fusion-bench 0.2.19__py3-none-any.whl → 0.2.21__py3-none-any.whl

fusion_bench_config/_get_started/clip_evaluate_single_model.yaml

@@ -0,0 +1,21 @@
+_target_: fusion_bench.programs.FabricModelFusionProgram
+_recursive_: false
+method:
+  _target_: fusion_bench.method.DummyAlgorithm
+modelpool:
+  _target_: fusion_bench.modelpool.CLIPVisionModelPool
+  models:
+    _pretrained_: openai/clip-vit-base-patch32
+taskpool:
+  _target_: fusion_bench.taskpool.CLIPVisionModelTaskPool
+  test_datasets:
+    sun397:
+      _target_: datasets.load_dataset
+      path: tanganke/sun397
+      split: test
+    stanford-cars:
+      _target_: datasets.load_dataset
+      path: tanganke/stanford_cars
+      split: test
+  clip_model: openai/clip-vit-base-patch32
+  processor: openai/clip-vit-base-patch32

fusion_bench_config/_get_started/clip_simple_average.yaml

@@ -0,0 +1,23 @@
+_target_: fusion_bench.programs.FabricModelFusionProgram # (1)!
+_recursive_: false
+method: # (2)!
+  _target_: fusion_bench.method.SimpleAverageAlgorithm
+modelpool: # (3)!
+  _target_: fusion_bench.modelpool.CLIPVisionModelPool
+  models:
+    _pretrained_: openai/clip-vit-base-patch32
+    sun397: tanganke/clip-vit-base-patch32_sun397
+    stanford-cars: tanganke/clip-vit-base-patch32_stanford-cars
+taskpool: # (4)!
+  _target_: fusion_bench.taskpool.CLIPVisionModelTaskPool
+  test_datasets:
+    sun397:
+      _target_: datasets.load_dataset
+      path: tanganke/sun397
+      split: test
+    stanford-cars:
+      _target_: datasets.load_dataset
+      path: tanganke/stanford_cars
+      split: test
+  clip_model: openai/clip-vit-base-patch32
+  processor: openai/clip-vit-base-patch32

fusion_bench_config/_get_started/clip_task_arithmetic.yaml

@@ -0,0 +1,24 @@
+_target_: fusion_bench.programs.FabricModelFusionProgram
+_recursive_: false
+method:
+  _target_: fusion_bench.method.TaskArithmeticAlgorithm
+  scaling_factor: 0.7
+modelpool:
+  _target_: fusion_bench.modelpool.CLIPVisionModelPool
+  models:
+    _pretrained_: openai/clip-vit-base-patch32
+    sun397: tanganke/clip-vit-base-patch32_sun397
+    stanford-cars: tanganke/clip-vit-base-patch32_stanford-cars
+taskpool:
+  _target_: fusion_bench.taskpool.CLIPVisionModelTaskPool
+  test_datasets:
+    sun397:
+      _target_: datasets.load_dataset
+      path: tanganke/sun397
+      split: test
+    stanford-cars:
+      _target_: datasets.load_dataset
+      path: tanganke/stanford_cars
+      split: test
+  clip_model: openai/clip-vit-base-patch32
+  processor: openai/clip-vit-base-patch32

fusion_bench_config/_get_started/greeting_program.yaml

@@ -0,0 +1,4 @@
+_target_: fusion_bench._get_started.greeting_program.GreetingProgram
+message: "Welcome to FusionBench"
+name: "Developer"
+repeat_count: 3

fusion_bench_config/fabric/loggers/csv_logger.yaml

@@ -3,9 +3,9 @@ _target_: lightning.fabric.loggers.CSVLogger
 # for example, `outputs/logs/lightning_logs/version_0` and `outputs/logs/lightning_logs/version_1` by default
 
