Upload folder using huggingface_hub

README.md

@@ -0,0 +1,77 @@
+---
+base_model:
+- facebook/convnext-base-224
+library_name: transformers
+tags:
+- fusion-bench
+- merge
+---
+# Deep Model Fusion
+
+Fine-tuned ResNet model on dataset emnist_letters.
+
+## Models Merged
+
+This is a merged model created using [fusion-bench](https://github.com/tanganke/fusion_bench).
+
+The following models were included in the merge:
+
+
+- base model: facebook/convnext-base-224
+
+
+## Configuration
+
+The following YAML configuration was used to produce this model:
+
+### Algorithm Configuration
+
+```yaml
+_recursive_: false
+_target_: fusion_bench.method.classification.image_classification_finetune.ImageClassificationFineTuning
+_usage_: null
+_version_: 0.2.31.dev0
+dataloader_kwargs:
+  batch_size: 64
+  num_workers: 8
+  pin_memory: true
+label_smoothing: 0
+lr_scheduler: null
+max_epochs: -1
+max_steps: 4000
+optimizer:
+  _target_: torch.optim.SGD
+  lr: 0.01
+  momentum: 0.9
+  weight_decay: 0.0001
+save_interval: 1000
+save_on_train_epoch_end: false
+save_top_k: -1
+training_data_ratio: null
+```
+
+### Model Pool Configuration
+
+```yaml
+_recursive_: false
+_target_: fusion_bench.modelpool.convnext_for_image_classification.ConvNextForImageClassificationPool
+_usage_: null
+_version_: 0.2.31.dev0
+models:
+  _pretrained_:
+    config_path: facebook/convnext-base-224
+    dataset_name: emnist_letters
+    pretrained: true
+test_datasets: null
+train_datasets:
+  emnist_letters:
+    _target_: datasets.load_dataset
+    path: tanganke/emnist_letters
+    split: train
+val_datasets:
+  emnist_letters:
+    _target_: datasets.load_dataset
+    path: tanganke/emnist_letters
+    split: test
+```
+

cli.log

@@ -0,0 +1,26 @@
+[2026-01-08 20:06:09,718][fusion_bench.programs.fusion_program][INFO] - Running the model fusion program.
+[2026-01-08 20:06:09,722][fusion_bench.programs.fusion_program][INFO] - loading model pool
+[2026-01-08 20:06:12,117][fusion_bench.programs.fusion_program][INFO] - loading method
+[2026-01-08 20:06:12,180][fusion_bench.method.classification.image_classification_finetune][INFO] - Training interval: step
+[2026-01-08 20:06:12,181][fusion_bench.method.classification.image_classification_finetune][INFO] - Max epochs: -1, max steps: 4000
+[2026-01-08 20:06:12,182][fusion_bench.programs.fusion_program][INFO] - loading task pool
+[2026-01-08 20:06:13,960][fusion_bench.method.classification.image_classification_finetune][INFO] - Number of classes for dataset emnist_letters: 26
+[2026-01-08 20:06:13,972][datasets.load][WARNING] - Using the latest cached version of the dataset since tanganke/emnist_letters couldn't be found on the Hugging Face Hub (offline mode is enabled).
+[2026-01-08 20:06:13,982][datasets.packaged_modules.cache.cache][WARNING] - Found the latest cached dataset configuration 'emnist-letters' at /data/dataset/datasets/tanganke___emnist_letters/emnist-letters/0.0.0/4236c26fe034bc5b23851adc00067fa87b8ac16b (last modified on Mon Jan  5 04:58:28 2026).
+[2026-01-08 20:06:14,656][fusion_bench.method.classification.image_classification_finetune][INFO] - Training dataset size: 124800
+[2026-01-08 20:06:14,687][datasets.load][WARNING] - Using the latest cached version of the dataset since tanganke/emnist_letters couldn't be found on the Hugging Face Hub (offline mode is enabled).
+[2026-01-08 20:06:14,696][datasets.packaged_modules.cache.cache][WARNING] - Found the latest cached dataset configuration 'emnist-letters' at /data/dataset/datasets/tanganke___emnist_letters/emnist-letters/0.0.0/4236c26fe034bc5b23851adc00067fa87b8ac16b (last modified on Mon Jan  5 04:58:28 2026).
+[2026-01-08 20:06:14,818][fusion_bench.method.classification.image_classification_finetune][INFO] - optimizer:
+SGD (
+Parameter Group 0
+    dampening: 0
+    differentiable: False
+    foreach: None
+    fused: None
+    lr: 0.01
+    maximize: False
+    momentum: 0.9
+    nesterov: False
+    weight_decay: 0.0001
+)
+[2026-01-08 20:29:04,217][fusion_bench.method.classification.image_classification_finetune][INFO] - Saving the final model to outputs/convnext-base-224/emnist_letters/batch_size=64,lr=0.01/raw_checkpoints/final

