Upload pairwise bidirectional merge evaluation
Browse filesThis view is limited to 50 files because it contains too many changes. See raw diff
- .gitattributes +1 -0
- Bengali/final_pairwise_bidirectional_test_scores.csv +5 -0
- Bengali/final_pairwise_bidirectional_test_scores.json +46 -0
- Bengali/mergekit_configs/dare_sparsity_0.1_scale_0.1.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.1_scale_0.2.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.1_scale_0.3.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.1_scale_0.4.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.1_scale_0.5.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.1_scale_0.6.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.1_scale_0.7.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.1_scale_0.8.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.1_scale_0.9.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.2_scale_0.1.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.2_scale_0.2.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.2_scale_0.3.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.2_scale_0.4.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.2_scale_0.5.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.2_scale_0.6.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.2_scale_0.7.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.2_scale_0.8.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.2_scale_0.9.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.3_scale_0.1.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.3_scale_0.2.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.3_scale_0.3.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.3_scale_0.4.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.3_scale_0.5.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.3_scale_0.6.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.3_scale_0.7.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.3_scale_0.8.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.3_scale_0.9.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.4_scale_0.1.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.4_scale_0.2.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.4_scale_0.3.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.4_scale_0.4.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.4_scale_0.5.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.4_scale_0.6.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.4_scale_0.7.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.4_scale_0.8.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.4_scale_0.9.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.5_scale_0.1.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.5_scale_0.2.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.5_scale_0.3.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.5_scale_0.4.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.5_scale_0.5.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.5_scale_0.6.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.5_scale_0.7.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.5_scale_0.8.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.5_scale_0.9.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.6_scale_0.1.yaml +19 -0
- Bengali/mergekit_configs/dare_sparsity_0.6_scale_0.2.yaml +19 -0
.gitattributes
CHANGED
|
@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
|
| 33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
| 34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
| 35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
| 33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
| 34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
| 35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
| 36 |
+
Telugu/model_eval/Telugu_valid_dare_sparsity_0.99_scale_0.3/valid/predictions/English_to_Telugu_valid.json filter=lfs diff=lfs merge=lfs -text
|
Bengali/final_pairwise_bidirectional_test_scores.csv
ADDED
|
@@ -0,0 +1,5 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
Language,Model,English_to_Bengali_BLEU,Bengali_to_English_BLEU,English_to_Bengali_CHRF,Bengali_to_English_CHRF,average_bleu,average_chrf,model_name
|
| 2 |
+
Bengali,Task Arithmetic,2.1614810439426426,24.333693547954816,22.568108124290124,52.23549734538883,13.24758729594873,37.40180273483948,pairwise_bidirectional_merge_eval/Bengali/merged_checkpoints/task_arithmetic_scale_0.2
|
| 3 |
+
Bengali,TIES,0.43511663013152446,21.242599910786527,0.5654513466140579,51.91443228153667,10.838858270459026,26.239941814075365,pairwise_bidirectional_merge_eval/Bengali/merged_checkpoints/ties_keep_0.1_scale_0.6
|
| 4 |
+
Bengali,DARE,0.466615269528136,24.922993135494796,0.5885999696772924,53.01294313519763,12.694804202511467,26.80077155243746,pairwise_bidirectional_merge_eval/Bengali/merged_checkpoints/dare_sparsity_0.2_scale_0.2
|
| 5 |
+
Bengali,SCE-Merging,6.5592573303312784,22.115773370680497,34.97884463377777,51.34366165959373,14.337515350505887,43.16125314668575,pairwise_bidirectional_merge_eval/Bengali/merged_checkpoints/sce_topk_0.2
|
Bengali/final_pairwise_bidirectional_test_scores.json
ADDED
|
@@ -0,0 +1,46 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"Language": "Bengali",
|
| 4 |
+
"Model": "Task Arithmetic",
|
| 5 |
+
"model_name": "pairwise_bidirectional_merge_eval/Bengali/merged_checkpoints/task_arithmetic_scale_0.2",
|
| 6 |
+
"average_bleu": 13.24758729594873,
|
| 7 |
+
"average_chrf": 37.40180273483948,
|
| 8 |
+
"English_to_Bengali_BLEU": 2.1614810439426426,
|
| 9 |
+
"Bengali_to_English_BLEU": 24.333693547954816,
|
| 10 |
+
"English_to_Bengali_CHRF": 22.568108124290124,
|
| 11 |
+
"Bengali_to_English_CHRF": 52.23549734538883
|
| 12 |
+
},
|
| 13 |
+
{
|
| 14 |
+
"Language": "Bengali",
|
| 15 |
+
"Model": "TIES",
|
| 16 |
+
"model_name": "pairwise_bidirectional_merge_eval/Bengali/merged_checkpoints/ties_keep_0.1_scale_0.6",
|
| 17 |
+
"average_bleu": 10.838858270459026,
|
| 18 |
+
"average_chrf": 26.239941814075365,
|
| 19 |
+
"English_to_Bengali_BLEU": 0.43511663013152446,
|
| 20 |
+
