Upload 17 files

model_index.json

@@ -0,0 +1,33 @@
+{
+  "_class_name": "StableDiffusionPipeline",
+  "_diffusers_version": "0.21.4",
+  "feature_extractor": [
+    null,
+    null
+  ],
+  "requires_safety_checker": null,
+  "safety_checker": [
+    null,
+    null
+  ],
+  "scheduler": [
+    "diffusers",
+    "DDPMScheduler"
+  ],
+  "text_encoder": [
+    "transformers",
+    "CLIPTextModel"
+  ],
+  "tokenizer": [
+    "transformers",
+    "CLIPTokenizer"
+  ],
+  "unet": [
+    "diffusers",
+    "UNet2DConditionModel"
+  ],
+  "vae": [
+    "diffusers",
+    "AutoencoderKL"
+  ]
+}

optimizer_rico.json

@@ -0,0 +1,43 @@
+{
+    "doc": {
+        "base": "base optimizer configuration for unet and text encoder",
+        "text_encoder_overrides": "text encoder config overrides",
+        "text_encoder_lr_scale": "if LR not set on text encoder, sets the Lr to a multiple of the Base LR. for example, if base `lr` is 2e-6 and `text_encoder_lr_scale` is 0.5, the text encoder's LR will be set to `1e-6`.",
+        "-----------------": "-----------------",
+        "optimizer": "adamw, adamw8bit, lion",
+        "optimizer_desc": "'adamw' in standard 32bit, 'adamw8bit' is bitsandbytes, 'lion' is lucidrains",
+        "lr": "learning rate, if null will use CLI or main JSON config value",
+        "lr_scheduler": "'constant' or 'cosine'",
+        "lr_warmup_steps": "number of steps to warmup LR to target LR, if null will use CLI or default a value based on max epochs",
+        "lr_decay_steps": "number of steps to decay LR to zero for cosine, if null will use CLI or default a value based on max epochs",
+        "betas": "exponential decay rates for the moment estimates",
+        "epsilon": "value added to denominator for numerical stability, unused for lion",
+        "weight_decay": "weight decay (L2 penalty)",
+        "------------------": "-----------------",
+        "unfreeze_last_n_layers": "if not null, freeze all parameters in the text encoder except for the last n layers and the final layer norm"
+    },
+    "base": {
+        "optimizer": "adamw8bit",
+        "lr": 1e-6,
+        "lr_scheduler": "constant",
+        "lr_decay_steps": null,
+        "lr_warmup_steps": null,
+        "betas": [0.9, 0.999],
+        "epsilon": 1e-8,
+        "weight_decay": 0.010
+    },
+    "text_encoder_overrides": {
+        "optimizer": null,
+        "lr": 5e-7,
+        "lr_scheduler": "cosine",
+        "lr_decay_steps": null,
+        "lr_warmup_steps": null,
+        "betas": null,
+        "epsilon": null,
+        "weight_decay": null
+    },
+    "text_encoder_freezing": {
+        "unfreeze_last_n_layers": null
+    },
+    "apply_grad_scaler_step_tweaks": true
+}

