fusion-bench 0.2.18__py3-none-any.whl → 0.2.20__py3-none-any.whl

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (50) hide show
  1. fusion_bench/__init__.py +6 -0
  2. fusion_bench/constants/banner.py +12 -0
  3. fusion_bench/method/__init__.py +2 -0
  4. fusion_bench/method/linear/simple_average_for_llama.py +30 -5
  5. fusion_bench/method/regmean_plusplus/__init__.py +3 -0
  6. fusion_bench/method/regmean_plusplus/clip_regmean_plusplus.py +192 -0
  7. fusion_bench/method/regmean_plusplus/regmean_plusplus.py +365 -0
  8. fusion_bench/method/simple_average.py +29 -3
  9. fusion_bench/modelpool/causal_lm/causal_lm.py +37 -6
  10. fusion_bench/modelpool/clip_vision/modelpool.py +45 -12
  11. fusion_bench/scripts/cli.py +1 -1
  12. fusion_bench/tasks/clip_classification/imagenet.py +1008 -2004
  13. fusion_bench/utils/lazy_state_dict.py +75 -3
  14. fusion_bench/utils/misc.py +66 -2
  15. fusion_bench/utils/modelscope.py +146 -0
  16. fusion_bench/utils/state_dict_arithmetic.py +10 -5
  17. {fusion_bench-0.2.18.dist-info → fusion_bench-0.2.20.dist-info}/METADATA +9 -1
  18. {fusion_bench-0.2.18.dist-info → fusion_bench-0.2.20.dist-info}/RECORD +50 -43
  19. fusion_bench_config/method/regmean/clip_regmean.yaml +1 -1
  20. fusion_bench_config/method/regmean_plusplus/clip_regmean_plusplus.yaml +11 -0
  21. fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-base-patch16_TALL20.yaml +73 -8
  22. fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-base-patch16_TALL20_model_only.yaml +27 -7
  23. fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-base-patch32_TA8.yaml +34 -4
  24. fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-base-patch32_TA8_control_task.yaml +14 -17
  25. fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-base-patch32_TA8_model_only.yaml +14 -3
  26. fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-base-patch32_TALL10.yaml +39 -5
  27. fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-base-patch32_TALL12.yaml +49 -5
  28. fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-base-patch32_TALL14.yaml +55 -5
  29. fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-base-patch32_TALL14_model_only.yaml +21 -4
  30. fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-base-patch32_TALL16.yaml +61 -5
  31. fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-base-patch32_TALL18.yaml +67 -5
  32. fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-base-patch32_TALL20.yaml +73 -5
  33. fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-base-patch32_TALL20_model_only.yaml +26 -3
  34. fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-base-patch32_single_finetuned.yaml +7 -5
  35. fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-base-patch32_single_task_projection.yaml +6 -10
  36. fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-base-patch32_sun397_and_cars.yaml +6 -7
  37. fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-base-patch32_sun397_and_dtd.yaml +6 -7
  38. fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-base-patch32_sun397_cars_and_dtd.yaml +7 -8
  39. fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-base-patch32_svhn_and_mnist.yaml +8 -0
  40. fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-base-patch32_two_tasks_control_task.yaml +4 -6
  41. fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-large-patch14_TA8.yaml +32 -7
  42. fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-large-patch14_TA8_model_only.yaml +14 -6
  43. fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-large-patch14_TALL20.yaml +73 -8
  44. fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-large-patch14_TALL20_model_only.yaml +27 -7
