Upload folder using huggingface_hub

.gitattributes

@@ -101,3 +101,13 @@ dlay_qwen_lora_v1/samples/1765212508678__000006800_3.jpg filter=lfs diff=lfs mer
 dlay_qwen_lora_v1/samples/1765214029041__000007600_3.jpg filter=lfs diff=lfs merge=lfs -text
 dlay_qwen_lora_v1/samples/1765214785978__000008000_3.jpg filter=lfs diff=lfs merge=lfs -text
 dlay_qwen_lora_v1/samples/1765215538813__000008400_3.jpg filter=lfs diff=lfs merge=lfs -text
+samples/.ipynb_checkpoints/1765229053127__000002800_3-checkpoint.jpg filter=lfs diff=lfs merge=lfs -text
+samples/.ipynb_checkpoints/1765232410784__000003600_3-checkpoint.jpg filter=lfs diff=lfs merge=lfs -text
+samples/.ipynb_checkpoints/1765234085028__000004000_3-checkpoint.jpg filter=lfs diff=lfs merge=lfs -text
+samples/.ipynb_checkpoints/1765235764342__000004400_3-checkpoint.jpg filter=lfs diff=lfs merge=lfs -text
+samples/.ipynb_checkpoints/1765239111417__000005200_3-checkpoint.jpg filter=lfs diff=lfs merge=lfs -text
+samples/1765229053127__000002800_3.jpg filter=lfs diff=lfs merge=lfs -text
+samples/1765232410784__000003600_3.jpg filter=lfs diff=lfs merge=lfs -text
+samples/1765234085028__000004000_3.jpg filter=lfs diff=lfs merge=lfs -text
+samples/1765235764342__000004400_3.jpg filter=lfs diff=lfs merge=lfs -text
+samples/1765239111417__000005200_3.jpg filter=lfs diff=lfs merge=lfs -text

config.yaml

@@ -1,119 +1,102 @@
 job: extension
 config:
-  name: Qwen-Image-LoKr-DLAY-v2
+  name: dlay_qwen_LoHA_v1
   process:
-  - type: diffusion_trainer
-    training_folder: /workspace/ai-toolkit/output
-    sqlite_db_path: /workspace/ai-toolkit/aitk_db.db
-    device: cuda
-    trigger_word: null
-    performance_log_every: 10
+  - type: sd_trainer
+    training_folder: /root/output
+    performance_log_every: 100
+    device: cuda:0
     network:
-      type: lokr
-      linear: 128
-      linear_alpha: 64
-      conv: 32
-      conv_alpha: 16
-      lokr_full_rank: true
-      lokr_factor: 4
+      type: loha
+      linear: 64
+      linear_alpha: 32
+      dropout: 0.0
+      use_tucker: false
+      use_scalar: true
+      weight_decompose: true
+      wd_on_out: true
+      rs_lora: false
+      rank_dropout_scale: false
       network_kwargs:
-        algo: lokr
-        factor: 4
-        dora_wd: true
-        ignore_if_contains: []
-    save:
-      dtype: bf16
-      save_every: 500
-      max_step_saves_to_keep: 6
-      save_format: safetensors
-      push_to_hub: true
+        rank_dropout: 0.02
+    model:
+      name_or_path: Qwen/Qwen-Image
+      arch: qwen_image
+      quantize: false
+      quantize_te: false
+      low_vram: false
+      layer_offloading: true
+      layer_offloading_transformer_percent: 0.99
     datasets:
-    - folder_path: /workspace/DLAY-1024/subject
-      mask_path: /workspace/DLAY-1024/mask
-      mask_min_value: 0.1
-      default_caption: ''
+    - folder_path: /root/DATASETS/DLAY/flux2
       caption_ext: txt
-      caption_dropout_rate: 0.05
-      cache_latents_to_disk: false
-      is_reg: false
-      network_weight: 1
+      default_caption: DLAY, a man with short dark brown hair and a slim build
+      mask_path: /root/DATASETS/DLAY/mask
+      mask_min_value: 0.0
+      invert_mask: false
+      loss_multiplier: 1.2
       resolution:
+      - 640
       - 1024
-      controls: []
-      shrink_video_to_frames: true
-      num_frames: 1
-      do_i2v: true
-      flip_x: false
-      flip_y: false
-      control_path_1: null
-      control_path_2: null
-      control_path_3: null
+      cache_latents: true
+      cache_latents_to_disk: true
+      cache_text_embeddings: true
+      cache_text_embeddings_to_disk: true
+      unload_text_encoder: true
+      caption_dropout_rate: 0.05
+      shuffle_tokens: false
     train:
-      batch_size: 1
-      bypass_guidance_embedding: false
-      steps: 13000
+      batch_size: 2
       gradient_accumulation: 1
-      train_unet: true
-      train_text_encoder: false
-      gradient_checkpointing: false
-      attention_backend: _flash_3
-      noise_scheduler: flowmatch
-      optimizer: adamw
-      timestep_type: weighted
-      content_or_style: balanced
+      steps: 6000
+      attention_backend: flash
+      optimizer: prodigy
+      lr: 1.0
       optimizer_params:
+        d_coef: 1.5
+        decouple: true
+        use_bias_correction: true
         weight_decay: 0.01
-      unload_text_encoder: false
-      cache_text_embeddings: false
-      lr: 0.0001
+        safeguard_warmup: true
+        betas:
+        - 0.9
+        - 0.999
+      lr_scheduler: cosine_with_warmup
+      lr_scheduler_params:
+        num_warmup_steps: 200
+        num_training_steps: 6000
+      dtype: bf16
       ema_config:
-        use_ema: true
-        ema_decay: 0.99
+        use_ema: false
+        ema_decay: 0.9995
+      gradient_checkpointing: false
+      noise_scheduler: flowmatch
+      timestep_type: sigmoid
+      max_grad_norm: 1.0
+      train_unet: true
+      train_text_encoder: false
       skip_first_sample: true
-      force_first_sample: false
-      disable_sampling: false
-      dtype: bf16
-      diff_output_preservation: false
-      diff_output_preservation_multiplier: 1
-      diff_output_preservation_class: person
-      switch_boundary_every: 10
-      inverted_mask_prior: true
-      inverted_mask_prior_multiplier: 0.5
-      loss_type: mse
-    model:
-      name_or_path: Qwen/Qwen-Image
-      quantize: false
-      qtype: qfloat8
-      quantize_te: false
-      qtype_te: qfloat8
-      arch: qwen_image
-      low_vram: false
-      model_kwargs: {}
-      layer_offloading: false
-      layer_offloading_text_encoder_percent: 1
-      layer_offloading_transformer_percent: 1
+    save:
+      dtype: bfloat16
+      save_every: 400
+      save_format: safetensors
+      max_step_saves_to_keep: 15
     sample:
       sampler: flowmatch
-      sample_every: 1000
+      sample_every: 400
       width: 1024
       height: 1024
-      samples:
-      - prompt: A realistic photo of dlay man on a rocky overlook, gazing out at a
-          sprawling forest and a winding blue river under a clear sky. Bright, direct
-          sunlight of a summer afternoon.
-      - prompt: Cinematic interior shot of dlay man standing in the center of the
-          Pantheon in Rome, looking up towards the oculus. The vast, coffered dome
-          is visible above, with light streaming in. Ornate marble floors and classical
-          architecture surround him.
-      - prompt: A photorealistic photo of dlay man standing behind a stone wall, looking
-          out at the Giza pyramids in the sandy desert under a hazy sky.
+      prompts:
+      - DLAY, a man in a professional studio lighting, white background
+      - DLAY, a man wearing a suit, confident expression, corporate headshot
+      - DLAY, a man in a cinematic photo, close up, detailed skin texture, dramatic
+        lighting
+      - DLAY, a man wearing casual attire, outdoor setting, natural sunlight
       neg: ''
       seed: 42
       walk_seed: true
-      guidance_scale: 4
+      guidance_scale: 3.5
       sample_steps: 25
-      num_frames: 1
-      fps: 1
 meta:
-  name: '[DLAY-Qwen-Image--LoKr]'
+  name: '[DLAY-man-Qwen-Image-LoHA]'
   version: '1.0'