optimizer_rico_comp.json

@@ -0,0 +1,43 @@
+{
+    "doc": {
+        "base": "base optimizer configuration for unet and text encoder",
+        "text_encoder_overrides": "text encoder config overrides",
+        "text_encoder_lr_scale": "if LR not set on text encoder, sets the Lr to a multiple of the Base LR. for example, if base `lr` is 2e-6 and `text_encoder_lr_scale` is 0.5, the text encoder's LR will be set to `1e-6`.",
+        "-----------------": "-----------------",
+        "optimizer": "adamw, adamw8bit, lion",
+        "optimizer_desc": "'adamw' in standard 32bit, 'adamw8bit' is bitsandbytes, 'lion' is lucidrains",
+        "lr": "learning rate, if null will use CLI or main JSON config value",
+        "lr_scheduler": "'constant' or 'cosine'",
+        "lr_warmup_steps": "number of steps to warmup LR to target LR, if null will use CLI or default a value based on max epochs",
+        "lr_decay_steps": "number of steps to decay LR to zero for cosine, if null will use CLI or default a value based on max epochs",
+        "betas": "exponential decay rates for the moment estimates",
+        "epsilon": "value added to denominator for numerical stability, unused for lion",
+        "weight_decay": "weight decay (L2 penalty)",
+        "------------------": "-----------------",
+        "unfreeze_last_n_layers": "if not null, freeze all parameters in the text encoder except for the last n layers and the final layer norm"
+    },
+    "base": {
+        "optimizer": "adamw8bit",
+        "lr": 2.5e-6,
+        "lr_scheduler": "constant",
+        "lr_decay_steps": null,
+        "lr_warmup_steps": null,
+        "betas": [0.9, 0.999],
+        "epsilon": 1e-8,
+        "weight_decay": 0.010
+    },
+    "text_encoder_overrides": {
+        "optimizer": null,
+        "lr": 1.25e-6,
+        "lr_scheduler": "cosine",
+        "lr_decay_steps": null,
+        "lr_warmup_steps": null,
+        "betas": null,
+        "epsilon": null,
+        "weight_decay": null
+    },
+    "text_encoder_freezing": {
+        "unfreeze_last_n_layers": null
+    },
+    "apply_grad_scaler_step_tweaks": true
+}

rico_diffusion-20250309-203322_main.json

@@ -0,0 +1,112 @@
+{
+  "config": "rico_diffusion.json",
+  "attn_type": "sdp",
+  "batch_size": 7,
+  "ckpt_every_n_minutes": 1000000000.0,
+  "clip_grad_norm": null,
+  "clip_skip": 2,
+  "cond_dropout": 0.1,
+  "data_root": "/home/azhure/Tesis/save_data",
+  "disable_amp": false,
+  "disable_textenc_training": false,
+  "flip_p": 0.0,
+  "gpuid": 0,
+  "gradient_checkpointing": true,
+  "grad_accum": 1,
+  "logdir": "logs",
+  "log_step": 25,
+  "loss_type": "huber",
+  "max_epochs": 5,
+  "optimizer_config": "optimizer_rico.json",
+  "project_name": "rico_diffusion",
+  "resolution": 512,
+  "resume_ckpt": "logs/rico_diffusion_v2_comp_big-20250307-090306/ckpts/last-rico_diffusion_v2_comp_big-ep04-gs32932",
+  "sample_prompts": "sample_prompts_rico.json",
+  "sample_steps": 15000,
+  "save_ckpt_dir": "model_output",
+  "save_ckpts_from_n_epochs": 0,
+  "save_every_n_epochs": 5,
+  "save_optimizer": false,
+  "scale_lr": false,
+  "seed": 555,
+  "keep_tags": 3,
+  "shuffle_tags": false,
+  "validation_config": "validation_rico_off.json",
+  "zero_frequency_noise_ratio": 0.02,
+  "min_snr_gamma": 5.0,
+  "enable_zero_terminal_snr": true,
+  "amp": true,
+  "disable_unet_training": false,
+  "embedding_perturbation": 0.0,
+  "lr": null,
+  "lr_decay_steps": 124140,
+  "lr_scheduler": "constant",
+  "lr_warmup_steps": 2482,
+  "no_prepend_last": false,
+  "no_save_ckpt": false,
+  "plugins": null,
+  "run_name": null,
+  "save_full_precision": false,
+  "timestep_start": 0,
+  "timestep_end": 1000,
+  "train_sampler": "ddpm",
+  "wandb": false,
+  "write_schedule": false,
+  "rated_dataset": false,
+  "rated_dataset_target_dropout_percent": 50,
+  "load_settings_every_epoch": null,
+  "ema_decay_rate": null,
+  "ema_strength_target": null,
+  "ema_update_interval": 500,
+  "ema_device": "cpu",
+  "ema_sample_nonema_model": false,
+  "ema_sample_ema_model": false,
+  "ema_resume_model": null,
+  "pyramid_noise_discount": null,
+  "aspects": [
+    [
+      512,
+      512
+    ],
+    [
+      576,
+      448
+    ],
+    [
+      448,
+      576
+    ],
+    [
+      640,
+      384
+    ],
+    [
+      384,
+      640
+    ],
+    [
+      768,
+      320
+    ],
+    [
+      320,
+      768
+    ],
+    [
+      896,
+      256
+    ],
+    [
+      256,
+      896
+    ],
+    [
+      1024,
+      256
+    ],
+    [
+      256,
+      1024
+    ]
+  ]
+}