  45. fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-large-patch14_individual.yaml +6 -10
  46. fusion_bench_config/modelpool/CausalLMPool/Qwen2.5-1.5B_math_and_coder.yaml +11 -0
  47. {fusion_bench-0.2.18.dist-info → fusion_bench-0.2.20.dist-info}/WHEEL +0 -0
  48. {fusion_bench-0.2.18.dist-info → fusion_bench-0.2.20.dist-info}/entry_points.txt +0 -0
  49. {fusion_bench-0.2.18.dist-info → fusion_bench-0.2.20.dist-info}/licenses/LICENSE +0 -0
  50. {fusion_bench-0.2.18.dist-info → fusion_bench-0.2.20.dist-info}/top_level.txt +0 -0
fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-base-patch32_TALL20_model_only.yaml CHANGED
@@ -1,6 +1,29 @@
1
1
  # The 20 task used in the paper:
2
2
  # Wang et al. Localizing Task Information for Improved Model Merging and Compression
3
3
  # http://arxiv.org/abs/2405.07813
4
- defaults:
5
- - CLIPVisionModelPool@: _template
6
- - /model/clip-vit@models: clip-vit-base-patch32_TALL20
4
+ _target_: fusion_bench.modelpool.CLIPVisionModelPool
5
+ _recursive_: False
6
+ processor: openai/clip-vit-base-patch32
7
+ models:
8
+ _pretrained_: openai/clip-vit-base-patch32
9
+ sun397: tanganke/clip-vit-base-patch32_sun397
10
+ stanford-cars: tanganke/clip-vit-base-patch32_stanford-cars
11
+ resisc45: tanganke/clip-vit-base-patch32_resisc45
12
+ eurosat: tanganke/clip-vit-base-patch32_eurosat
13
+ svhn: tanganke/clip-vit-base-patch32_svhn
14
+ gtsrb: tanganke/clip-vit-base-patch32_gtsrb
15
+ mnist: tanganke/clip-vit-base-patch32_mnist
16
+ dtd: tanganke/clip-vit-base-patch32_dtd
17
+ oxford_flowers102: tanganke/clip-vit-base-patch32_oxford_flowers102
18
+ pcam: tanganke/clip-vit-base-patch32_pcam
19
+ fer2013: tanganke/clip-vit-base-patch32_fer2013
20
+ oxford-iiit-pet: tanganke/clip-vit-base-patch32_oxford-iiit-pet
21
+ stl10: tanganke/clip-vit-base-patch32_stl10
22
+ cifar100: tanganke/clip-vit-base-patch32_cifar100
23
+ cifar10: tanganke/clip-vit-base-patch32_cifar10
24
+ food101: tanganke/clip-vit-base-patch32_food101
25
+ fashion_mnist: tanganke/clip-vit-base-patch32_fashion_mnist
26
+ emnist_letters: tanganke/clip-vit-base-patch32_emnist_letters
27
+ kmnist: tanganke/clip-vit-base-patch32_kmnist
28
+ rendered-sst2: tanganke/clip-vit-base-patch32_rendered-sst2
29
+ platform: hf
fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-base-patch32_single_finetuned.yaml CHANGED
@@ -1,5 +1,7 @@
1
- defaults:
2
- - CLIPVisionModelPool@: _template
3
- - /model/clip-vit@models:
4
- - clip-vit-base-patch32
5
- - clip-vit-base-patch32_stanford-cars
1
+ _target_: fusion_bench.modelpool.CLIPVisionModelPool
2
+ _recursive_: False
3
+ processor: openai/clip-vit-base-patch32
4
+ models:
5
+ _pretrained_: openai/clip-vit-base-patch32
6
+ stanford-cars: tanganke/clip-vit-base-patch32_stanford-cars
7
+ platform: hf
fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-base-patch32_single_task_projection.yaml CHANGED
@@ -1,12 +1,8 @@
1
- defaults:
2
- - /model/clip-vit@models:
3
- - clip-vit-base-patch32
4
- - clip-vit-base-patch32_sun397
5
- - clip-vit-base-patch32_stanford-cars
6
1
  _target_: fusion_bench.modelpool.CLIPVisionModelPool
7
2
  _recursive_: false
8
- train_datasets: null
9
- test_datasets: null
10
- processor:
11
- _target_: transformers.CLIPProcessor.from_pretrained
12
- pretrained_model_name_or_path: openai/clip-vit-base-patch32
3
+ processor: openai/clip-vit-base-patch32
4
+ models:
5
+ _pretrained_: openai/clip-vit-base-patch32
6
+ sun397: tanganke/clip-vit-base-patch32_sun397
7
+ stanford-cars: tanganke/clip-vit-base-patch32_stanford-cars
8
+ platform: hf
fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-base-patch32_sun397_and_cars.yaml CHANGED