 # root directory for all logging
-root_dir: outputs/logs
+root_dir: ${path.log_dir}
 # the name of the experiment
-name: lightning_logs
-version: null
+name: ""
+version: ""
 prefix: ""
 flush_logs_every_n_steps: 100

fusion_bench_config/fabric/loggers/tensorboard_logger.yaml

@@ -3,9 +3,9 @@ _target_: lightning.fabric.loggers.TensorBoardLogger
 # for example, `outputs/logs/lightning_logs/version_0` and `outputs/logs/lightning_logs/version_1` by default
 
 # root directory for all logging
-root_dir: outputs/logs
+root_dir: ${path.log_dir}
 # the name of the experiment
-name: "lightning_logs"
-version: null
+name: ""
+version: ""
 sub_dir: null
 default_hp_metric: false

fusion_bench_config/fabric_model_fusion.yaml

@@ -1,19 +1,47 @@
+# =============================================================================
+# FusionBench Fabric Model Fusion Configuration
+# =============================================================================
+# This configuration file defines the settings for running model fusion experiments
+# using PyTorch Lightning Fabric framework within FusionBench.
+#
+# The configuration includes:
+#
+# - Hydra framework settings and overrides
+# - PyTorch Lightning Fabric configuration for distributed training
+# - Path management for data, outputs, and logs
+# - (core components) Model pool, fusion method, and task pool specifications
+# - Experiment execution parameters and debugging options
+#
+# =============================================================================
+# Hydra Configuration Defaults
+# =============================================================================
 defaults:
-  - hydra: default
-  - fabric: auto
-  # --- Model, Method, Task ---
-  - modelpool: CLIPVisionModelPool/clip-vit-base-patch32_TA8
-  - method: dummy
-  - taskpool: dummy
-  - _self_
+  - hydra: default # Hydra framework configuration
+  - fabric: auto # PyTorch Lightning Fabric auto-configuration
+  - path: default # Path management configuration
+  # --- Core Components ---
+  - modelpool: CLIPVisionModelPool/clip-vit-base-patch32_TA8 # Model pool specification
+  - method: dummy # Fusion method (placeholder)
+  - taskpool: dummy # Task pool specification (placeholder)
+  - _self_ # Self-reference for override priority
+# =============================================================================
+# Program Configuration
+# =============================================================================
 _target_: fusion_bench.programs.FabricModelFusionProgram
-_recursive_: false
-fast_dev_run: false # Run a single batch of data to test the model or method
-# Run the script without actually running the experiment, use with `print_config=true`.
-# You can also use `--cfg` or `-c` to show the configuration instead of running.
-dry_run: false
-print_config: true # Print the configuration to the console
-merged_model_save_path: null # path to save the merged model, use "{log_dir}" to refer to the logger directory, for example `merged_model_save_path=\{log_dir\}/merged_model`
-merged_model_save_kwargs: null
-report_save_path: "{log_dir}/program_report.json" # path to save the result report
-print_function_call: true # set to false if you don't want to print the details of instantiate calls
+_recursive_: false # Disable recursive instantiation
+# =============================================================================
+# Experiment Execution Settings
+# =============================================================================
+# Development and debugging options
+fast_dev_run: false # This option is for quick testing. For example, run single batch instead of full dataset
+dry_run: false # Show configuration without running experiment
+print_config: true # Display full configuration before execution
+print_function_call: true # Show detailed instantiation calls
+# =============================================================================
+# Output and Logging Configuration
+# =============================================================================
+# Model saving configuration
+merged_model_save_path: null # Path to save merged model.
+merged_model_save_kwargs: null # Additional kwargs for model saving.
+# Report generation
+report_save_path: "{log_dir}/program_report.json" # Experiment results report path

fusion_bench_config/hydra/default.yaml

@@ -2,7 +2,11 @@ defaults:
   - override help: fusion_bench_help
   - override job_logging: rich_logging
 run:
-  dir: outputs/${hydra.job.name}/${now:%Y-%m-%d_%H-%M-%S}
+  dir: ${path.log_dir}
 sweep:
-  dir: outputs/${hydra.job.name}/${now:%Y-%m-%d_%H-%M-%S}
+  dir: ${path.log_dir}
   subdir: ${hydra.job.num}
+job:
+  env_set:
+    HYDRA_FULL_ERROR: ${oc.env:HYDRA_FULL_ERROR,1}
+output_subdir: ""