rico_diffusion-20250309-203322_opt.json

@@ -0,0 +1,32 @@
+{
+  "base": {
+    "optimizer": "adamw8bit",
+    "lr": 1e-06,
+    "lr_scheduler": "constant",
+    "lr_decay_steps": 124140,
+    "lr_warmup_steps": 2482,
+    "betas": [
+      0.9,
+      0.999
+    ],
+    "epsilon": 1e-08,
+    "weight_decay": 0.01
+  },
+  "text_encoder_overrides": {
+    "optimizer": "adamw8bit",
+    "lr": 5e-07,
+    "lr_scheduler": "cosine",
+    "lr_decay_steps": 124140,
+    "lr_warmup_steps": 2482,
+    "betas": [
+      0.9,
+      0.999
+    ],
+    "epsilon": 1e-08,
+    "weight_decay": 0.01
+  },
+  "text_encoder_freezing": {
+    "unfreeze_last_n_layers": null
+  },
+  "apply_grad_scaler_step_tweaks": true
+}

rico_diffusion.json

@@ -0,0 +1,37 @@
+{
+  "attn_type": "sdp",
+  "batch_size": 7,
+  "ckpt_every_n_minutes": null,
+  "clip_grad_norm": null,
+  "clip_skip": 2,
+  "cond_dropout": 0.10,
+  "data_root": "/home/azhure/Tesis/save_data",
+  "disable_amp": false,
+  "disable_textenc_training": false,
+  "flip_p": 0.0,
+  "gpuid": 0,
+  "gradient_checkpointing": true,
+  "grad_accum": 1,
+  "logdir": "logs",
+  "log_step": 25,
+  "loss_type": "huber",
+  "max_epochs": 5,
+  "optimizer_config": "optimizer_rico.json",
+  "project_name": "rico_diffusion",
+  "resolution": 512,
+  "resume_ckpt": "logs/rico_diffusion_v2_comp_big-20250307-090306/ckpts/last-rico_diffusion_v2_comp_big-ep04-gs32932",
+  "sample_prompts": "sample_prompts_rico.json",
+  "sample_steps": 15000,
+  "save_ckpt_dir": "model_output",
+  "save_ckpts_from_n_epochs": 0,
+  "save_every_n_epochs": 5,
+  "save_optimizer": false,
+  "scale_lr": false,
+  "seed": 555,
+  "keep_tags": 3,
+  "shuffle_tags": false,
+  "validation_config": "validation_rico_off.json",
+  "zero_frequency_noise_ratio": 0.02,
+  "min_snr_gamma": 5.0,
+  "enable_zero_terminal_snr": true
+}

rico_diffusion.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d0318d1fc435edcbfde87f7df5c892dd5a347aa719323d75a222ec5d9a595aeb
+size 2132625158