@@ -1,8 +1,4 @@
1
1
  defaults:
2
- - /model/clip-vit@models:
3
- - clip-vit-base-patch32
4
- - clip-vit-base-patch32_sun397
5
- - clip-vit-base-patch32_stanford-cars
6
2
  - /dataset/image_classification/train@train_datasets:
7
3
  - sun397
8
4
  - stanford-cars
@@ -11,6 +7,9 @@ defaults:
11
7
  - stanford-cars
12
8
  _target_: fusion_bench.modelpool.CLIPVisionModelPool
13
9
  _recursive_: False
14
- processor:
15
- _target_: transformers.CLIPProcessor.from_pretrained
16
- pretrained_model_name_or_path: openai/clip-vit-base-patch32
10
+ processor: openai/clip-vit-base-patch32
11
+ models:
12
+ _pretrained_: openai/clip-vit-base-patch32
13
+ sun397: tanganke/clip-vit-base-patch32_sun397
14
+ stanford-cars: tanganke/clip-vit-base-patch32_stanford-cars
15
+ platform: hf
fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-base-patch32_sun397_and_dtd.yaml CHANGED
@@ -1,8 +1,4 @@
1
1
  defaults:
2
- - /model/clip-vit@models:
3
- - clip-vit-base-patch32
4
- - clip-vit-base-patch32_sun397
5
- - clip-vit-base-patch32_dtd
6
2
  - /dataset/image_classification/train@train_datasets:
7
3
  - sun397
8
4
  - dtd
@@ -11,6 +7,9 @@ defaults:
11
7
  - dtd
12
8
  _target_: fusion_bench.modelpool.CLIPVisionModelPool
13
9
  _recursive_: False
14
- processor:
15
- _target_: transformers.CLIPProcessor.from_pretrained
16
- pretrained_model_name_or_path: openai/clip-vit-base-patch32
10
+ processor: openai/clip-vit-base-patch32
11
+ models:
12
+ _pretrained_: openai/clip-vit-base-patch32
13
+ sun397: tanganke/clip-vit-base-patch32_sun397
14
+ dtd: tanganke/clip-vit-base-patch32_dtd
15
+ platform: hf
fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-base-patch32_sun397_cars_and_dtd.yaml CHANGED
@@ -1,9 +1,4 @@
1
1
  defaults:
2
- - /model/clip-vit@models:
3
- - clip-vit-base-patch32
4
- - clip-vit-base-patch32_sun397
5
- - clip-vit-base-patch32_stanford-cars
6
- - clip-vit-base-patch32_dtd
7
2
  - /dataset/image_classification/train@train_datasets:
8
3
  - sun397
9
4
  - stanford-cars
@@ -14,6 +9,10 @@ defaults:
14
9
  - dtd
15
10
  _target_: fusion_bench.modelpool.CLIPVisionModelPool
16
11
  _recursive_: False
17
- processor:
18
- _target_: transformers.CLIPProcessor.from_pretrained
19
- pretrained_model_name_or_path: openai/clip-vit-base-patch32
12
+ processor: openai/clip-vit-base-patch32
13
+ models:
14
+ _pretrained_: openai/clip-vit-base-patch32
15
+ sun397: tanganke/clip-vit-base-patch32_sun397
16
+ stanford-cars: tanganke/clip-vit-base-patch32_stanford-cars
17
+ dtd: tanganke/clip-vit-base-patch32_dtd
18
+ platform: hf
fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-base-patch32_svhn_and_mnist.yaml CHANGED
@@ -4,3 +4,11 @@ defaults:
4
4
  - clip-vit-base-patch32
5
5
  - clip-vit-base-patch32_svhn
6
6
  - clip-vit-base-patch32_mnist
7
+ _target_: fusion_bench.modelpool.CLIPVisionModelPool
8
+ _recursive_: False
9
+ processor: openai/clip-vit-base-patch32
10
+ models:
11
+ _pretrained_: openai/clip-vit-base-patch32
12
+ svhn: tanganke/clip-vit-base-patch32_svhn
13
+ mnist: tanganke/clip-vit-base-patch32_mnist
14
+ platform: hf
fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-base-patch32_two_tasks_control_task.yaml CHANGED
@@ -1,14 +1,12 @@
1
1
  defaults:
2
- - _self_
3
2
  - /dataset/image_classification/train@train_datasets:
4
3
  - tiny-imagenet
4
+ - _self_
5
5
  _target_: fusion_bench.modelpool.CLIPVisionModelPool
6
- _recursive_: false
6
+ _recursive_: False
7
7
  models:
8
8
  _pretrained_: openai/clip-vit-base-patch32
9
9
  model_1: tanganke/clip-vit-base-patch32_sun397
10
10
  model_2: tanganke/clip-vit-base-patch32_stanford-cars
11
- train_datasets: ???
