Publish MOSS-Image experiment recipes, adapter checkpoints, and validation previews

README.md

@@ -48,5 +48,5 @@ the JSON runs use the shared schema implementation from [PR #3273](https://githu
 | [base-1024](https://huggingface.co/SimpleTuner/moss-image-base-domokun-1024) | 10,000 / 10,000 | [10,000](https://huggingface.co/SimpleTuner/moss-image-base-domokun-1024/tree/main/checkpoint-10000) | completed |
 | [sft-512](https://huggingface.co/SimpleTuner/moss-image-sft-domokun-512) | 10,000 / 10,000 | [10,000](https://huggingface.co/SimpleTuner/moss-image-sft-domokun-512/tree/main/checkpoint-10000) | completed |
 | [sft-1024](https://huggingface.co/SimpleTuner/moss-image-sft-domokun-1024) | 10,000 / 10,000 | [10,000](https://huggingface.co/SimpleTuner/moss-image-sft-domokun-1024/tree/main/checkpoint-10000) | completed |
-| [json-base-512](https://huggingface.co/SimpleTuner/moss-image-base-domokun-json-512) | 0 / 10,000 | — | queued |
-| [json-sft-512](https://huggingface.co/SimpleTuner/moss-image-sft-domokun-json-512) | 0 / 10,000 | — | queued |
+| [json-base-512](https://huggingface.co/SimpleTuner/moss-image-base-domokun-json-512) | 0 / 10,000 | — | failed |
+| [json-sft-512](https://huggingface.co/SimpleTuner/moss-image-sft-domokun-json-512) | 0 / 10,000 | — | starting |

manifest.json

@@ -290,20 +290,20 @@
       "repo_id": "SimpleTuner/moss-image-base-domokun-json-512",
       "flavour": "base",
       "resolution": 512,
-      "status": "queued",
+      "status": "failed",
       "observed_step": 0,
       "checkpoints": {},
-      "revision": "7cf3b4db55906c5449ba74f89ef18f8cdd1b72b2"
+      "revision": "fe45b3f2f86a8d4c61720e6df18581732843dd3b"
     },
     {
       "name": "json-sft-512",
       "repo_id": "SimpleTuner/moss-image-sft-domokun-json-512",
       "flavour": "sft",
       "resolution": 512,
-      "status": "queued",
+      "status": "starting",
       "observed_step": 0,
       "checkpoints": {},
-      "revision": "3ee4ed803eb4cdfd2eac4391e8e724c698f8893d"
+      "revision": "240b8e70c9acce7c951377684eec4ff793b84e05"
     }
   ]
 }

recipes/json-base-512/dataloader.json

@@ -0,0 +1,23 @@
+[
+  {
+    "id": "domokun",
+    "type": "local",
+    "dataset_type": "image",
+    "instance_data_dir": "datasets/domokun",
+    "caption_strategy": "textfile",
+    "resolution": 0.262144,
+    "resolution_type": "area",
+    "crop": false,
+    "minimum_image_size": 0,
+    "repeats": 10,
+    "cache_dir_vae": "cache/json-base-512/vae",
+    "text_embeds": "text"
+  },
+  {
+    "id": "text",
+    "type": "local",
+    "dataset_type": "text_embeds",
+    "default": true,
+    "cache_dir": "cache/json-base-512/text"
+  }
+]

recipes/json-base-512/prompts.json

@@ -0,0 +1,6 @@
+{
+  "studio": "A studio photograph of 🟫 standing on a wooden table, soft window light, with the entire character visible against a pale gray background.",
+  "garden": "A photograph of 🟫 standing in a sunny garden surrounded by flowers, natural colors and a softly blurred background.",
+  "astronaut": "An illustration of 🟫 wearing a white astronaut suit on the surface of the moon, Earth visible in the sky.",
+  "retention": "A photograph of a cobalt blue ceramic teapot on a wooden table beside a window, soft natural light."
+}