rico_diffusion_v2_comp.json

@@ -0,0 +1,38 @@
+{
+  "attn_type": "sdp",
+  "batch_size": 11,
+  "ckpt_every_n_minutes": null,
+  "clip_grad_norm": null,
+  "clip_skip": 2,
+  "cond_dropout": 0.00,
+  "data_root": "/home/azhure/Tesis/save_data_comp",
+  "disable_amp": false,
+  "disable_textenc_training": false,
+  "flip_p": 0.0,
+  "gpuid": 0,
+  "gradient_checkpointing": true,
+  "grad_accum": 1,
+  "logdir": "logs",
+  "log_step": 25,
+  "loss_type": "huber",
+  "max_epochs": 3,
+  "optimizer_config": "optimizer_rico_comp.json",
+  "project_name": "rico_diffusion_v2_comp",
+  "resolution": 384,
+  "resume_ckpt": "stable-diffusion-v1-5/stable-diffusion-v1-5",
+  "sample_prompts": "sample_prompts.txt",
+  "sample_steps": 999999999,
+  "save_ckpt_dir": "model_output",
+  "save_ckpts_from_n_epochs": 0,
+  "save_every_n_epochs": 10,
+  "save_optimizer": false,
+  "scale_lr": false,
+  "seed": 555,
+  "keep_tags": 3,
+  "shuffle_tags": false,
+  "validation_config": "validation_rico_off.json",
+  "zero_frequency_noise_ratio": 0.02,
+  "min_snr_gamma": 5.0,
+  "enable_zero_terminal_snr": true,
+  "no_save_ckpt": true
+}

rico_diffusion_v2_comp_big.json

@@ -0,0 +1,38 @@
+{
+  "attn_type": "sdp",
+  "batch_size": 9,
+  "ckpt_every_n_minutes": null,
+  "clip_grad_norm": null,
+  "clip_skip": 2,
+  "cond_dropout": 0.00,
+  "data_root": "/home/azhure/Tesis/save_data_comp_big",
+  "disable_amp": false,
+  "disable_textenc_training": false,
+  "flip_p": 0.0,
+  "gpuid": 0,
+  "gradient_checkpointing": true,
+  "grad_accum": 1,
+  "logdir": "logs",
+  "log_step": 25,
+  "loss_type": "huber",
+  "max_epochs": 4,
+  "optimizer_config": "optimizer_rico_comp.json",
+  "project_name": "rico_diffusion_v2_comp_big",
+  "resolution": 448,
+  "resume_ckpt": "findlast",
+  "sample_prompts": "sample_prompts.txt",
+  "sample_steps": 999999999,
+  "save_ckpt_dir": "model_output",
+  "save_ckpts_from_n_epochs": 0,
+  "save_every_n_epochs": 10,
+  "save_optimizer": false,
+  "scale_lr": false,
+  "seed": 555,
+  "keep_tags": 3,
+  "shuffle_tags": false,
+  "validation_config": "validation_rico_off.json",
+  "zero_frequency_noise_ratio": 0.02,
+  "min_snr_gamma": 5.0,
+  "enable_zero_terminal_snr": true,
+  "no_save_ckpt": true
+}

rico_diffusion_v2_comp_button.json

@@ -0,0 +1,38 @@
+{
+  "attn_type": "sdp",
+  "batch_size": 11,
+  "ckpt_every_n_minutes": null,
+  "clip_grad_norm": null,
+  "clip_skip": 2,
+  "cond_dropout": 0.00,
+  "data_root": "/home/azhure/Tesis/Button",
+  "disable_amp": false,
+  "disable_textenc_training": false,
+  "flip_p": 0.0,
+  "gpuid": 0,
+  "gradient_checkpointing": true,
+  "grad_accum": 1,
+  "logdir": "logs",
+  "log_step": 25,
+  "loss_type": "huber",
+  "max_epochs": 2,
+  "optimizer_config": "optimizer_rico_comp.json",
+  "project_name": "rico_diffusion_v2_comp_button",
+  "resolution": 384,
+  "resume_ckpt": "findlast",
+  "sample_prompts": "sample_prompts.txt",
+  "sample_steps": 999999999,
+  "save_ckpt_dir": "model_output",
+  "save_ckpts_from_n_epochs": 0,
+  "save_every_n_epochs": 10,
+  "save_optimizer": false,
+  "scale_lr": false,
+  "seed": 555,
+  "keep_tags": 3,
+  "shuffle_tags": false,
+  "validation_config": "validation_rico_off.json",
+  "zero_frequency_noise_ratio": 0.02,
+  "min_snr_gamma": 5.0,
+  "enable_zero_terminal_snr": true,
+  "no_save_ckpt": true
+}