"Bengali_to_English_BLEU": 21.242599910786527,
|
| 21 |
+
"English_to_Bengali_CHRF": 0.5654513466140579,
|
| 22 |
+
"Bengali_to_English_CHRF": 51.91443228153667
|
| 23 |
+
},
|
| 24 |
+
{
|
| 25 |
+
"Language": "Bengali",
|
| 26 |
+
"Model": "DARE",
|
| 27 |
+
"model_name": "pairwise_bidirectional_merge_eval/Bengali/merged_checkpoints/dare_sparsity_0.2_scale_0.2",
|
| 28 |
+
"average_bleu": 12.694804202511467,
|
| 29 |
+
"average_chrf": 26.80077155243746,
|
| 30 |
+
"English_to_Bengali_BLEU": 0.466615269528136,
|
| 31 |
+
"Bengali_to_English_BLEU": 24.922993135494796,
|
| 32 |
+
"English_to_Bengali_CHRF": 0.5885999696772924,
|
| 33 |
+
"Bengali_to_English_CHRF": 53.01294313519763
|
| 34 |
+
},
|
| 35 |
+
{
|
| 36 |
+
"Language": "Bengali",
|
| 37 |
+
"Model": "SCE-Merging",
|
| 38 |
+
"model_name": "pairwise_bidirectional_merge_eval/Bengali/merged_checkpoints/sce_topk_0.2",
|
| 39 |
+
"average_bleu": 14.337515350505887,
|
| 40 |
+
"average_chrf": 43.16125314668575,
|
| 41 |
+
"English_to_Bengali_BLEU": 6.5592573303312784,
|
| 42 |
+
"Bengali_to_English_BLEU": 22.115773370680497,
|
| 43 |
+
"English_to_Bengali_CHRF": 34.97884463377777,
|
| 44 |
+
"Bengali_to_English_CHRF": 51.34366165959373
|
| 45 |
+
}
|
| 46 |
+
]
|
Bengali/mergekit_configs/dare_sparsity_0.1_scale_0.1.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.1
|
| 7 |
+
density: 0.9
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.1
|
| 11 |
+
density: 0.9
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.1_scale_0.2.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.2
|
| 7 |
+
density: 0.9
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.2
|
| 11 |
+
density: 0.9
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.1_scale_0.3.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.3
|
| 7 |
+
density: 0.9
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.3
|
| 11 |
+
density: 0.9
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.1_scale_0.4.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.4
|
| 7 |
+
density: 0.9
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.4
|
| 11 |
+
density: 0.9
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.1_scale_0.5.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.5
|
| 7 |
+
density: 0.9
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.5
|
| 11 |
+
density: 0.9
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.1_scale_0.6.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.6
|
| 7 |
+
density: 0.9
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.6
|
| 11 |
+
density: 0.9
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.1_scale_0.7.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.7
|
| 7 |
+
density: 0.9
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.7
|
| 11 |
+
density: 0.9
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.1_scale_0.8.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.8
|
| 7 |
+
density: 0.9
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.8
|
| 11 |
+
density: 0.9
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.1_scale_0.9.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.9
|
| 7 |
+
density: 0.9
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.9
|
| 11 |
+
density: 0.9
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.2_scale_0.1.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.1
|
| 7 |
+
density: 0.8
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.1
|
| 11 |
+
density: 0.8
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.2_scale_0.2.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.2
|
| 7 |
+
density: 0.8
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.2
|
| 11 |
+
density: 0.8
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.2_scale_0.3.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.3
|
| 7 |
+
density: 0.8
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.3
|
| 11 |
+
density: 0.8
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.2_scale_0.4.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.4
|
| 7 |
+
density: 0.8
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.4
|
| 11 |
+
density: 0.8
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.2_scale_0.5.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.5
|
| 7 |
+
density: 0.8
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.5
|
| 11 |
+
density: 0.8
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.2_scale_0.6.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.6
|
| 7 |
+
density: 0.8
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.6
|
| 11 |
+
density: 0.8
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.2_scale_0.7.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.7
|
| 7 |
+
density: 0.8
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.7
|
| 11 |
+
density: 0.8
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.2_scale_0.8.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.8
|
| 7 |
+
density: 0.8
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.8
|
| 11 |
+
density: 0.8
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.2_scale_0.9.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.9
|
| 7 |
+
density: 0.8
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.9
|
| 11 |
+
density: 0.8
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.3_scale_0.1.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.1
|
| 7 |
+
density: 0.7
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.1
|
| 11 |
+
density: 0.7
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.3_scale_0.2.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.2
|
| 7 |
+
density: 0.7
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.2
|
| 11 |
+
density: 0.7
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.3_scale_0.3.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.3
|
| 7 |
+
density: 0.7
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.3
|
| 11 |
+
density: 0.7
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.3_scale_0.4.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.4
|
| 7 |
+
density: 0.7
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.4