12
- processor:
13
- _target_: transformers.CLIPProcessor.from_pretrained
14
- pretrained_model_name_or_path: openai/clip-vit-base-patch32
11
+ processor: openai/clip-vit-base-patch32
12
+ platform: hf
fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-large-patch14_TA8.yaml CHANGED
@@ -1,8 +1,33 @@
1
1
  defaults:
2
- - CLIPVisionModelPool@: _template
3
- - /model/clip-vit@models: clip-vit-large-patch14_eight_tasks
4
- - /dataset/image_classification/train@train_datasets: the_eight_tasks
5
- - /dataset/image_classification/test@test_datasets: the_eight_tasks
6
- processor:
7
- _target_: transformers.CLIPProcessor.from_pretrained
8
- pretrained_model_name_or_path: openai/clip-vit-large-patch14
2
+ - /dataset/image_classification/train@train_datasets:
3
+ - sun397
4
+ - stanford-cars
5
+ - resisc45
6
+ - eurosat
7
+ - svhn
8
+ - gtsrb
9
+ - mnist
10
+ - dtd
11
+ - /dataset/image_classification/test@test_datasets:
12
+ - sun397
13
+ - stanford-cars
14
+ - resisc45
15
+ - eurosat
16
+ - svhn
17
+ - gtsrb
18
+ - mnist
19
+ - dtd
20
+ _target_: fusion_bench.modelpool.CLIPVisionModelPool
21
+ _recursive_: False
22
+ processor: openai/clip-vit-large-patch14
23
+ models:
24
+ _pretrained_: openai/clip-vit-large-patch14
25
+ sun397: tanganke/clip-vit-large-patch14_sun397
26
+ stanford-cars: tanganke/clip-vit-large-patch14_stanford-cars
27
+ resisc45: tanganke/clip-vit-large-patch14_dtd
28
+ eurosat: tanganke/clip-vit-large-patch14_eurosat
29
+ svhn: tanganke/clip-vit-large-patch14_svhn
30
+ gtsrb: tanganke/clip-vit-large-patch14_gtsrb
31
+ mnist: tanganke/clip-vit-large-patch14_mnist
32
+ dtd: tanganke/clip-vit-large-patch14_dtd
33
+ platform: hf
fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-large-patch14_TA8_model_only.yaml CHANGED
@@ -1,6 +1,14 @@
1
- defaults:
2
- - CLIPVisionModelPool@: _template
3
- - /model/clip-vit@models: clip-vit-large-patch14_eight_tasks
4
- processor:
5
- _target_: transformers.CLIPProcessor.from_pretrained
6
- pretrained_model_name_or_path: openai/clip-vit-large-patch14
1
+ _target_: fusion_bench.modelpool.CLIPVisionModelPool
2
+ _recursive_: False
3
+ processor: openai/clip-vit-large-patch14
4
+ models:
5
+ _pretrained_: openai/clip-vit-large-patch14
6
+ sun397: tanganke/clip-vit-large-patch14_sun397
7
+ stanford-cars: tanganke/clip-vit-large-patch14_stanford-cars
8
+ resisc45: tanganke/clip-vit-large-patch14_dtd
9
+ eurosat: tanganke/clip-vit-large-patch14_eurosat
10
+ svhn: tanganke/clip-vit-large-patch14_svhn
11
+ gtsrb: tanganke/clip-vit-large-patch14_gtsrb
12
+ mnist: tanganke/clip-vit-large-patch14_mnist
13
+ dtd: tanganke/clip-vit-large-patch14_dtd
14
+ platform: hf
fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-large-patch14_TALL20.yaml CHANGED
@@ -1,11 +1,76 @@
1
- # The 20 task used in the paper:
1
+ # The 20 task used in the paper:
2
2
  # Wang et al. Localizing Task Information for Improved Model Merging and Compression
3
3
  # http://arxiv.org/abs/2405.07813
4