rico_diffusion_v2_comp_icon.json

@@ -0,0 +1,38 @@
+{
+  "attn_type": "sdp",
+  "batch_size": 11,
+  "ckpt_every_n_minutes": null,
+  "clip_grad_norm": null,
+  "clip_skip": 2,
+  "cond_dropout": 0.00,
+  "data_root": "/home/azhure/Tesis/Icon",
+  "disable_amp": false,
+  "disable_textenc_training": false,
+  "flip_p": 0.0,
+  "gpuid": 0,
+  "gradient_checkpointing": true,
+  "grad_accum": 1,
+  "logdir": "logs",
+  "log_step": 25,
+  "loss_type": "huber",
+  "max_epochs": 2,
+  "optimizer_config": "optimizer_rico_comp.json",
+  "project_name": "rico_diffusion_v2_comp_icon",
+  "resolution": 384,
+  "resume_ckpt": "findlast",
+  "sample_prompts": "sample_prompts.txt",
+  "sample_steps": 999999999,
+  "save_ckpt_dir": "model_output",
+  "save_ckpts_from_n_epochs": 0,
+  "save_every_n_epochs": 10,
+  "save_optimizer": false,
+  "scale_lr": false,
+  "seed": 555,
+  "keep_tags": 3,
+  "shuffle_tags": false,
+  "validation_config": "validation_rico_off.json",
+  "zero_frequency_noise_ratio": 0.02,
+  "min_snr_gamma": 5.0,
+  "enable_zero_terminal_snr": true,
+  "no_save_ckpt": true
+}

rico_diffusion_v2_comp_image.json

@@ -0,0 +1,38 @@
+{
+  "attn_type": "sdp",
+  "batch_size": 11,
+  "ckpt_every_n_minutes": null,
+  "clip_grad_norm": null,
+  "clip_skip": 2,
+  "cond_dropout": 0.00,
+  "data_root": "/home/azhure/Tesis/Image",
+  "disable_amp": false,
+  "disable_textenc_training": false,
+  "flip_p": 0.0,
+  "gpuid": 0,
+  "gradient_checkpointing": true,
+  "grad_accum": 1,
+  "logdir": "logs",
+  "log_step": 25,
+  "loss_type": "huber",
+  "max_epochs": 2,
+  "optimizer_config": "optimizer_rico_comp.json",
+  "project_name": "rico_diffusion_v2_comp_image",
+  "resolution": 384,
+  "resume_ckpt": "findlast",
+  "sample_prompts": "sample_prompts.txt",
+  "sample_steps": 999999999,
+  "save_ckpt_dir": "model_output",
+  "save_ckpts_from_n_epochs": 0,
+  "save_every_n_epochs": 10,
+  "save_optimizer": false,
+  "scale_lr": false,
+  "seed": 555,
+  "keep_tags": 3,
+  "shuffle_tags": false,
+  "validation_config": "validation_rico_off.json",
+  "zero_frequency_noise_ratio": 0.02,
+  "min_snr_gamma": 5.0,
+  "enable_zero_terminal_snr": true,
+  "no_save_ckpt": true
+}

rico_diffusion_v2_comp_list_item.json

@@ -0,0 +1,38 @@
+{
+  "attn_type": "sdp",
+  "batch_size": 11,
+  "ckpt_every_n_minutes": null,
+  "clip_grad_norm": null,
+  "clip_skip": 2,
+  "cond_dropout": 0.00,
+  "data_root": "/home/azhure/Tesis/List_Item",
+  "disable_amp": false,
+  "disable_textenc_training": false,
+  "flip_p": 0.0,
+  "gpuid": 0,
+  "gradient_checkpointing": true,
+  "grad_accum": 1,
+  "logdir": "logs",
+  "log_step": 25,
+  "loss_type": "huber",
+  "max_epochs": 2,
+  "optimizer_config": "optimizer_rico_comp.json",
+  "project_name": "rico_diffusion_v2_comp_list_item",
+  "resolution": 384,
+  "resume_ckpt": "findlast",
+  "sample_prompts": "sample_prompts.txt",
+  "sample_steps": 999999999,
+  "save_ckpt_dir": "model_output",
+  "save_ckpts_from_n_epochs": 0,
+  "save_every_n_epochs": 10,
+  "save_optimizer": false,
+  "scale_lr": false,
+  "seed": 555,
+  "keep_tags": 3,
+  "shuffle_tags": false,
+  "validation_config": "validation_rico_off.json",
+  "zero_frequency_noise_ratio": 0.02,
+  "min_snr_gamma": 5.0,
+  "enable_zero_terminal_snr": true,
+  "no_save_ckpt": true
+}

