Merge pull request #15 from neph1/update-v0.9.3

add more config options

Author: neph1

config/config_categories.yaml

@@ -1,4 +1,4 @@
-Dataset: data_root, video_column, caption_column, id_token, video_resolution_buckets, caption_dropout_p
+Dataset: data_root, video_column, caption_column, dataset_file, id_token, image_resolution_buckets, video_resolution_buckets, caption_dropout_p
 Training: training_type, seed, mixed_precision, train_steps, rank, lora_alpha, target_modules, gradient_accumulation_steps, checkpointing_steps, checkpointing_limit, enable_slicing, enable_tiling, batch_size, resume_from_checkpoint
 Optimizer: optimizer, lr, beta1, beta2, epsilon, weight_decay, max_grad_norm, lr_scheduler, lr_num_cycles, lr_warmup_steps
 Validation: validation_steps, validation_epochs, num_validation_videos, validation_prompts, validation_prompt_separator

config/config_template.yaml

@@ -10,14 +10,17 @@ checkpointing_limit: 102
 checkpointing_steps: 500
 data_root: ''
 dataloader_num_workers: 0
+dataset_file: ''
 diffusion_options: ''
+enable_model_cpu_offload: false
 enable_slicing: true
 enable_tiling: true
 epsilon: 1e-8
 gpu_ids: '0'
 gradient_accumulation_steps: 4
 gradient_checkpointing: true
 id_token: afkx
+image_resolution_buckets: 512x768
 lora_alpha: 128
 lr: 0.0001
 lr_num_cycles: 1

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "finetrainers-ui"
-version = "0.9.1"
+version = "0.9.3"
 dependencies = [
     "gradio",
     "torch>=2.4.1"

run_trainer.py

@@ -28,13 +28,17 @@ def run(self, config: Config, finetrainers_path: str, log_file: str):
                    "--id_token", config.get('id_token'),
                    "--video_resolution_buckets"]
         dataset_cmd += config.get('video_resolution_buckets').split(' ')
+        dataset_cmd += ["--image_resolution_buckets"]
+        dataset_cmd += config.get('image_resolution_buckets').split(' ')
         dataset_cmd += ["--caption_dropout_p", config.get('caption_dropout_p'),
                    "--caption_dropout_technique", config.get('caption_dropout_technique'),
                    "--text_encoder_dtype", config.get('text_encoder_dtype'),
                    "--text_encoder_2_dtype", config.get('text_encoder_2_dtype'),
                    "--text_encoder_3_dtype", config.get('text_encoder_3_dtype'),
                    "--vae_dtype", config.get('vae_dtype'),
                    '--precompute_conditions' if config.get('precompute_conditions') else '']
+        if config.get('dataset_file'):
+            dataset_cmd += ["--dataset_file", config.get('dataset_file')]
 
         dataloader_cmd = ["--dataloader_num_workers", config.get('dataloader_num_workers')]
 
@@ -56,6 +60,8 @@ def run(self, config: Config, finetrainers_path: str, log_file: str):
                 "--checkpointing_limit", config.get('checkpointing_limit'),
                 '--enable_slicing' if config.get('enable_slicing') else '',
                 '--enable_tiling' if config.get('enable_tiling') else '']
+        if config.get('enable_model_cpu_offload'):
+            training_cmd += ["--enable_model_cpu_offload"]
 
         if config.get('resume_from_checkpoint'):
             training_cmd += ["--resume_from_checkpoint", config.get('resume_from_checkpoint')]