|
| 11 |
+
density: 0.7
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.3_scale_0.5.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.5
|
| 7 |
+
density: 0.7
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.5
|
| 11 |
+
density: 0.7
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.3_scale_0.6.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.6
|
| 7 |
+
density: 0.7
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.6
|
| 11 |
+
density: 0.7
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.3_scale_0.7.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.7
|
| 7 |
+
density: 0.7
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.7
|
| 11 |
+
density: 0.7
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.3_scale_0.8.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.8
|
| 7 |
+
density: 0.7
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.8
|
| 11 |
+
density: 0.7
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.3_scale_0.9.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.9
|
| 7 |
+
density: 0.7
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.9
|
| 11 |
+
density: 0.7
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.4_scale_0.1.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.1
|
| 7 |
+
density: 0.6
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.1
|
| 11 |
+
density: 0.6
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.4_scale_0.2.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.2
|
| 7 |
+
density: 0.6
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.2
|
| 11 |
+
density: 0.6
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.4_scale_0.3.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.3
|
| 7 |
+
density: 0.6
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.3
|
| 11 |
+
density: 0.6
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.4_scale_0.4.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.4
|
| 7 |
+
density: 0.6
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.4
|
| 11 |
+
density: 0.6
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.4_scale_0.5.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.5
|
| 7 |
+
density: 0.6
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.5
|
| 11 |
+
density: 0.6
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.4_scale_0.6.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.6
|
| 7 |
+
density: 0.6
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.6
|
| 11 |
+
density: 0.6
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.4_scale_0.7.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.7
|
| 7 |
+
density: 0.6
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.7
|
| 11 |
+
density: 0.6
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.4_scale_0.8.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.8
|
| 7 |
+
density: 0.6
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.8
|
| 11 |
+
density: 0.6
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.4_scale_0.9.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.9
|
| 7 |
+
density: 0.6
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.9
|
| 11 |
+
density: 0.6
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.5_scale_0.1.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.1
|
| 7 |
+
density: 0.5
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.1
|
| 11 |
+
density: 0.5
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.5_scale_0.2.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.2
|
| 7 |
+
density: 0.5
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.2
|
| 11 |
+
density: 0.5
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.5_scale_0.3.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.3
|
| 7 |
+
density: 0.5
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.3
|
| 11 |
+
density: 0.5
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.5_scale_0.4.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.4
|
| 7 |
+
density: 0.5
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.4
|
| 11 |
+
density: 0.5
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.5_scale_0.5.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.5
|
| 7 |
+
density: 0.5
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.5
|
| 11 |
+
density: 0.5
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.5_scale_0.6.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.6
|
| 7 |
+
density: 0.5
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.6
|
| 11 |
+
density: 0.5
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.5_scale_0.7.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.7
|
| 7 |
+
density: 0.5
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.7
|
| 11 |
+
density: 0.5
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.5_scale_0.8.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.8
|
| 7 |
+
density: 0.5
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.8
|
| 11 |
+
density: 0.5
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.5_scale_0.9.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.9
|
| 7 |
+
density: 0.5
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.9
|
| 11 |
+
density: 0.5
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.6_scale_0.1.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.1
|
| 7 |
+
density: 0.4
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.1
|
| 11 |
+
density: 0.4
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|
Bengali/mergekit_configs/dare_sparsity_0.6_scale_0.2.yaml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
merge_method: dare_linear
|
| 2 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 3 |
+
models:
|
| 4 |
+
- model: baban/QwenTranslate_English_Bengali
|
| 5 |
+
parameters:
|
| 6 |
+
weight: 0.2
|
| 7 |
+
density: 0.4
|
| 8 |
+
- model: baban/QwenTranslate_Bengali_English
|
| 9 |
+
parameters:
|
| 10 |
+
weight: 0.2
|
| 11 |
+
density: 0.4
|
| 12 |
+
parameters:
|
| 13 |
+
normalize: true
|
| 14 |
+
int8_mask: false
|
| 15 |
+
rescale: true
|
| 16 |
+
dtype: bfloat16
|
| 17 |
+
tokenizer:
|
| 18 |
+
source: base
|
| 19 |
+
chat_template: auto
|