README.md

---
base_model:
- unsloth/DeepSeek-R1-Distill-Qwen-7B
- Qwen/Qwen2.5-Math-7B-Instruct
- nvidia/AceMath-7B-Instruct
library_name: transformers
tags:
- mergekit
- merge

---
# merged-model-ACE

This is a merge of pre-trained language models created using [mergekit](https://github.com/cg123/mergekit).

## Merge Details
### Merge Method

This model was merged using the [TIES](https://arxiv.org/abs/2306.01708) merge method using [unsloth/DeepSeek-R1-Distill-Qwen-7B](https://huggingface.co/unsloth/DeepSeek-R1-Distill-Qwen-7B) as a base.

### Models Merged

The following models were included in the merge:
* [Qwen/Qwen2.5-Math-7B-Instruct](https://huggingface.co/Qwen/Qwen2.5-Math-7B-Instruct)
* [nvidia/AceMath-7B-Instruct](https://huggingface.co/nvidia/AceMath-7B-Instruct)

### Configuration

The following YAML configuration was used to produce this model:

```yaml
# merge_ties.yml

# 1. Overall merge method: TIES (sign-elect sparse task arithmetic)
merge_method: ties                                      

# 2. Base model (all task vectors are computed relative to this checkpoint)
base_model: unsloth/DeepSeek-R1-Distill-Qwen-7B   

# 3. Full models to merge (base first, then others)
models:
  - model: unsloth/DeepSeek-R1-Distill-Qwen-7B       # base has no extra params
  - model: nvidia/AceMath-7B-Instruct
    parameters:
      weight: 0.7
      density: 0.7
  - model: Qwen/Qwen2.5-Math-7B-Instruct
    parameters:
      weight: 0.3
      density: 0.7

# 4. Global merge parameters
parameters:
  normalize: true        # normalize weights across models
  int8_mask: true        # mask small values when using int8 backing

# 5. Data type for merged tensors
dtype: bfloat16

```