fusion_bench_config/llama_full_finetune.yaml

@@ -1,6 +1,7 @@
 defaults:
   - hydra: default
   - fabric: llama_fsdp
+  - path: default
   # --- Model, Method, Task ---
   - method: lm_finetune/fullfinetune_sft.yaml
   - modelpool: CausalLMPool/llama_alpaca_cleaned.yaml

fusion_bench_config/method/adamerging/clip.yaml

@@ -1,5 +1,5 @@
 # this option can be "clip_task_wise_adamerging"
-name: ???
+name: clip_layer_wise_adamerging
 # this weights can be a list of float, or a string that points to a *.np, *.pt file containing the weights
 # if weights is specified, skip the test-time adaptation training
 weights: null

fusion_bench_config/method/bitdelta/bitdelta.yaml

@@ -0,0 +1,12 @@
+_target_: fusion_bench.method.bitdelta.BitDeltaAlgorithm
+save_dir: null
+save_full_model: false
+# training arguments
+lr: 1e-4
+batch_size: 4
+num_steps: 100
+# dataset arguments
+dataset_name: c4
+subset: en
+split: train
+max_length: 128

fusion_bench_config/method/depth_upscaling.yaml

@@ -1,5 +1,8 @@
 _target_: DepthUpscalingAlgorithm
-# this should be a list of integers or string, indicating the sequence of layers. If the entry is an integer, it will use the n-th layer of the model. If the entry is a string, it will use the layers specified by the string. The string should be a valid python expression that evaluates to a list of integers.
+# this should be a list of integers or string, indicating the sequence of layers. 
+# If the entry is an integer, it will use the n-th layer of the model. 
+# If the entry is a string, it will use the layers specified by the string. 
+# The string should be a valid python expression that evaluates to a list of integers.
 # for example, ["range(0,12)", "range(6,12)"] will use the first 12 layers and the last 6 layers of the model to construct the new model
 # [0, 2, 4, "range(6,12)"] will use the 1st, 3rd, 5th, and the 7th to 12th layers of the model to construct the new model
 layer_indices: null

fusion_bench_config/method/regmean/clip_regmean.yaml

@@ -5,7 +5,7 @@ exclude_param_names_regex: []
 num_regmean_examples: 256
 weight_transpose: true
 # float, reduce non-diagonal elements in regmean weights by multiplying this scalar
-reduce_non_diagonal_ratio: 0.6
+reduce_non_diagonal_ratio: 0.95
 dataloader_kwargs:
   batch_size: 32
   num_workers: 0

fusion_bench_config/method/regmean_plusplus/clip_regmean_plusplus.yaml

@@ -0,0 +1,11 @@
+_target_: fusion_bench.method.RegMeanAlgorithmForCLIPPlusPlus
+# list, regular expression of names of parameters that need to be excluded
+exclude_param_names_regex: []
+# numbers of examples to compute regmean weights
+num_regmean_examples: 256
+weight_transpose: true
+# float, reduce non-diagonal elements in regmean weights by multiplying this scalar
+reduce_non_diagonal_ratio: 0.95
+dataloader_kwargs:
+  batch_size: 32
+  num_workers: 0

fusion_bench_config/method/smile_upscaling/error_accumulation.yaml

@@ -0,0 +1,5 @@
+# Measure error accumulation
+_target_: fusion_bench.method.smile_upscaling.error_accumulation.ErrorAccumulationAnalysisForCLIP
+gate_k: 16
+k: 128
+top_k: 1

fusion_bench_config/method/smile_upscaling/projected_energy.yaml

@@ -0,0 +1,2 @@
+# Measure projected energy
+_target_: fusion_bench.method.smile_upscaling.projected_energy.ProjectedEnergyAnalysis