rico_diffusion_v2_full.json

@@ -0,0 +1,37 @@
+{
+  "attn_type": "sdp",
+  "batch_size": 7,
+  "ckpt_every_n_minutes": null,
+  "clip_grad_norm": null,
+  "clip_skip": 2,
+  "cond_dropout": 0.10,
+  "data_root": "/home/azhure/Tesis/save_data",
+  "disable_amp": false,
+  "disable_textenc_training": false,
+  "flip_p": 0.0,
+  "gpuid": 0,
+  "gradient_checkpointing": true,
+  "grad_accum": 1,
+  "logdir": "logs",
+  "log_step": 25,
+  "loss_type": "huber",
+  "max_epochs": 5,
+  "optimizer_config": "optimizer_rico.json",
+  "project_name": "rico_diffusion_v2_full",
+  "resolution": 512,
+  "resume_ckpt": "findlast",
+  "sample_prompts": "sample_prompts_rico.json",
+  "sample_steps": 15000,
+  "save_ckpt_dir": "model_output",
+  "save_ckpts_from_n_epochs": 0,
+  "save_every_n_epochs": 5,
+  "save_optimizer": false,
+  "scale_lr": false,
+  "seed": 555,
+  "keep_tags": 3,
+  "shuffle_tags": false,
+  "validation_config": "validation_rico.json",
+  "zero_frequency_noise_ratio": 0.02,
+  "min_snr_gamma": 5.0,
+  "enable_zero_terminal_snr": true
+}

sample_prompts_rico.json

@@ -0,0 +1,19 @@
+{
+  "batch_size": 1,		
+  "seed": 555,
+  "cfgs": [7],
+  "scheduler": "ddim",		
+  "num_inference_steps": 20,	
+  "show_progress_bars": true, 
+  "generate_pretrain_samples": true,
+  "samples": [
+    {
+      "prompt": "red Toolbar Upper Top containing Text Left Login inside and white Input Upper Top and white Input Lower Top and red Button login Upper Middle, Android UI, Medical, white background",      
+      "size": [384, 640]
+    },
+    {
+      "prompt": "blue Toolbar Upper Top containing Text League Left inside and blue Multi Tab Top containing Text Button News Left Side Left and Text Button Schedule Right Side Left inside and white Card Middle containing Image soccer Cover Top and Text Soccer Match Bottom Left inside and white Card Bottom containing Image basketball Cover Top and Text Basketball Match Bottom Left inside, Android UI, Sports, white background",  
+      "size": [384, 640]
+    }
+  ]
+}