recipes/json-base-512/training_config.json

@@ -0,0 +1,93 @@
+{
+  "base_model_precision": "no_change",
+  "caption_dropout_probability": 0.0,
+  "checkpoint_step_interval": 500,
+  "checkpoints_total_limit": 20,
+  "compress_disk_cache": false,
+  "data_backend_config": "dataloader.json",
+  "disable_benchmark": true,
+  "disable_bucket_pruning": true,
+  "gradient_checkpointing": true,
+  "hub_model_id": "SimpleTuner/moss-image-base-domokun-json-512",
+  "strict_epoch_limit": false,
+  "learning_rate": 0.0001,
+  "lora_alpha": 32,
+  "lora_rank": 32,
+  "lora_type": "standard",
+  "lr_scheduler": "constant_with_warmup",
+  "lr_warmup_steps": 25,
+  "max_grad_norm": 1.0,
+  "max_train_steps": 10000,
+  "minimum_image_size": 0,
+  "mixed_precision": "bf16",
+  "model_family": "moss_image",
+  "model_flavour": "base",
+  "model_type": "lora",
+  "num_train_epochs": 0,
+  "optimizer": "adamw_bf16",
+  "output_dir": "output/json-base-512",
+  "push_checkpoints_to_hub": false,
+  "push_to_hub": false,
+  "report_to": "none",
+  "resolution": 512,
+  "resolution_type": "pixel_area",
+  "seed": 42,
+  "skip_file_discovery": false,
+  "tracker_project_name": "lora-training",
+  "tracker_run_name": "moss-domokun-json-base-512",
+  "train_batch_size": 4,
+  "use_ema": false,
+  "vae_batch_size": 1,
+  "validation_guidance": 3.0,
+  "validation_guidance_rescale": 0.0,
+  "validation_lycoris_strength": 1.0,
+  "validation_num_inference_steps": 50,
+  "validation_prompt_library": false,
+  "validation_resolution": "512x512",
+  "validation_seed": 42,
+  "validation_guidance_real": 3.0,
+  "user_prompt_library": "prompts.json",
+  "dynamo_backend": "inductor",
+  "dynamo_mode": "default",
+  "dynamo_use_regional_compilation": true,
+  "validation_disable_unconditional": true,
+  "validation_negative_prompt": "",
+  "gradient_checkpointing_interval": 2,
+  "gradient_accumulation_steps": 1,
+  "grad_clip_method": "norm",
+  "lora_dropout": 0.0,
+  "vae_enable_tiling": false,
+  "num_validation_images": 1,
+  "dynamo_dynamic": true,
+  "crepa_enabled": true,
+  "crepa_block_index": 8,
+  "crepa_lambda": 0.5,
+  "crepa_encoder_image_size": 518,
+  "crepa_adjacent_distance": 0,
+  "crepa_spatial_align": true,
+  "crepa_encoder": "dinov2_vitg14",
+  "data_backend_sampling": "uniform",
+  "aspect_bucket_alignment": 16,
+  "dataloader_prefetch": false,
+  "flow_schedule_auto_shift": false,
+  "flow_schedule_shift": 1.0,
+  "validation_step_interval": 500,
+  "resume_from_checkpoint": "",
+  "vae_dtype": "fp32",
+  "tokenizer_max_length": 4096,
+  "train_text_encoder": false,
+  "moss_image_validation_shift": 5.0,
+  "moss_image_uncond_skip_layers": [
+    22
+  ],
+  "gradient_checkpointing_backend": "torch",
+  "gradient_checkpointing_offload_attention": false,
+  "pretrained_model_name_or_path": "OpenMOSS-Team/MOSS-Image-Base",
+  "model_card_base_model": "OpenMOSS-Team/MOSS-Image-Base",
+  "validation_on_startup": true,
+  "disable_assistant_lora": true,
+  "attention_mechanism": "kohaku-fa-auto",
+  "moss_image_auto_json": true,
+  "moss_image_prompt_upsample": false,
+  "revision": "1ceede80badddd6a7812069619edaa5633587192"
+}

recipes/json-sft-512/dataloader.json

@@ -0,0 +1,23 @@
+[
+  {
+    "id": "domokun",
+    "type": "local",
+    "dataset_type": "image",
+    "instance_data_dir": "datasets/domokun",
+    "caption_strategy": "textfile",
+    "resolution": 0.262144,
+    "resolution_type": "area",
+    "crop": false,
+    "minimum_image_size": 0,
+    "repeats": 10,
+    "cache_dir_vae": "cache/json-sft-512/vae",
+    "text_embeds": "text"
+  },
+  {
+    "id": "text",
+    "type": "local",
+    "dataset_type": "text_embeds",
+    "default": true,
+    "cache_dir": "cache/json-sft-512/text"
+  }
+]

