Upload config.yaml

config.yaml

@@ -0,0 +1,180 @@
+accelerator:
+  _target_: trainer.accelerators.deepspeed_accelerator.DeepSpeedAccelerator
+  output_dir: outputs/Multi-CLIP_ViT_H_14-clean/from_con_image/stage2_eval_from_latest
+  mixed_precision: BF16
+  gradient_accumulation_steps: 1
+  log_with: WANDB
+  debug:
+    activate: false
+    port: 5900
+  seed: 42
+  resume_from_latest: true
+  resume_from_name_or_path: outputs/Multi-CLIP_ViT_H_14-clean/from_con_image/stage1_neg/checkpoint-latest
+  train_from_scratch: true
+  max_steps: 4000
+  num_epochs: 10
+  validate_steps: 100
+  eval_on_start: true
+  project_name: Multi-CLIP_ViT_H_14-clean
+  max_grad_norm: 1.0
+  save_steps: 100
+  metric_name: loss
+  metric_mode: MIN
+  limit_num_checkpoints: 3
+  save_only_if_best: true
+  dynamo_backend: 'NO'
+  keep_best_ckpts: true
+  deepspeed:
+    fp16:
+      enabled: false
+    bf16:
+      enabled: true
+    optimizer:
+      type: AdamW
+      params:
+        lr: auto
+        weight_decay: auto
+        torch_adam: true
+        adam_w_mode: true
+    scheduler:
+      type: WarmupDecayLR
+      params:
+        warmup_min_lr: auto
+        warmup_max_lr: auto
+        warmup_num_steps: auto
+        total_num_steps: auto
+    zero_optimization:
+      stage: 2
+      allgather_partitions: true
+      allgather_bucket_size: 200000000.0
+      overlap_comm: true
+      reduce_scatter: true
+      reduce_bucket_size: 500000000
+      contiguous_gradients: true
+    gradient_accumulation_steps: 16
+    gradient_clipping: 1.0
+    steps_per_print: 1
+    train_batch_size: auto
+    train_micro_batch_size_per_gpu: auto
+    wall_clock_breakdown: false
+  deepspeed_final:
+    fp16:
+      enabled: false
+    bf16:
+      enabled: true
+    optimizer:
+      type: AdamW
+      params:
+        lr: auto
+        weight_decay: auto
+        torch_adam: true
+        adam_w_mode: true
+    scheduler:
+      type: WarmupDecayLR
+      params:
+        warmup_min_lr: auto
+        warmup_max_lr: auto
+        warmup_num_steps: auto
+        total_num_steps: auto
+    zero_optimization:
+      stage: 2
+      allgather_partitions: true
+      allgather_bucket_size: 200000000.0
+      overlap_comm: true
+      reduce_scatter: true
+      reduce_bucket_size: 500000000
+      contiguous_gradients: true
+    gradient_accumulation_steps: 16
+    gradient_clipping: 1.0
+    steps_per_print: .inf
+    train_batch_size: auto
+    train_micro_batch_size_per_gpu: auto
+    wall_clock_breakdown: false
+task:
+  limit_examples_to_wandb: 50
+  _target_: trainer.tasks.mvclip_task.MVCLIPTask
+  pretrained_clip_model_name_or_path: laion/CLIP-ViT-H-14-laion2B-s32B-b79K
+  label_0_column_name: label_0
+  label_1_column_name: label_1
+  reference_type_column_name: reference_type
+  reference_input_column_name: reference_input
+  reference_idx_column_name: reference_idx
+  normal_pixels_0_column_name: normal_pixel_values_0
+  normal_pixels_1_column_name: normal_pixel_values_1
+  rgb_pixels_0_column_name: rgb_pixel_values_0
+  rgb_pixels_1_column_name: rgb_pixel_values_1
+model:
+  _target_: trainer.models.mvclip_model.MVCLIPModel
+  pretrained_clip_model_name_or_path: laion/CLIP-ViT-H-14-laion2B-s32B-b79K
+  freeze_clip: true
+  freeze_vision: false
+  freeze_logit_scale: false
+  freeze_logit_proj: false
+criterion:
+  _target_: trainer.criterions.criterion.MVCriterion
+  is_distributed: true
+  reference_type_column_name: reference_type
+  reference_input_column_name: reference_input
+  label_0_column_name: label_0
+  label_1_column_name: label_1
+  normal_pixels_0_column_name: normal_pixel_values_0
+  normal_pixels_1_column_name: normal_pixel_values_1
+  rgb_pixels_0_column_name: rgb_pixel_values_0
+  rgb_pixels_1_column_name: rgb_pixel_values_1
+  num_examples_per_prompt_column_name: num_example_per_prompt
+  in_batch_negatives: false
+dataset:
+  train_split_name: train_valid
+  valid_split_name: valid
+  test_split_name: test
+  batch_size: 16
+  num_workers: 2
+  drop_last: true
+  _target_: trainer.datasetss.dataset.MVDataset
+  records_dir: /mnt/petrelfs/zhangyuhan/data/evaluation/records/battle/clean/splits
+  gallery_dir: /mnt/petrelfs/zhangyuhan/data/evaluation/gallery510
+  text_gallery_path: /mnt/petrelfs/zhangyuhan/data/evaluation/gallery510/prompts_510.json
+  image_gallery_dir: /mnt/petrelfs/zhangyuhan/data/evaluation/gallery510/rgba
+  con_images_dir: /mnt/petrelfs/zhangyuhan/data/evaluation/con_images
+  split_con_image: false
+  cache_dir: null
+  reference_type_column_name: reference_type
+  reference_idx_column_name: reference_idx
+  reference_column_name: reference
+  normal_image_0_column_name: normal_image_0
+  normal_image_1_column_name: normal_image_1
+  rgb_image_0_column_name: rgb_image_0
+  rgb_image_1_column_name: rgb_image_1
+  eval_dims_column_name: eval_dims
+  label_0_column_name: label_0
+  label_1_column_name: label_1
+  are_different_column_name: are_different
+  has_label_column_name: has_label
+  reference_input_column_name: reference_input
+  normal_pixels_0_column_name: normal_pixel_values_0
+  normal_pixels_1_column_name: normal_pixel_values_1
+  rgb_pixels_0_column_name: rgb_pixel_values_0
+  rgb_pixels_1_column_name: rgb_pixel_values_1
+  num_examples_per_prompt_column_name: num_example_per_prompt
+  shuffle: true
+  keep_only_different: false
+  keep_only_with_label: false
+  keep_only_with_label_in_non_train: true
+  processor:
+    _target_: transformers.AutoProcessor.from_pretrained
+    pretrained_model_name_or_path: laion/CLIP-ViT-H-14-laion2B-s32B-b79K
+  limit_examples_per_prompt: -1
+  only_on_best: false
+optimizer:
+  _target_: trainer.optimizers.dummy_optimizer.BaseDummyOptim
+  lr: 3.0e-06
+  weight_decay: 0.3
+lr_scheduler:
+  _target_: trainer.lr_schedulers.dummy_lr_scheduler.instantiate_dummy_lr_scheduler
+  lr: 3.0e-06
+  lr_warmup_steps: 500
+  total_num_steps: 4000
+debug:
+  activate: false
+  port: 5900
+output_dir: outputs/Multi-CLIP_ViT_H_14-clean/from_con_image/stage2_eval_from_latest