dlay_qwen_LoHA_v1.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:78fcc8062ad5fd0db1481a9a05fcc3354bbeca5c9300c666cbfde54a597a40f2
+size 2371152168

dlay_qwen_LoHA_v1_000000400.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:17102dc1754a8afd4830b01e2cf18dbb1509e29c2c66195837a1806c9dc278ac
+size 2371152160

dlay_qwen_LoHA_v1_000000800.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6ad0b76ed31e8f5fdef58ece7ab394ccdbc9d78a2336288fcf2063c070d48580
+size 2371152160

dlay_qwen_LoHA_v1_000001200.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7cb659fbd8edbf05ee515a5df4d341d239d0eaba438a75adfda416dcc5ba61b2
+size 2371152168

dlay_qwen_LoHA_v1_000001600.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a0e07c0660c7e59846b2defb9ab061ad7e7d54a8cecc4f26fd6fd764f3c29d64
+size 2371152168

dlay_qwen_LoHA_v1_000002000.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3f40af4d10d5eead0143cb7165e4b9acecacfe07ef09113fa08f6f960a5254df
+size 2371152168

dlay_qwen_LoHA_v1_000002400.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0e5398aaeb550fbba7509b5ca9fbbe3b187fe4c9b915dacf0f3255326f02b9c7
+size 2371152168

dlay_qwen_LoHA_v1_000002800.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7b48b62cba199d238bf1e32d22494fa5205c4335480c0147407f3019b14b0cdd
+size 2371152168

dlay_qwen_LoHA_v1_000003200.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1653543068554963818136ae4ca0d6dbb8a9eea19c1351a940fa73c7a62fc310
+size 2371152168

dlay_qwen_LoHA_v1_000003600.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1b6f7685003b5069232a2cb08ad269a84f8e79712eb179fcb7bc4ca525a5c80a
+size 2371152168

dlay_qwen_LoHA_v1_000004000.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:33a68017b2e24df507cd72bcc9ba1ff98c7efd3d7f42a01a05b2c83f74626b3e
+size 2371152168

dlay_qwen_LoHA_v1_000004400.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7f640b07ac25c4f157e405de547b3f8ee0d8079226d931812258fdbf46bf2e2b
+size 2371152168

dlay_qwen_LoHA_v1_000004800.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3352e1331ae392c865e44cb65a72a1ef671e0edadfd8bf95bcc40613c30e5d07
+size 2371152168

dlay_qwen_LoHA_v1_000005200.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5314a9fcc2e35aa9fe9b65e64f4edff04e4f0f27ef00556bbe73b352fd458138
+size 2371152168

dlay_qwen_LoHA_v1_000005600.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5532d29aec89ff9fb2db45a57d6616210b44bf4c18253deda4ea7507eeee0d93
+size 2371152168

optimizer.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b5e403d95d9c98ac08b1d0b29a348a12d8e362edfcd56308f8581822885edfc9
+size 18893089275