LoRA_type
stringclasses 1
value | LyCORIS_preset
stringclasses 1
value | adaptive_noise_scale
int64 0
0
| additional_parameters
stringclasses 1
value | async_upload
bool 1
class | block_alphas
stringclasses 1
value | block_dims
stringclasses 1
value | block_lr_zero_threshold
stringclasses 1
value | bucket_no_upscale
bool 1
class | bucket_reso_steps
int64 64
64
| bypass_mode
bool 1
class | cache_latents
bool 1
class | cache_latents_to_disk
bool 1
class | caption_dropout_every_n_epochs
int64 0
0
| caption_dropout_rate
int64 0
0
| caption_extension
stringclasses 1
value | clip_skip
int64 1
1
| color_aug
bool 1
class | constrain
int64 0
0
| conv_alpha
int64 1
1
| conv_block_alphas
stringclasses 1
value | conv_block_dims
stringclasses 1
value | conv_dim
int64 1
1
| dataset_config
stringclasses 1
value | debiased_estimation_loss
bool 1
class | decompose_both
bool 1
class | dim_from_weights
bool 1
class | dora_wd
bool 1
class | down_lr_weight
stringclasses 1
value | dynamo_backend
stringclasses 1
value | dynamo_mode
stringclasses 1
value | dynamo_use_dynamic
bool 1
class | dynamo_use_fullgraph
bool 1
class | enable_bucket
bool 1
class | epoch
int64 100
100
| extra_accelerate_launch_args
stringclasses 1
value | factor
int64 -1
-1
| flip_aug
bool 1
class | fp8_base
bool 1
class | full_bf16
bool 1
class | full_fp16
bool 1
class | gpu_ids
stringclasses 1
value | gradient_accumulation_steps
int64 1
1
| gradient_checkpointing
bool 1
class | huber_c
float64 0.1
0.1
| huber_schedule
stringclasses 1
value | huggingface_path_in_repo
stringclasses 1
value | huggingface_repo_id
stringclasses 1
value | huggingface_repo_type
stringclasses 1
value | huggingface_repo_visibility
stringclasses 1
value | huggingface_token
stringclasses 1
value | ip_noise_gamma
int64 0
0
| ip_noise_gamma_random_strength
bool 1
class | keep_tokens
int64 0
0
| learning_rate
float64 0
0
| log_tracker_config
stringclasses 1
value | log_tracker_name
stringclasses 1
value | log_with
stringclasses 1
value | logging_dir
stringclasses 1
value | loss_type
stringclasses 1
value | lr_scheduler
stringclasses 1
value | lr_scheduler_args
stringclasses 1
value | lr_scheduler_num_cycles
int64 1
1
| lr_scheduler_power
int64 1
1
| lr_warmup
int64 10
10
| main_process_port
int64 0
0
| masked_loss
bool 1
class | max_bucket_reso
int64 2.05k
2.05k
| max_data_loader_n_workers
int64 0
0
| max_grad_norm
int64 1
1
| max_resolution
stringclasses 1
value | max_timestep
int64 1k
1k
| max_token_length
int64 75
75
| max_train_epochs
int64 0
0
| max_train_steps
int64 1.6k
1.6k
| mem_eff_attn
bool 1
class | metadata_author
stringclasses 1
value | metadata_description
stringclasses 1
value | metadata_license
stringclasses 1
value | metadata_tags
stringclasses 1
value | metadata_title
stringclasses 1
value | mid_lr_weight
stringclasses 1
value | min_bucket_reso
int64 256
256
| min_snr_gamma
int64 0
0
| min_timestep
int64 0
0
| mixed_precision
stringclasses 1
value | model_list
stringclasses 1
value | module_dropout
int64 0
0
| multi_gpu
bool 1
class | multires_noise_discount
float64 0.3
0.3
| multires_noise_iterations
int64 0
0
| network_alpha
int64 32
32
| network_dim
int64 64
64
| network_dropout
int64 0
0
| network_weights
stringclasses 1
value | noise_offset
int64 0
0
| noise_offset_random_strength
bool 1
class | noise_offset_type
stringclasses 1
value | num_cpu_threads_per_process
int64 2
2
| num_machines
int64 1
1
| num_processes
int64 1
1
| optimizer
stringclasses 1
value | optimizer_args
stringclasses 1
value | output_dir
stringclasses 1
value | output_name
stringclasses 1
value | persistent_data_loader_workers
bool 1
class | pretrained_model_name_or_path
stringclasses 1
value | prior_loss_weight
int64 1
1
| random_crop
bool 1
class | rank_dropout
int64 0
0
| rank_dropout_scale
bool 1
class | reg_data_dir
stringclasses 1
value | rescaled
bool 1
class | resume
stringclasses 1
value | resume_from_huggingface
stringclasses 1
value | sample_every_n_epochs
int64 0
0
| sample_every_n_steps
int64 0
0
| sample_prompts
stringclasses 1
value | sample_sampler
stringclasses 1
value | save_as_bool
bool 1
class | save_every_n_epochs
int64 1
1
| save_every_n_steps
int64 0
0
| save_last_n_steps
int64 0
0
| save_last_n_steps_state
int64 0
0
| save_model_as
stringclasses 1
value | save_precision
stringclasses 1
value | save_state
bool 1
class | save_state_on_train_end
bool 1
class | save_state_to_huggingface
bool 1
class | scale_v_pred_loss_like_noise_pred
bool 1
class | scale_weight_norms
