---
base_model:
- Qwen/Qwen2.5-14B
- Qwen/Qwen2.5-14B-Instruct
- CultriX/Qwen2.5-14B-Coder_r128
- CultriX/Qwen2.5-14B_Virtuoso-small-v2-LoRA_r128
- CultriX/Qwen2.5-14B-DeepSeek_r128
- CultriX/Qwen2.5-14B-Hyperionv3_r128
- CultriX/Qwen2.5-14B-SuperNova-Medius_r128
library_name: transformers
tags:
- mergekit
- merge

---
# merge

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 [DARE TIES](https://arxiv.org/abs/2311.03099) merge method using [Qwen/Qwen2.5-14B](https://huggingface.co/Qwen/Qwen2.5-14B) as a base.

### Models Merged

The following models were included in the merge:
* [Qwen/Qwen2.5-14B-Instruct](https://huggingface.co/Qwen/Qwen2.5-14B-Instruct)
* [CultriX/Qwen2.5-14B-Coder_r128](https://huggingface.co/CultriX/Qwen2.5-14B-Coder_r128)
* [CultriX/Qwen2.5-14B_Virtuoso-small-v2-LoRA_r128](https://huggingface.co/CultriX/Qwen2.5-14B_Virtuoso-small-v2-LoRA_r128)
* [CultriX/Qwen2.5-14B-DeepSeek_r128](https://huggingface.co/CultriX/Qwen2.5-14B-DeepSeek_r128)
* [CultriX/Qwen2.5-14B-Hyperionv3_r128](https://huggingface.co/CultriX/Qwen2.5-14B-Hyperionv3_r128)
* [CultriX/Qwen2.5-14B-SuperNova-Medius_r128](https://huggingface.co/CultriX/Qwen2.5-14B-SuperNova-Medius_r128)

### Configuration

The following YAML configuration was used to produce this model:

```yaml
# Optimized MergeKit configuration for merging extracted LoRA adapters
# into the Qwen2.5-14B-Instruct model.
base_model: Qwen/Qwen2.5-14B

models:
  # Each adapter was extracted (rank=128) from its respective finetuned model.
  # Their weights are set lower than the full instruct model (which is now the base)
  - model: CultriX/Qwen2.5-14B-Hyperionv3_r128
    parameters:
      weight: 0.15      # Reduced weight relative to base
      density: 1.0
      lora_rank: 128    # Mark as extracted LoRA adapter

  - model: CultriX/Qwen2.5-14B-Coder_r128
    parameters:
      weight: 0.15
      density: 1.0
      lora_rank: 128

  - model: CultriX/Qwen2.5-14B_Virtuoso-small-v2-LoRA_r128
    parameters:
      weight: 0.15
      density: 1.0
      lora_rank: 128

  - model: CultriX/Qwen2.5-14B-SuperNova-Medius_r128
    parameters:
      weight: 0.15
      density: 1.0
      lora_rank: 128

  - model: CultriX/Qwen2.5-14B-DeepSeek_r128
    parameters:
      weight: 0.15
      density: 1.0
      lora_rank: 128

# (Optionally, if you wish to “re-add” a full instruct copy you could include it here 
#  with a higher weight—but note that Qwen2.5-14B-Instruct is already the base.)
  - model: Qwen/Qwen2.5-14B-Instruct
    parameters:
      weight: 0.40
      density: 1.0

# Merging method and overall parameters
merge_method: dare_ties         # Ties corresponding weights across sources.
parameters:
  weight: 1.0                 # Overall scaling factor.
  density: 1.0                # Overall density (typically left at 1.0).
  normalize: true             # Normalize each set of weights before merging.
  int8_mask: true             # Enable masking if using int8 quantized weights.

# Use the instruct tokenizer to ensure compatibility.
tokenizer_source: CultriX/Qwen2.5-14B_Virtuoso-small-v2-LoRA_r128

# Data type for merged weights.
dtype: bfloat16

```