4
  defaults:
5
- - CLIPVisionModelPool@: _template
6
- - /model/clip-vit@models: clip-vit-large-patch14_TALL20
7
- - /dataset/image_classification/train@train_datasets: TALL20
8
- - /dataset/image_classification/test@test_datasets: TALL20
9
- processor:
10
- _target_: transformers.CLIPProcessor.from_pretrained
11
- pretrained_model_name_or_path: openai/clip-vit-large-patch14
5
+ - /dataset/image_classification/train@train_datasets:
6
+ # -- begin of eight tasks in the task arithmetic paper ---
7
+ - sun397
8
+ - stanford-cars
9
+ - resisc45
10
+ - eurosat
11
+ - svhn
12
+ - gtsrb
13
+ - mnist
14
+ - dtd
15
+ # -- end of eight tasks in the task arithmetic paper ---
16
+ - oxford_flowers102
17
+ - pcam
18
+ - fer2013
19
+ - oxford-iiit-pet
20
+ - stl10
21
+ - cifar100
22
+ - cifar10
23
+ - food101
24
+ - fashion_mnist
25
+ - emnist_letters
26
+ - kmnist
27
+ - rendered-sst2
28
+ - /dataset/image_classification/test@test_datasets:
29
+ # -- begin of eight tasks in the task arithmetic paper ---
30
+ - sun397
31
+ - stanford-cars
32
+ - resisc45
33
+ - eurosat
34
+ - svhn
35
+ - gtsrb
36
+ - mnist
37
+ - dtd
38
+ # -- end of eight tasks in the task arithmetic paper ---
39
+ - oxford_flowers102
40
+ - pcam
41
+ - fer2013
42
+ - oxford-iiit-pet
43
+ - stl10
44
+ - cifar100
45
+ - cifar10
46
+ - food101
47
+ - fashion_mnist
48
+ - emnist_letters
49
+ - kmnist
50
+ - rendered-sst2
51
+ _target_: fusion_bench.modelpool.CLIPVisionModelPool
52
+ _recursive_: False
53
+ processor: openai/clip-vit-large-patch14
54
+ models:
55
+ _pretrained_: openai/clip-vit-large-patch14
56
+ sun397: tanganke/clip-vit-large-patch14_sun397
57
+ stanford-cars: tanganke/clip-vit-large-patch14_stanford-cars
58
+ resisc45: tanganke/clip-vit-large-patch14_resisc45
59
+ eurosat: tanganke/clip-vit-large-patch14_eurosat
60
+ svhn: tanganke/clip-vit-large-patch14_svhn
61
+ gtsrb: tanganke/clip-vit-large-patch14_gtsrb
62
+ mnist: tanganke/clip-vit-large-patch14_mnist
63
+ dtd: tanganke/clip-vit-large-patch14_dtd
64
+ oxford_flowers102: tanganke/clip-vit-large-patch14_oxford_flowers102
65
+ pcam: tanganke/clip-vit-large-patch14_pcam
66
+ fer2013: tanganke/clip-vit-large-patch14_fer2013
67
+ oxford-iiit-pet: tanganke/clip-vit-large-patch14_oxford-iiit-pet
68
+ stl10: tanganke/clip-vit-large-patch14_stl10
69
+ cifar100: tanganke/clip-vit-large-patch14_cifar100
70
+ cifar10: tanganke/clip-vit-large-patch14_cifar10
71
+ food101: tanganke/clip-vit-large-patch14_food101
72
+ fashion_mnist: tanganke/clip-vit-large-patch14_fashion_mnist
73
+ emnist_letters: tanganke/clip-vit-large-patch14_emnist_letters
74
+ kmnist: tanganke/clip-vit-large-patch14_kmnist
75
+ rendered-sst2: tanganke/clip-vit-large-patch14_rendered-sst2
76
+ platform: hf
fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-large-patch14_TALL20_model_only.yaml CHANGED
@@ -1,9 +1,29 @@
1
- # The 20 task used in the paper:
1
+ # The 20 task used in the paper:
2
2
