113 lines
4.7 KiB
Markdown
113 lines
4.7 KiB
Markdown
|
---
|
||
|
|
base_model:
|
||
|
|
- qingy2019/Qwen2.5-Math-14B-Instruct
|
||
|
|
- djuna/Q2.5-Veltha-14B-0.5
|
||
|
|
- CultriX/Qwen2.5-14B-Wernickev3
|
||
|
|
- CultriX/Qwenfinity-2.5-14B
|
||
|
|
- allknowingroger/QwenSlerp6-14B
|
||
|
|
- CultriX/Qwen2.5-14B-Broca
|
||
|
|
- sometimesanotion/Qwen2.5-14B-Vimarckoso
|
||
|
|
- CultriX/SeQwence-14Bv1
|
||
|
|
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 della_linear merge method using [CultriX/Qwen2.5-14B-Wernickev3](https://huggingface.co/CultriX/Qwen2.5-14B-Wernickev3) as a base.
|
||
|
|
|
||
|
|
### Models Merged
|
||
|
|
|
||
|
|
The following models were included in the merge:
|
||
|
|
* [qingy2019/Qwen2.5-Math-14B-Instruct](https://huggingface.co/qingy2019/Qwen2.5-Math-14B-Instruct)
|
||
|
|
* [djuna/Q2.5-Veltha-14B-0.5](https://huggingface.co/djuna/Q2.5-Veltha-14B-0.5)
|
||
|
|
* [CultriX/Qwenfinity-2.5-14B](https://huggingface.co/CultriX/Qwenfinity-2.5-14B)
|
||
|
|
* [allknowingroger/QwenSlerp6-14B](https://huggingface.co/allknowingroger/QwenSlerp6-14B)
|
||
|
|
* [CultriX/Qwen2.5-14B-Broca](https://huggingface.co/CultriX/Qwen2.5-14B-Broca)
|
||
|
|
* [sometimesanotion/Qwen2.5-14B-Vimarckoso](https://huggingface.co/sometimesanotion/Qwen2.5-14B-Vimarckoso)
|
||
|
|
* [CultriX/SeQwence-14Bv1](https://huggingface.co/CultriX/SeQwence-14Bv1)
|
||
|
|
|
||
|
|
### Configuration
|
||
|
|
|
||
|
|
The following YAML configuration was used to produce this model:
|
||
|
|
|
||
|
|
```yaml
|
||
|
|
merge_method: della_linear
|
||
|
|
base_model: CultriX/Qwen2.5-14B-Wernickev3
|
||
|
|
dtype: bfloat16
|
||
|
|
parameters:
|
||
|
|
epsilon: 0.012 # Further reduced to ensure ultra-fine parameter scaling for precision.
|
||
|
|
lambda: 1.4 # Stronger emphasis on significant model contributions.
|
||
|
|
normalize: true # Balances the parameter integration for stability.
|
||
|
|
|
||
|
|
adaptive_merge_parameters:
|
||
|
|
task_weights:
|
||
|
|
tinyArc: 1.6 # Prioritizes logical reasoning improvements.
|
||
|
|
tinyHellaswag: 1.5 # Strengthened contextual understanding and consistency.
|
||
|
|
tinyMMLU: 1.65 # Enhanced domain knowledge for multitask benchmarks.
|
||
|
|
tinyTruthfulQA: 1.9 # Maximized for accurate factual reasoning and QA.
|
||
|
|
tinyTruthfulQA_mc1: 1.7 # Balanced focus for multiple-choice reasoning.
|
||
|
|
tinyWinogrande: 1.75 # Advanced reasoning and contextual prediction improvement.
|
||
|
|
IFEval: 1.9 # Instruction-following tasks boosted by multitask contributors.
|
||
|
|
BBH: 1.7 # Complex reasoning is supported by logical base models.
|
||
|
|
MATH: 2.1 # Highest priority, focusing on mathematical excellence.
|
||
|
|
GPQA: 1.8 # Boosted graduate-level QA capabilities.
|
||
|
|
MUSR: 1.9 # Nuanced multi-step reasoning strengthened further.
|
||
|
|
MMLU-PRO: 1.8 # Domain multitask performance maximized.
|
||
|
|
smoothing_factor: 0.1 # Precisely tuned for smooth task-specific blending.
|
||
|
|
|
||
|
|
gradient_clipping:
|
||
|
|
CultriX/Qwen2.5-14B-Wernickev3: 0.86 # Backbone stability with slightly reduced clipping.
|
||
|
|
CultriX/Qwenfinity-2.5-14B: 0.83 # Consistent multitask integration.
|
||
|
|
djuna/Q2.5-Veltha-14B-0.5: 0.91 # Strengthened advanced reasoning contributions.
|
||
|
|
CultriX/Qwen2.5-14B-Broca: 0.85 # Logical reasoning enhancements stabilized.
|
||
|
|
qingy2019/Qwen2.5-Math-14B-Instruct: 0.93 # Mathematically focused tasks maximized.
|
||
|
|
CultriX/SeQwence-14Bv1: 0.88 # Generalist multitask support.
|
||
|
|
sometimesanotion/Qwen2.5-14B-Vimarckoso: 0.89 # Balanced multi-step reasoning contributions.
|
||
|
|
allknowingroger/QwenSlerp6-14B: 0.87 # Contextual and logical reasoning integration refined.
|
||
|
|
|
||
|
|
models:
|
||
|
|
- model: CultriX/Qwen2.5-14B-Wernickev3
|
||
|
|
parameters:
|
||
|
|
weight: 0.26 # Core backbone for multitask reasoning.
|
||
|
|
density: 0.7 # Slight increase to preserve critical reasoning parameters.
|
||
|
|
- model: CultriX/Qwenfinity-2.5-14B
|
||
|
|
parameters:
|
||
|
|
weight: 0.23 # Comprehensive multitask performer.
|
||
|
|
density: 0.65
|
||
|
|
- model: djuna/Q2.5-Veltha-14B-0.5
|
||
|
|
parameters:
|
||
|
|
weight: 0.22 # Advanced reasoning support for GPQA and MUSR.
|
||
|
|
density: 0.72
|
||
|
|
- model: CultriX/Qwen2.5-14B-Broca
|
||
|
|
parameters:
|
||
|
|
weight: 0.15 # Logical reasoning and factual QA enhancements.
|
||
|
|
density: 0.65
|
||
|
|
- model: qingy2019/Qwen2.5-Math-14B-Instruct
|
||
|
|
parameters:
|
||
|
|
weight: 0.18 # Mathematical reasoning priority.
|
||
|
|
density: 0.73
|
||
|
|
- model: CultriX/SeQwence-14Bv1
|
||
|
|
parameters:
|
||
|
|
weight: 0.14 # Generalist multitask backbone.
|
||
|
|
density: 0.63
|
||
|
|
- model: sometimesanotion/Qwen2.5-14B-Vimarckoso
|
||
|
|
parameters:
|
||
|
|
weight: 0.12 # Multi-step reasoning tasks contributor.
|
||
|
|
density: 0.6
|
||
|
|
- model: allknowingroger/QwenSlerp6-14B
|
||
|
|
parameters:
|
||
|
|
weight: 0.1 # Contextual reasoning improvements.
|
||
|
|
density: 0.62
|
||
|
|
|
||
|
|
tokenizer_source: CultriX/Qwen2.5-14B-Wernickev3
|
||
|
|
|
||
|
|
```
|