int64 0
0
| sdxl
bool 1
class | sdxl_cache_text_encoder_outputs
bool 1
class | sdxl_no_half_vae
bool 1
class | seed
int64 0
0
| shuffle_caption
bool 1
class | stop_text_encoder_training
int64 0
0
| text_encoder_lr
float64 0
0
| train_batch_size
int64 1
1
| train_data_dir
stringclasses 1
value | train_norm
bool 1
class | train_on_input
bool 1
class | training_comment
stringclasses 1
value | unet_lr
float64 0
0
| unit
int64 1
1
| up_lr_weight
stringclasses 1
value | use_cp
bool 1
class | use_scalar
bool 1
class | use_tucker
bool 1
class | v2
bool 1
class | v_parameterization
bool 1
class | v_pred_like_loss
int64 0
0
| vae
stringclasses 1
value | vae_batch_size
int64 0
0
| wandb_api_key
stringclasses 1
value | wandb_run_name
stringclasses 1
value | weighted_captions
bool 1
class | xformers
stringclasses 1
value |
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Standard | full | 0 | false | true | 64 | false | true | false | 0 | 0 | .txt | 1 | false | 0 | 1 | 1 | false | false | false | false | no | default | false | false | true | 100 | -1 | false | false | false | false | 1 | false | 0.1 | snr | 0 | false | 0 | 0.0001 | l2 | cosine | 1 | 1 | 10 | 0 | false | 2,048 | 0 | 1 | 512,512 | 1,000 | 75 | 0 | 1,600 | false | 256 | 0 | 0 | fp16 | 0 | false | 0.3 | 0 | 32 | 64 | 0 | 0 | false | Original | 2 | 1 | 1 | AdamW8bit | /workspace/kohya_ss/outputs | wpt | false | /workspace/kohya_ss/models/deliberate_v6.safetensors | 1 | false | 0 | false | false | 0 | 0 | euler_a | false | 1 | 0 | 0 | 0 | safetensors | fp16 | false | false | false | false | 0 | false | false | false | 0 | false | 0 | 0.0001 | 1 | /workspace/kohya_ss/dataset/images/ | false | true | 0.0001 | 1 | false | false | false | false | false | 0 | 0 | false | xformers |
||||||||||||||||||||||||||||||||||||||
Standard | full | 0 | false | true | 64 | false | true | false | 0 | 0 | .txt | 1 | false | 0 | 1 | 1 | false | false | false | false | no | default | false | false | true | 100 | -1 | false | false | false | false | 1 | false | 0.1 | snr | 0 | false | 0 | 0.0001 | l2 | cosine | 1 | 1 | 10 | 0 | false | 2,048 | 0 | 1 | 512,512 | 1,000 | 75 | 0 | 1,600 | false | 256 | 0 | 0 | fp16 | 0 | false | 0.3 | 0 | 32 | 64 | 0 | 0 | false | Original | 2 | 1 | 1 | AdamW8bit | /workspace/kohya_ss/outputs | wpt | false | /workspace/kohya_ss/models/deliberate_v6.safetensors | 1 | false | 0 | false | false | 0 | 0 | euler_a | false | 1 | 0 | 0 | 0 | safetensors | fp16 | false | false | false | false | 0 | false | false | false | 0 | false | 0 | 0.0001 | 1 | /workspace/kohya_ss/dataset/images/ | false | true | 0.0001 | 1 | false | false | false | false | false | 0 | 0 | false | xformers |
Hi, community!
I'm a newcomer to model creation and LoRA fine-tuning. My goal is to create a model that can generate realistic images of tennis players in action. I've already made about 10 attempts to train a LoRA using a dataset I collected myself. I've tried different settings and parameters during the training, but unfortunately, I haven't yet achieved the desired results.
I understand that creating a quality model is not always easy and takes time and experience. I'm here to learn and to get advice from more experienced community members. I would appreciate any recommendations or feedback on my approach. I want to learn how to do it right and hopefully, I'll be able to achieve my goal.
If anyone can share their thoughts on what I could change or improve in my process, I would be very grateful!
Thanks for your attention and support.
RU
Привет, сообщество!
Я совсем недавно начал заниматься созданием моделей и дообучением LoRA. Моя цель — создать модель, которая сможет генерировать реалистичные изображения теннисисток во время игры. Я уже провел около 10 попыток тренировки LoRA с использованием набора данных, который я собрал сам. В процессе обучения я использовал различные настройки и параметры, но, к сожалению, пока что не получил желаемого результата.
Понимаю, что путь создания качественной модели не всегда прост и требует времени и опыта. Я здесь, чтобы учиться и получать советы от более опытных участников сообщества. Буду благодарен за любые рекомендации или критику по моему подходу. Хочу научиться делать это правильно и, надеюсь, смогу добиться результата.
Если кто-то может поделиться своими мыслями о том, что я мог бы изменить или улучшить в своем процессе, я буду очень признателен!
Спасибо за ваше внимание и поддержку.
- Downloads last month
- 66