fusion_bench_config/method/smile_upscaling/smile_qwen2_upscaling.yaml

@@ -11,3 +11,4 @@ num_experts_per_tok: 1
 rank_of_router: 8
 # if rank_of_expert < 0, dense expert is used.
 rank_of_expert: 64
+save_with_remote_code: true

fusion_bench_config/modelpool/CLIPVisionModelPool/_template.yaml

@@ -2,11 +2,8 @@ _usage_: |
   defaults:
     - CLIPVisionModelPool@: _template
 _target_: fusion_bench.modelpool.CLIPVisionModelPool
-_version_: "0.2"
 _recursive_: False
 models: ???
 train_datasets: null
 test_datasets: null
-processor:
-  _target_: transformers.CLIPProcessor.from_pretrained
-  pretrained_model_name_or_path: openai/clip-vit-base-patch32
+processor: openai/clip-vit-base-patch32

fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-base-patch16_TALL20.yaml

@@ -1,11 +1,76 @@
-# The 20 task used in the paper: 
+# The 20 task used in the paper:
 # Wang et al. Localizing Task Information for Improved Model Merging and Compression
 # http://arxiv.org/abs/2405.07813
 defaults:
-  - CLIPVisionModelPool@: _template
-  - /model/clip-vit@models: clip-vit-base-patch16_TALL20
-  - /dataset/image_classification/train@train_datasets: TALL20
-  - /dataset/image_classification/test@test_datasets: TALL20
-processor:
-  _target_: transformers.CLIPProcessor.from_pretrained
-  pretrained_model_name_or_path: openai/clip-vit-base-patch16
+  - /dataset/image_classification/train@train_datasets:
+      # -- begin of eight tasks in the task arithmetic paper ---
+      - sun397
+      - stanford-cars
+      - resisc45
+      - eurosat
+      - svhn
+      - gtsrb
+      - mnist
+      - dtd
+      # -- end of eight tasks in the task arithmetic paper ---
+      - oxford_flowers102
+      - pcam
+      - fer2013
+      - oxford-iiit-pet
+      - stl10
+      - cifar100
+      - cifar10
+      - food101
+      - fashion_mnist
+      - emnist_letters
+      - kmnist
+      - rendered-sst2
+  - /dataset/image_classification/test@test_datasets:
+      # -- begin of eight tasks in the task arithmetic paper ---
+      - sun397
+      - stanford-cars
+      - resisc45
+      - eurosat
+      - svhn
+      - gtsrb
+      - mnist
+      - dtd
+      # -- end of eight tasks in the task arithmetic paper ---
+      - oxford_flowers102
+      - pcam
+      - fer2013
+      - oxford-iiit-pet
+      - stl10
+      - cifar100
+      - cifar10
+      - food101
+      - fashion_mnist
+      - emnist_letters
+      - kmnist
+      - rendered-sst2
+_target_: fusion_bench.modelpool.CLIPVisionModelPool
+_recursive_: False
+processor: openai/clip-vit-base-patch16
+models:
+  _pretrained_: openai/clip-vit-base-patch16
+  sun397: tanganke/clip-vit-base-patch16_sun397
+  stanford-cars: tanganke/clip-vit-base-patch16_stanford-cars
+  resisc45: tanganke/clip-vit-base-patch16_resisc45
+  eurosat: tanganke/clip-vit-base-patch16_eurosat
+  svhn: tanganke/clip-vit-base-patch16_svhn
+  gtsrb: tanganke/clip-vit-base-patch16_gtsrb
+  mnist: tanganke/clip-vit-base-patch16_mnist
+  dtd: tanganke/clip-vit-base-patch16_dtd
+  oxford_flowers102: tanganke/clip-vit-base-patch16_oxford_flowers102
+  pcam: tanganke/clip-vit-base-patch16_pcam
+  fer2013: tanganke/clip-vit-base-patch16_fer2013
+  oxford-iiit-pet: tanganke/clip-vit-base-patch16_oxford-iiit-pet
+  stl10: tanganke/clip-vit-base-patch16_stl10
+  cifar100: tanganke/clip-vit-base-patch16_cifar100
+  cifar10: tanganke/clip-vit-base-patch16_cifar10
+  food101: tanganke/clip-vit-base-patch16_food101
+  fashion_mnist: tanganke/clip-vit-base-patch16_fashion_mnist
+  emnist_letters: tanganke/clip-vit-base-patch16_emnist_letters
+  kmnist: tanganke/clip-vit-base-patch16_kmnist
+  rendered-sst2: tanganke/clip-vit-base-patch16_rendered-sst2
+platform: hf

fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-base-patch16_TALL20_model_only.yaml

