gsm8k-reasoning | |
| This is a merge of pre-trained language models created using [mergekit](https://github.com/cg123/mergekit) as described in the paper "Reasoning Vectors: Transferring Chain-of-Thought Capabilities via Task Arithmetic". | |
| - Paper: [Reasoning Vectors: Transferring Chain-of-Thought Capabilities via Task Arithmetic](https://arxiv.org/abs/2509.01363) | |
| - Code: [https://github.com/rohitgandikota/erasing-llm](https://github.com/rohitgandikota/erasing-llm) | |
| - Project page: [https://elm.baulab.info](https://elm.baulab.info) | |
| ## Merge Details | |
| ### Merge Method | |
| This model was merged using the [Task Arithmetic](https://arxiv.org/abs/2212.04089) merge method using Qwen-Qwen2.5-1.5B-Instruct as a base. | |
| ### Models Merged | |
| The following models were included in the merge: | |
| * Qwen2.5-1.5B-gsm8k-sft-full | |
| * Qwen2.5-1.5B-gsm8k-grpo-full | |
| ### Configuration | |
| The following YAML configuration was used to produce this model: | |
| ```yaml | |
| base_model: Qwen-Qwen2.5-1.5B-Instruct | |
| models: | |
| - model: Qwen2.5-1.5B-gsm8k-grpo-full | |
| parameters: | |
| weight: 1.0 | |
| - model: Qwen2.5-1.5B-gsm8k-sft-full | |
| parameters: | |
| weight: -1.0 | |
| merge_method: task_arithmetic | |
| dtype: bfloat16 | |
| ``` |