  # Wang et al. Localizing Task Information for Improved Model Merging and Compression
3
3
  # http://arxiv.org/abs/2405.07813
4
- defaults:
5
- - CLIPVisionModelPool@: _template
6
- - /model/clip-vit@models: clip-vit-large-patch14_TALL20
7
- processor:
8
- _target_: transformers.CLIPProcessor.from_pretrained
9
- pretrained_model_name_or_path: openai/clip-vit-large-patch14
4
+ _target_: fusion_bench.modelpool.CLIPVisionModelPool
5
+ _recursive_: False
6
+ processor: openai/clip-vit-large-patch14
7
+ models:
8
+ _pretrained_: openai/clip-vit-large-patch14
9
+ sun397: tanganke/clip-vit-large-patch14_sun397
10
+ stanford-cars: tanganke/clip-vit-large-patch14_stanford-cars
11
+ resisc45: tanganke/clip-vit-large-patch14_resisc45
12
+ eurosat: tanganke/clip-vit-large-patch14_eurosat
13
+ svhn: tanganke/clip-vit-large-patch14_svhn
14
+ gtsrb: tanganke/clip-vit-large-patch14_gtsrb
15
+ mnist: tanganke/clip-vit-large-patch14_mnist
16
+ dtd: tanganke/clip-vit-large-patch14_dtd
17
+ oxford_flowers102: tanganke/clip-vit-large-patch14_oxford_flowers102
18
+ pcam: tanganke/clip-vit-large-patch14_pcam
19
+ fer2013: tanganke/clip-vit-large-patch14_fer2013
20
+ oxford-iiit-pet: tanganke/clip-vit-large-patch14_oxford-iiit-pet
21
+ stl10: tanganke/clip-vit-large-patch14_stl10
22
+ cifar100: tanganke/clip-vit-large-patch14_cifar100
23
+ cifar10: tanganke/clip-vit-large-patch14_cifar10
24
+ food101: tanganke/clip-vit-large-patch14_food101
25
+ fashion_mnist: tanganke/clip-vit-large-patch14_fashion_mnist
26
+ emnist_letters: tanganke/clip-vit-large-patch14_emnist_letters
27
+ kmnist: tanganke/clip-vit-large-patch14_kmnist
28
+ rendered-sst2: tanganke/clip-vit-large-patch14_rendered-sst2
29
+ platform: hf
fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-large-patch14_individual.yaml CHANGED
@@ -2,15 +2,11 @@
2
2
  #
3
3
  # fusion_bench \
4
4
  # modelpool=CLIPVisionModelPool/clip-vit-large-patch14_individual \
5
- # modelpool.base_model=${MODEL_PATH}
5
+ # modelpool.models._pretrained_=${MODEL_PATH}
6
6
  # ...
7
- defaults:
8
- - CLIPVisionModelPool@: _template
7
+ _target_: fusion_bench.modelpool.CLIPVisionModelPool
8
+ _recursive_: False
9
9
  models:
10
- _pretrained_:
11
- _target_: transformers.CLIPVisionModel.from_pretrained
12
- pretrained_model_name_or_path: ${...base_model}
13
- processor:
14
- _target_: transformers.CLIPProcessor.from_pretrained
15
- pretrained_model_name_or_path: ${..base_model}
16
- base_model: openai/clip-vit-large-patch14
10
+ _pretrained_: openai/clip-vit-large-patch14
11
+ processor: openai/clip-vit-large-patch14
12
+ platform: hf
fusion_bench_config/modelpool/CausalLMPool/Qwen2.5-1.5B_math_and_coder.yaml ADDED
@@ -0,0 +1,11 @@
1
+ _target_: fusion_bench.modelpool.CausalLMPool
2
+ _recursive_: false
3
+
4
+ load_lazy: false
5
+ models:
6
+ _pretrained_: Qwen/Qwen2.5-1.5B
7
+ expert_1: Qwen/Qwen2.5-Math-1.5B
8
+ expert_2: Qwen/Qwen2.5-Coder-1.5B
9
+ model_kwargs:
10
+ torch_dtype: bfloat16
11
+ tokenizer: Qwen/Qwen2.5-1.5B