config.json

@@ -0,0 +1,98 @@
+{
+  "architectures": [
+    "ConvNextForImageClassification"
+  ],
+  "depths": [
+    3,
+    3,
+    27,
+    3
+  ],
+  "drop_path_rate": 0.0,
+  "dtype": "float32",
+  "hidden_act": "gelu",
+  "hidden_sizes": [
+    128,
+    256,
+    512,
+    1024
+  ],
+  "id2label": {
+    "0": "A",
+    "1": "B",
+    "2": "C",
+    "3": "D",
+    "4": "E",
+    "5": "F",
+    "6": "G",
+    "7": "H",
+    "8": "I",
+    "9": "J",
+    "10": "K",
+    "11": "L",
+    "12": "M",
+    "13": "N",
+    "14": "O",
+    "15": "P",
+    "16": "Q",
+    "17": "R",
+    "18": "S",
+    "19": "T",
+    "20": "U",
+    "21": "V",
+    "22": "W",
+    "23": "X",
+    "24": "Y",
+    "25": "Z"
+  },
+  "image_size": 224,
+  "initializer_range": 0.02,
+  "label2id": {
+    "A": 0,
+    "B": 1,
+    "C": 2,
+    "D": 3,
+    "E": 4,
+    "F": 5,
+    "G": 6,
+    "H": 7,
+    "I": 8,
+    "J": 9,
+    "K": 10,
+    "L": 11,
+    "M": 12,
+    "N": 13,
+    "O": 14,
+    "P": 15,
+    "Q": 16,
+    "R": 17,
+    "S": 18,
+    "T": 19,
+    "U": 20,
+    "V": 21,
+    "W": 22,
+    "X": 23,
+    "Y": 24,
+    "Z": 25
+  },
+  "layer_norm_eps": 1e-12,
+  "layer_scale_init_value": 1e-06,
+  "model_type": "convnext",
+  "num_channels": 3,
+  "num_stages": 4,
+  "out_features": [
+    "stage4"
+  ],
+  "out_indices": [
+    4
+  ],
+  "patch_size": 4,
+  "stage_names": [
+    "stem",
+    "stage1",
+    "stage2",
+    "stage3",
+    "stage4"
+  ],
+  "transformers_version": "4.57.3"
+}

config.yaml

@@ -0,0 +1,58 @@
+path:
+  root_dir: ${oc.env:FUSION_BENCH_PROJECT_ROOT,"."}
+  output_dir: ${.root_dir}/outputs
+  data_dir: ${oc.env:FUSION_BENCH_DATA_DIR,${.root_dir}/data}
+  cache_dir: ${oc.env:FUSION_BENCH_CACHE_DIR,${.output_dir}/cache}
+  log_dir: outputs/convnext-base-224/emnist_letters/batch_size=64,lr=0.01
+  work_dir: ${hydra:runtime.cwd}
+modelpool:
+  _target_: fusion_bench.modelpool.ConvNextForImageClassificationPool
+  _recursive_: false
+  models:
+    _pretrained_:
+      config_path: facebook/convnext-base-224
+      pretrained: true
+      dataset_name: emnist_letters
+  train_datasets:
+    emnist_letters:
+      _target_: datasets.load_dataset
+      path: tanganke/emnist_letters
+      split: train
+  val_datasets:
+    emnist_letters:
+      _target_: datasets.load_dataset
+      path: tanganke/emnist_letters
+      split: test
+  test_datasets: null
+method:
+  _target_: fusion_bench.method.classification.ImageClassificationFineTuning
+  max_epochs: -1
+  max_steps: 4000
+  save_top_k: -1
+  save_interval: 1000
+  save_on_train_epoch_end: false
+  training_data_ratio: null
+  label_smoothing: 0
+  optimizer:
+    _target_: torch.optim.SGD
+    lr: 0.01
+    momentum: 0.9
+    weight_decay: 0.0001
+  lr_scheduler: null
+  dataloader_kwargs:
+    batch_size: 64
+    num_workers: 8
+    pin_memory: true
+taskpool:
+  _target_: fusion_bench.taskpool.DummyTaskPool
+  model_save_path: null
+_target_: fusion_bench.programs.ModelFusionProgram
+_recursive_: false
+seed: 0
+fast_dev_run: false
+dry_run: false
+print_config: true
+print_function_call: true
+merged_model_save_path: null
+merged_model_save_kwargs: null
+report_save_path: '{log_dir}/program_report.json'

