Update README.md
Browse files
README.md
CHANGED
|
@@ -7,43 +7,78 @@ tags:
|
|
| 7 |
- merge
|
| 8 |
|
| 9 |
---
|
| 10 |
-
#
|
| 11 |
|
| 12 |
This is a merge of pre-trained language models created using [mergekit](https://github.com/cg123/mergekit).
|
| 13 |
|
| 14 |
-
## Merge Details
|
| 15 |
-
### Merge Method
|
| 16 |
-
|
| 17 |
-
This model was merged using the [Model Stock](https://arxiv.org/abs/2403.19522) merge method using [Qwen/QwQ-32B](https://huggingface.co/Qwen/QwQ-32B) as a base.
|
| 18 |
-
|
| 19 |
-
### Models Merged
|
| 20 |
-
|
| 21 |
-
The following models were included in the merge:
|
| 22 |
-
* ./merge_model_3
|
| 23 |
-
* ./merge_model_4
|
| 24 |
-
* ./merge_model_2
|
| 25 |
-
* ./merge_model_1
|
| 26 |
-
|
| 27 |
### Configuration
|
| 28 |
|
| 29 |
The following YAML configuration was used to produce this model:
|
| 30 |
|
| 31 |
```yaml
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 32 |
merge_method: model_stock
|
| 33 |
|
| 34 |
base_model: Qwen/QwQ-32B
|
| 35 |
|
| 36 |
models:
|
| 37 |
- model: Qwen/QwQ-32B
|
| 38 |
-
- model:
|
| 39 |
-
- model:
|
| 40 |
-
- model:
|
| 41 |
-
- model:
|
| 42 |
|
| 43 |
dtype: bfloat16
|
| 44 |
|
| 45 |
pad_to_multiple_of: 512
|
| 46 |
tokenizer_source: base
|
| 47 |
|
| 48 |
-
name:
|
| 49 |
```
|
|
|
|
| 7 |
- merge
|
| 8 |
|
| 9 |
---
|
| 10 |
+
# QwQ-32B Kumo
|
| 11 |
|
| 12 |
This is a merge of pre-trained language models created using [mergekit](https://github.com/cg123/mergekit).
|
| 13 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 14 |
### Configuration
|
| 15 |
|
| 16 |
The following YAML configuration was used to produce this model:
|
| 17 |
|
| 18 |
```yaml
|
| 19 |
+
merge_method: slerp
|
| 20 |
+
base_model: Qwen/QwQ-32B
|
| 21 |
+
models:
|
| 22 |
+
- model: Qwen/QwQ-32B
|
| 23 |
+
- model: NovaSky-AI/Sky-T1-32B-Flash
|
| 24 |
+
parameters:
|
| 25 |
+
t: 0.4
|
| 26 |
+
dtype: bfloat16
|
| 27 |
+
name: merge_model_1
|
| 28 |
+
---
|
| 29 |
+
merge_method: breadcrumbs_ties
|
| 30 |
+
base_model: Qwen/QwQ-32B
|
| 31 |
+
tokenizer_source: Qwen/QwQ-32B
|
| 32 |
+
name: merge_model_2
|
| 33 |
+
models:
|
| 34 |
+
- model: Qwen/QwQ-32B
|
| 35 |
+
parameters:
|
| 36 |
+
weight: 1.0
|
| 37 |
+
- model: FuseAI/FuseO1-DeepSeekR1-Qwen2.5-Coder-32B-Preview
|
| 38 |
+
parameters:
|
| 39 |
+
weight: 0.75
|
| 40 |
+
dtype: bfloat16
|
| 41 |
+
---
|
| 42 |
+
merge_method: task_arithmetic
|
| 43 |
+
base_model: Qwen/Qwen2.5-32B
|
| 44 |
+
tokenizer_source: Qwen/QwQ-32B
|
| 45 |
+
name: merge_model_3
|
| 46 |
+
models:
|
| 47 |
+
- model: rinna/deepseek-r1-distill-qwen2.5-bakeneko-32b
|
| 48 |
+
parameters:
|
| 49 |
+
weight: 1.0
|
| 50 |
+
- model: cyberagent/DeepSeek-R1-Distill-Qwen-32B-Japanese
|
| 51 |
+
parameters:
|
| 52 |
+
weight: 0.9
|
| 53 |
+
tokenizer_source: base
|
| 54 |
+
dtype: bfloat16
|
| 55 |
+
---
|
| 56 |
+
merge_method: slerp
|
| 57 |
+
base_model: Qwen/QwQ-32B
|
| 58 |
+
models:
|
| 59 |
+
- model: Qwen/QwQ-32B
|
| 60 |
+
- model: TeamDelta/ABEJA-Qwen2.5-32B-base-jp-v0.1
|
| 61 |
+
parameters:
|
| 62 |
+
t: 0.5
|
| 63 |
+
tokenizer_source: base
|
| 64 |
+
dtype: bfloat16
|
| 65 |
+
name: merge_model_4
|
| 66 |
+
---
|
| 67 |
merge_method: model_stock
|
| 68 |
|
| 69 |
base_model: Qwen/QwQ-32B
|
| 70 |
|
| 71 |
models:
|
| 72 |
- model: Qwen/QwQ-32B
|
| 73 |
+
- model: merge_model_1
|
| 74 |
+
- model: merge_model_2
|
| 75 |
+
- model: merge_model_3
|
| 76 |
+
- model: merge_model_4
|
| 77 |
|
| 78 |
dtype: bfloat16
|
| 79 |
|
| 80 |
pad_to_multiple_of: 512
|
| 81 |
tokenizer_source: base
|
| 82 |
|
| 83 |
+
name: QwQ-32B-Kumo
|
| 84 |
```
|