recipes/json-sft-512/prompts.json

@@ -0,0 +1,6 @@
+{
+  "studio": "A studio photograph of 🟫 standing on a wooden table, soft window light, with the entire character visible against a pale gray background.",
+  "garden": "A photograph of 🟫 standing in a sunny garden surrounded by flowers, natural colors and a softly blurred background.",
+  "astronaut": "An illustration of 🟫 wearing a white astronaut suit on the surface of the moon, Earth visible in the sky.",
+  "retention": "A photograph of a cobalt blue ceramic teapot on a wooden table beside a window, soft natural light."
+}

recipes/json-sft-512/training_config.json

@@ -0,0 +1,93 @@
+{
+  "base_model_precision": "no_change",
+  "caption_dropout_probability": 0.0,
+  "checkpoint_step_interval": 500,
+  "checkpoints_total_limit": 20,
+  "compress_disk_cache": false,
+  "data_backend_config": "dataloader.json",
+  "disable_benchmark": true,
+  "disable_bucket_pruning": true,
+  "gradient_checkpointing": true,
+  "hub_model_id": "SimpleTuner/moss-image-sft-domokun-json-512",
+  "strict_epoch_limit": false,
+  "learning_rate": 0.0001,
+  "lora_alpha": 32,
+  "lora_rank": 32,
+  "lora_type": "standard",
+  "lr_scheduler": "constant_with_warmup",
+  "lr_warmup_steps": 25,
+  "max_grad_norm": 1.0,
+  "max_train_steps": 10000,
+  "minimum_image_size": 0,
+  "mixed_precision": "bf16",
+  "model_family": "moss_image",
+  "model_flavour": "sft",
+  "model_type": "lora",
+  "num_train_epochs": 0,
+  "optimizer": "adamw_bf16",
+  "output_dir": "output/json-sft-512",
+  "push_checkpoints_to_hub": false,
+  "push_to_hub": false,
+  "report_to": "none",
+  "resolution": 512,
+  "resolution_type": "pixel_area",
+  "seed": 42,
+  "skip_file_discovery": false,
+  "tracker_project_name": "lora-training",
+  "tracker_run_name": "moss-domokun-json-sft-512",
+  "train_batch_size": 4,
+  "use_ema": false,
+  "vae_batch_size": 1,
+  "validation_guidance": 3.0,
+  "validation_guidance_rescale": 0.0,
+  "validation_lycoris_strength": 1.0,
+  "validation_num_inference_steps": 50,
+  "validation_prompt_library": false,
+  "validation_resolution": "512x512",
+  "validation_seed": 42,
+  "validation_guidance_real": 3.0,
+  "user_prompt_library": "prompts.json",
+  "dynamo_backend": "inductor",
+  "dynamo_mode": "default",
+  "dynamo_use_regional_compilation": true,
+  "validation_disable_unconditional": true,
+  "validation_negative_prompt": "",
+  "gradient_checkpointing_interval": 2,
+  "gradient_accumulation_steps": 1,
+  "grad_clip_method": "norm",
+  "lora_dropout": 0.0,
+  "vae_enable_tiling": false,
+  "num_validation_images": 1,
+  "dynamo_dynamic": true,
+  "crepa_enabled": true,
+  "crepa_block_index": 8,
+  "crepa_lambda": 0.5,
+  "crepa_encoder_image_size": 518,
+  "crepa_adjacent_distance": 0,
+  "crepa_spatial_align": true,
+  "crepa_encoder": "dinov2_vitg14",
+  "data_backend_sampling": "uniform",
+  "aspect_bucket_alignment": 16,
+  "dataloader_prefetch": false,
+  "flow_schedule_auto_shift": false,
+  "flow_schedule_shift": 1.0,
+  "validation_step_interval": 500,
+  "resume_from_checkpoint": "",
+  "vae_dtype": "fp32",
+  "tokenizer_max_length": 4096,
+  "train_text_encoder": false,
+  "moss_image_validation_shift": 5.0,
+  "moss_image_uncond_skip_layers": [
+    22
+  ],
+  "gradient_checkpointing_backend": "torch",
+  "gradient_checkpointing_offload_attention": false,
+  "pretrained_model_name_or_path": "OpenMOSS-Team/MOSS-Image-SFT",
+  "model_card_base_model": "OpenMOSS-Team/MOSS-Image-SFT",
+  "validation_on_startup": true,
+  "disable_assistant_lora": true,
+  "attention_mechanism": "kohaku-fa-auto",
+  "moss_image_auto_json": true,
+  "moss_image_prompt_upsample": false,
+  "revision": "fc5de0d56fc910cbca6f6625a70f5f582b26deb1"
+}