events.out.tfevents.1767902775.pt-b7552130803b44b59bdfb04f6bb732f9-worker-0.101887.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e46a86e54b27209a56ddfef49a9d6001709ed226ab1af9b1cef53425467915b5
+size 2406120

hparams.yaml

@@ -0,0 +1 @@
+{}

hydra.yaml

@@ -0,0 +1,208 @@
+hydra:
+  run:
+    dir: ${path.log_dir}
+  sweep:
+    dir: ${oc.env:FUSION_BENCH_PROJECT_ROOT,"."}/outputs/multirun/${hydra.job.config_name}/${now:%Y-%m-%d_%H-%M-%S}
+    subdir: ${hydra.job.num}
+  launcher:
+    _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher
+  sweeper:
+    _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper
+    max_batch_size: null
+    params: null
+  help:
+    app_name: fusion_bench
+    header: == ${hydra.help.app_name} ==
+    footer: 'Powered by Hydra (https://hydra.cc)
+
+      Use --hydra-help to view Hydra specific help'
+    template: '${hydra.help.header}
+
+
+      fusion_bench is the command line interface for running model fusion benchmarks
+      in the FusionBench project.
+
+      It provides a flexible way to configure and execute various fusion algorithms
+      on different model pools and evaluate them across multiple tasks.
+
+
+      == Configuration groups ==
+
+      Compose your configuration from these groups (method, modelpool, taskpool are
+      the most important):
+
+
+      $APP_CONFIG_GROUPS
+
+
+      == Config ==
+
+      You can override options, for example:
+
+      fusion_bench method=task_arithmetic modelpool=clip-vit-base-patch32_svhn_and_mnist
+      taskpool=clip-vit-base-patch32_svhn_and_mnist
+
+
+      == Basic usage ==
+
+      fusion_bench [--config-path CONFIG_PATH] [--config-name CONFIG_NAME] OPTION_1=VALUE_1
+      OPTION_2=VALUE_2 ...
+
+
+      == Key options ==
+
+      --help, -h            : Print this help message and exit
+
+      --hydra-help          : Hydra''s help
+
+      --cfg, -c             : Show config instead of running [job|hydra|all]
+
+      --config-path, -cp    : Overrides the config_path
+
+      --config-name, -cn    : Overrides the config_name
+
+      --shell-completion, -sc : Install or Uninstall shell completion
+
+
+      For more detailed information on options and usage, please refer to the online
+      documentation:
+
+      https://tanganke.github.io/fusion_bench/cli/fusion_bench/
+
+
+      ${hydra.help.footer}'
+  hydra_help:
+    template: 'Hydra (${hydra.runtime.version})
+
+      See https://hydra.cc for more info.
+
+
+      == Flags ==
+
+      $FLAGS_HELP
+
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (For example, append hydra/job_logging=disabled
+      to command line)
+
+
+      $HYDRA_CONFIG_GROUPS
+
+
+      Use ''--cfg hydra'' to Show the Hydra config.
+
+      '
+    hydra_help: ???
+  hydra_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][HYDRA] %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+    root:
+      level: INFO
+      handlers:
+      - console
+    loggers:
+      logging_example:
+        level: DEBUG
+    disable_existing_loggers: false
+  job_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s'
+      rich_handler:
+        format: '%(message)s'
+    handlers:
+      console:
+        class: rich.logging.RichHandler
+        formatter: rich_handler
+      file:
+        class: logging.FileHandler
+        formatter: simple
+        filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log
+    root:
+      level: INFO
+      handlers:
+      - console
+      - file
+    disable_existing_loggers: false
+  env: {}
+  mode: RUN
+  searchpath: []
+  callbacks: {}
+  output_subdir: ''
+  overrides:
+    hydra:
+    - hydra.mode=RUN
+    task:
+    - path.log_dir="outputs/convnext-base-224/emnist_letters/batch_size=64,lr=0.01"
+    - seed=0
+    - method=classification/image_classification_finetune
+    - method.max_epochs=-1
+    - method.max_steps=4000
+    - method.save_top_k=-1
+    - method.save_interval=1000
+    - method.save_on_train_epoch_end=false
+    - method.optimizer.lr=0.01
+    - method.lr_scheduler=null
+    - method.dataloader_kwargs.batch_size=64
+    - modelpool=ConvNextForImageClassification/convnext-base-224
+    - modelpool.models._pretrained_.dataset_name=emnist_letters
+    - +dataset/image_classification/train@modelpool.train_datasets=emnist_letters
+    - +dataset/image_classification/test@modelpool.val_datasets=emnist_letters
+  job:
+    name: cli
+    chdir: null
+    override_dirname: +dataset/image_classification/test@modelpool.val_datasets=emnist_letters,+dataset/image_classification/train@modelpool.train_datasets=emnist_letters,method.dataloader_kwargs.batch_size=64,method.lr_scheduler=null,method.max_epochs=-1,method.max_steps=4000,method.optimizer.lr=0.01,method.save_interval=1000,method.save_on_train_epoch_end=false,method.save_top_k=-1,method=classification/image_classification_finetune,modelpool.models._pretrained_.dataset_name=emnist_letters,modelpool=ConvNextForImageClassification/convnext-base-224,path.log_dir="outputs/convnext-base-224/emnist_letters/batch_size=64,lr=0.01",seed=0
+    id: ???
+    num: ???
+    config_name: model_fusion
+    env_set:
+      HYDRA_FULL_ERROR: ${oc.env:HYDRA_FULL_ERROR,1}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.3'
+    cwd: /data/users/anke/fusion_bench
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /data/users/anke/fusion_bench/config
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /data/users/anke/fusion_bench/outputs/convnext-base-224/emnist_letters/batch_size=64,lr=0.01
+    choices:
+      dataset/image_classification/test@modelpool.val_datasets: emnist_letters
+      dataset/image_classification/train@modelpool.train_datasets: emnist_letters
+      taskpool: dummy
+      method: classification/image_classification_finetune
+      modelpool: ConvNextForImageClassification/convnext-base-224
+      path: default
+      hydra: default
+      hydra/env: default
+      hydra/callbacks: null
+      hydra/job_logging: rich_logging
+      hydra/hydra_logging: default
+      hydra/hydra_help: default
+      hydra/help: fusion_bench_help
+      hydra/sweeper: basic
+      hydra/launcher: basic
+      hydra/output: default
+  verbose: false