@@ -1,9 +1,29 @@
-# The 20 task used in the paper: 
+# The 20 task used in the paper:
 # Wang et al. Localizing Task Information for Improved Model Merging and Compression
 # http://arxiv.org/abs/2405.07813
-defaults:
-  - CLIPVisionModelPool@: _template
-  - /model/clip-vit@models: clip-vit-base-patch16_TALL20
-processor:
-  _target_: transformers.CLIPProcessor.from_pretrained
-  pretrained_model_name_or_path: openai/clip-vit-base-patch16
+_target_: fusion_bench.modelpool.CLIPVisionModelPool
+_recursive_: False
+processor: openai/clip-vit-base-patch16
+models:
+  _pretrained_: openai/clip-vit-base-patch16
+  sun397: tanganke/clip-vit-base-patch16_sun397
+  stanford-cars: tanganke/clip-vit-base-patch16_stanford-cars
+  resisc45: tanganke/clip-vit-base-patch16_resisc45
+  eurosat: tanganke/clip-vit-base-patch16_eurosat
+  svhn: tanganke/clip-vit-base-patch16_svhn
+  gtsrb: tanganke/clip-vit-base-patch16_gtsrb
+  mnist: tanganke/clip-vit-base-patch16_mnist
+  dtd: tanganke/clip-vit-base-patch16_dtd
+  oxford_flowers102: tanganke/clip-vit-base-patch16_oxford_flowers102
+  pcam: tanganke/clip-vit-base-patch16_pcam
+  fer2013: tanganke/clip-vit-base-patch16_fer2013
+  oxford-iiit-pet: tanganke/clip-vit-base-patch16_oxford-iiit-pet
+  stl10: tanganke/clip-vit-base-patch16_stl10
+  cifar100: tanganke/clip-vit-base-patch16_cifar100
+  cifar10: tanganke/clip-vit-base-patch16_cifar10
+  food101: tanganke/clip-vit-base-patch16_food101
+  fashion_mnist: tanganke/clip-vit-base-patch16_fashion_mnist
+  emnist_letters: tanganke/clip-vit-base-patch16_emnist_letters
+  kmnist: tanganke/clip-vit-base-patch16_kmnist
+  rendered-sst2: tanganke/clip-vit-base-patch16_rendered-sst2
+platform: hf

fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-base-patch32_TA8.yaml

@@ -1,5 +1,35 @@
+# eight image classification tasks defined in task arithmetic paper
 defaults:
-  - CLIPVisionModelPool@: _template
-  - /model/clip-vit@models: clip-vit-base-patch32_eight_tasks
-  - /dataset/image_classification/train@train_datasets: the_eight_tasks
-  - /dataset/image_classification/test@test_datasets: the_eight_tasks
+  - /dataset/image_classification/train@train_datasets:
+      - sun397
+      - stanford-cars
+      - resisc45
+      - eurosat
+      - svhn
+      - gtsrb
+      - mnist
+      - dtd
+  - /dataset/image_classification/test@test_datasets:
+      - sun397
+      - stanford-cars
+      - resisc45
+      - eurosat
+      - svhn
+      - gtsrb
+      - mnist
+      - dtd
+  - _self_
+_target_: fusion_bench.modelpool.CLIPVisionModelPool
+_recursive_: False
+processor: openai/clip-vit-base-patch32
+models:
+  _pretrained_: openai/clip-vit-base-patch32
+  sun397: tanganke/clip-vit-base-patch32_sun397
+  stanford-cars: tanganke/clip-vit-base-patch32_stanford-cars
+  resisc45: tanganke/clip-vit-base-patch32_resisc45
+  eurosat: tanganke/clip-vit-base-patch32_eurosat
+  svhn: tanganke/clip-vit-base-patch32_svhn
+  gtsrb: tanganke/clip-vit-base-patch32_gtsrb
+  mnist: tanganke/clip-vit-base-patch32_mnist
+  dtd: tanganke/clip-vit-base-patch32_dtd
+platform: hf

fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-base-patch32_TA8_control_task.yaml