validation_rico.json

@@ -0,0 +1,24 @@
+{
+    "documentation": {
+        "validate_training": "If true, validate the training using a separate set of image/caption pairs, and log the results as `loss/val`. The curve will trend downwards as the model trains, then flatten and start to trend upwards as effective training finishes and the model begins to overfit the training data. Very useful for preventing overfitting, for checking if your learning rate is too low or too high, and for deciding when to stop training.",
+        "val_split_mode": "Either 'automatic' or 'manual', ignored if validate_training is false. 'automatic' val_split_mode picks a random subset of the training set (the number of items is controlled by auto_split_proportion) and removes them from training to use as a validation set. 'manual' val_split_mode lets you provide your own folder of validation items (images+captions), specified using 'manual_data_root'.",
+        "auto_split_proportion": "For 'automatic' val_split_mode, how much of the train dataset that should be removed to use for validation. Typical values are 0.15-0.2 (15-20% of the total dataset). Higher is more accurate but slower.",
+        "manual_data_root": "For 'manual' val_split_mode, the path to a folder containing validation items.",
+        "extra_manual_datasets": "Dictionary of 'name':'path' pairs defining additional validation datasets to load and log. eg { 'santa_suit': '/path/to/captioned_santa_suit_images', 'flamingo_suit': '/path/to/flamingo_suit_images' }",
+        "stabilize_training_loss": "If true, stabilize the train loss curves for `loss/epoch` and `loss/log step` by re-calculating training loss with a fixed random seed, and log the results as `loss/train-stabilized`. This more clearly shows the training progress, but it is not enough alone to tell you if you're overfitting.",
+        "stabilize_split_proportion": "For stabilize_training_loss, the proportion of the train dataset to overlap for stabilizing the train loss graph. Typical values are 0.15-0.2 (15-20% of the total dataset). Higher is more accurate but slower.",
+        "every_n_epochs": "How often to run validation (1=every epoch, 2=every second epoch; 0.5=twice per epoch, 0.33=three times per epoch, etc.).",
+        "seed": "The seed to use when running validation and stabilization passes.",
+        "use_relative_loss": "logs val/loss as negative relative to first pre-train val/loss value"
+    },
+    "validate_training": true,
+    "val_split_mode": "automatic",
+    "auto_split_proportion": 0.05,
+    "manual_data_root": null,
+    "extra_manual_datasets" : {},
+    "stabilize_training_loss": true,
+    "stabilize_split_proportion": 0.05,
+    "every_n_epochs": 1,
+    "seed": 555,
+    "use_relative_loss": false
+}

validation_rico_off.json

@@ -0,0 +1,24 @@
+{
+    "documentation": {
+        "validate_training": "If true, validate the training using a separate set of image/caption pairs, and log the results as `loss/val`. The curve will trend downwards as the model trains, then flatten and start to trend upwards as effective training finishes and the model begins to overfit the training data. Very useful for preventing overfitting, for checking if your learning rate is too low or too high, and for deciding when to stop training.",
+        "val_split_mode": "Either 'automatic' or 'manual', ignored if validate_training is false. 'automatic' val_split_mode picks a random subset of the training set (the number of items is controlled by auto_split_proportion) and removes them from training to use as a validation set. 'manual' val_split_mode lets you provide your own folder of validation items (images+captions), specified using 'manual_data_root'.",
+        "auto_split_proportion": "For 'automatic' val_split_mode, how much of the train dataset that should be removed to use for validation. Typical values are 0.15-0.2 (15-20% of the total dataset). Higher is more accurate but slower.",
+        "manual_data_root": "For 'manual' val_split_mode, the path to a folder containing validation items.",
+        "extra_manual_datasets": "Dictionary of 'name':'path' pairs defining additional validation datasets to load and log. eg { 'santa_suit': '/path/to/captioned_santa_suit_images', 'flamingo_suit': '/path/to/flamingo_suit_images' }",
+        "stabilize_training_loss": "If true, stabilize the train loss curves for `loss/epoch` and `loss/log step` by re-calculating training loss with a fixed random seed, and log the results as `loss/train-stabilized`. This more clearly shows the training progress, but it is not enough alone to tell you if you're overfitting.",
+        "stabilize_split_proportion": "For stabilize_training_loss, the proportion of the train dataset to overlap for stabilizing the train loss graph. Typical values are 0.15-0.2 (15-20% of the total dataset). Higher is more accurate but slower.",
+        "every_n_epochs": "How often to run validation (1=every epoch, 2=every second epoch; 0.5=twice per epoch, 0.33=three times per epoch, etc.).",
+        "seed": "The seed to use when running validation and stabilization passes.",
+        "use_relative_loss": "logs val/loss as negative relative to first pre-train val/loss value"
+    },
+    "validate_training": false,
+    "val_split_mode": "automatic",
+    "auto_split_proportion": 0.05,
+    "manual_data_root": null,
+    "extra_manual_datasets" : {},
+    "stabilize_training_loss": false,
+    "stabilize_split_proportion": 0.05,
+    "every_n_epochs": 1,
+    "seed": 555,
+    "use_relative_loss": false
+}