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:73b1109ca51d01400419806a0192054b97d126c40e5e6e7f95bb9dfd00c92b2b
+size 350413648

overrides.yaml

@@ -0,0 +1,15 @@
+- path.log_dir="outputs/convnext-base-224/emnist_letters/batch_size=64,lr=0.01"
+- seed=0
+- method=classification/image_classification_finetune
+- method.max_epochs=-1
+- method.max_steps=4000
+- method.save_top_k=-1
+- method.save_interval=1000
+- method.save_on_train_epoch_end=false
+- method.optimizer.lr=0.01
+- method.lr_scheduler=null
+- method.dataloader_kwargs.batch_size=64
+- modelpool=ConvNextForImageClassification/convnext-base-224
+- modelpool.models._pretrained_.dataset_name=emnist_letters
+- +dataset/image_classification/train@modelpool.train_datasets=emnist_letters
+- +dataset/image_classification/test@modelpool.val_datasets=emnist_letters

preprocessor_config.json

@@ -0,0 +1,22 @@
+{
+  "crop_pct": 0.875,
+  "do_normalize": true,
+  "do_rescale": true,
+  "do_resize": true,
+  "image_mean": [
+    0.485,
+    0.456,
+    0.406
+  ],
+  "image_processor_type": "ConvNextImageProcessor",
+  "image_std": [
+    0.229,
+    0.224,
+    0.225
+  ],
+  "resample": 3,
+  "rescale_factor": 0.00392156862745098,
+  "size": {
+    "shortest_edge": 224
+  }
+}

program_report.json

@@ -0,0 +1 @@
+{"model_info": {"trainable_params": 87593114, "all_params": 87593114, "trainable_percentage": 1.0}}