@@ -1,21 +1,18 @@
 defaults:
-  - _self_
-  - /model/clip-vit@models:
-      - clip-vit-base-patch32
-      - clip-vit-base-patch32_sun397
-      - clip-vit-base-patch32_stanford-cars
-      - clip-vit-base-patch32_resisc45
-      - clip-vit-base-patch32_eurosat
-      - clip-vit-base-patch32_svhn
-      - clip-vit-base-patch32_gtsrb
-      - clip-vit-base-patch32_mnist
-      - clip-vit-base-patch32_dtd
   - /dataset/image_classification/train@train_datasets:
       - tiny-imagenet
+  - _self_
 _target_: fusion_bench.modelpool.CLIPVisionModelPool
-_recursive_: false
-models: ???
-train_datasets: ???
-processor:
-  _target_: transformers.CLIPProcessor.from_pretrained
-  pretrained_model_name_or_path: openai/clip-vit-base-patch32
+_recursive_: False
+processor: openai/clip-vit-base-patch32
+models:
+  _pretrained_: openai/clip-vit-base-patch32
+  sun397: tanganke/clip-vit-base-patch32_sun397
+  stanford-cars: tanganke/clip-vit-base-patch32_stanford-cars
+  resisc45: tanganke/clip-vit-base-patch32_resisc45
+  eurosat: tanganke/clip-vit-base-patch32_eurosat
+  svhn: tanganke/clip-vit-base-patch32_svhn
+  gtsrb: tanganke/clip-vit-base-patch32_gtsrb
+  mnist: tanganke/clip-vit-base-patch32_mnist
+  dtd: tanganke/clip-vit-base-patch32_dtd
+platform: hf

fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-base-patch32_TA8_model_only.yaml

@@ -1,3 +1,14 @@
-defaults:
-  - CLIPVisionModelPool@: _template
-  - /model/clip-vit@models: clip-vit-base-patch32_eight_tasks
+_target_: fusion_bench.modelpool.CLIPVisionModelPool
+_recursive_: False
+processor: openai/clip-vit-base-patch32
+models:
+  _pretrained_: openai/clip-vit-base-patch32
+  sun397: tanganke/clip-vit-base-patch32_sun397
+  stanford-cars: tanganke/clip-vit-base-patch32_stanford-cars
+  resisc45: tanganke/clip-vit-base-patch32_resisc45
+  eurosat: tanganke/clip-vit-base-patch32_eurosat
+  svhn: tanganke/clip-vit-base-patch32_svhn
+  gtsrb: tanganke/clip-vit-base-patch32_gtsrb
+  mnist: tanganke/clip-vit-base-patch32_mnist
+  dtd: tanganke/clip-vit-base-patch32_dtd
+platform: hf

fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-base-patch32_TALL10.yaml

@@ -1,8 +1,42 @@
-# The 20 task used in the paper: 
+# The 10 task used in the paper (TALL mask):
 # Wang et al. Localizing Task Information for Improved Model Merging and Compression
 # http://arxiv.org/abs/2405.07813
 defaults:
-  - CLIPVisionModelPool@: _template
-  - /model/clip-vit@models: clip-vit-base-patch32_TALL10
-  - /dataset/image_classification/train@train_datasets: TALL10
-  - /dataset/image_classification/test@test_datasets: TALL10
+  - /dataset/image_classification/train@train_datasets:
+      - sun397
+      - stanford-cars
+      - resisc45
+      - eurosat
+      - svhn
+      - gtsrb
+      - mnist
+      - dtd
+      - oxford_flowers102
+      - pcam
+  - /dataset/image_classification/test@test_datasets:
+      - sun397
+      - stanford-cars
+      - resisc45
+      - eurosat
+      - svhn
+      - gtsrb
+      - mnist
+      - dtd
+      - oxford_flowers102
+      - pcam
+_target_: fusion_bench.modelpool.CLIPVisionModelPool
+_recursive_: False
+processor: openai/clip-vit-base-patch32
+models:
+  _pretrained_: openai/clip-vit-base-patch32
+  sun397: tanganke/clip-vit-base-patch32_sun397
+  stanford-cars: tanganke/clip-vit-base-patch32_stanford-cars
+  resisc45: tanganke/clip-vit-base-patch32_resisc45
+  eurosat: tanganke/clip-vit-base-patch32_eurosat
+  svhn: tanganke/clip-vit-base-patch32_svhn
+  gtsrb: tanganke/clip-vit-base-patch32_gtsrb
+  mnist: tanganke/clip-vit-base-patch32_mnist
+  dtd: tanganke/clip-vit-base-patch32_dtd
+  oxford_flowers102: tanganke/clip-vit-base-patch32_oxford_flowers102
+  pcam: tanganke/clip-vit-base-patch32_pcam
+platform: hf

fusion_bench_config/modelpool/CLIPVisionModelPool/clip-vit-base-patch32_TALL12.yaml

@@ -1,8 +1,52 @@
-# The 20 task used in the paper: 
+# The 12 task used in the paper (TALL mask):
 # Wang et al. Localizing Task Information for Improved Model Merging and Compression
 # http://arxiv.org/abs/2405.07813
 defaults:
-  - CLIPVisionModelPool@: _template
-  - /model/clip-vit@models: clip-vit-base-patch32_TALL12
-  - /dataset/image_classification/train@train_datasets: TALL12
-  - /dataset/image_classification/test@test_datasets: TALL12
+  - /dataset/image_classification/train@train_datasets:
+      # -- begin of eight tasks in the task arithmetic paper ---
+      - sun397
+      - stanford-cars
+      - resisc45
+      - eurosat
+      - svhn
+      - gtsrb
+      - mnist
+      - dtd
+      # -- end of eight tasks in the task arithmetic paper ---
+      - oxford_flowers102
+      - pcam
+      - fer2013
+      - oxford-iiit-pet
+  - /dataset/image_classification/test@test_datasets:
+      # -- begin of eight tasks in the task arithmetic paper ---
+      - sun397
+      - stanford-cars
+      - resisc45
+      - eurosat
+      - svhn
+      - gtsrb
+      - mnist
+      - dtd
+      # -- end of eight tasks in the task arithmetic paper ---
+      - oxford_flowers102
+      - pcam
+      - fer2013
+      - oxford-iiit-pet
+_target_: fusion_bench.modelpool.CLIPVisionModelPool
+_recursive_: False
+processor: openai/clip-vit-base-patch32
+models:
+  _pretrained_: openai/clip-vit-base-patch32
+  sun397: tanganke/clip-vit-base-patch32_sun397
+  stanford-cars: tanganke/clip-vit-base-patch32_stanford-cars
+  resisc45: tanganke/clip-vit-base-patch32_resisc45
+  eurosat: tanganke/clip-vit-base-patch32_eurosat
+  svhn: tanganke/clip-vit-base-patch32_svhn
+  gtsrb: tanganke/clip-vit-base-patch32_gtsrb
+  mnist: tanganke/clip-vit-base-patch32_mnist
+  dtd: tanganke/clip-vit-base-patch32_dtd
+  oxford_flowers102: tanganke/clip-vit-base-patch32_oxford_flowers102
+  pcam: tanganke/clip-vit-base-patch32_pcam
+  fer2013: tanganke/clip-vit-base-patch32_fer2013
+  oxford-iiit-pet: tanganke/clip-vit-base-patch32_oxford-iiit-pet
+platform: hf