30000 steps (approx 4 epochs) with terminal SNR on 22k Midjourney 5.1 images plus 7200 real photographs as balance data with complete BLIP captions on all data. BS=4, LR=4e-7 to 1e-8

config.yaml

@@ -65,4 +65,4 @@ model:
       target: ldm.modules.encoders.modules.FrozenOpenCLIPEmbedder
       params:
         freeze: True
-        layer: "penultimate"
+        layer: "penultimate"

model_index.json

@@ -12,7 +12,7 @@
   ],
   "scheduler": [
     "diffusers",
-    "DDIMScheduler"
+    "DDPMScheduler"
   ],
   "text_encoder": [
     "transformers",

scheduler/scheduler_config.json

@@ -1,5 +1,5 @@
 {
-  "_class_name": "DDIMScheduler",
+  "_class_name": "DDPMScheduler",
   "_diffusers_version": "0.17.0.dev0",
   "beta_end": 0.012,
   "beta_schedule": "scaled_linear",
@@ -14,5 +14,6 @@
   "skip_prk_steps": true,
   "steps_offset": 1,
   "thresholding": false,
-  "trained_betas": null
+  "trained_betas": null,
+  "variance_type": "fixed_small"
 }

text_encoder/config.json

@@ -1,5 +1,5 @@
 {
-  "_name_or_path": "/notebooks/images/datasets/models//checkpoint-550/text_encoder",
+  "_name_or_path": "/notebooks/datasets/models/checkpoint-30000/text_encoder",
   "architectures": [
     "CLIPTextModel"
   ],
@@ -20,6 +20,6 @@
   "pad_token_id": 1,
   "projection_dim": 512,
   "torch_dtype": "float32",
-  "transformers_version": "4.29.2",
+  "transformers_version": "4.28.1",
   "vocab_size": 49408
 }

text_encoder/model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d7f8471deacb5f976911bfcf000ec5ea27e912b491dfd5041d2bbeefe88ebf8a
+oid sha256:0029ad268db15f83a8454ed1d5a99c6d712915f7f1dd04b948264a09efb35f43
 size 1361597016

unet/config.json

@@ -1,7 +1,7 @@
 {
   "_class_name": "UNet2DConditionModel",
   "_diffusers_version": "0.17.0.dev0",
-  "_name_or_path": "/notebooks/images/datasets/models//checkpoint-550/unet",
+  "_name_or_path": "/notebooks/datasets/models/checkpoint-30000/unet",
   "act_fn": "silu",
   "addition_embed_type": null,
   "addition_embed_type_num_heads": 64,

unet/diffusion_pytorch_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:644a1dd448e2ac8faba304c89ae1c0f65e9d73ef84144b3d5d3ef22915b81f76
+oid sha256:19b83cdbc9c86dd0cbee9d8bc9dcb76f81615c3fc1a399a7f1c7eebda8c026ed
 size 3463726504

vae/config.json

@@ -1,7 +1,7 @@
 {
   "_class_name": "AutoencoderKL",
   "_diffusers_version": "0.17.0.dev0",
-  "_name_or_path": "/notebooks/datasets/models/pipeline/vae",
+  "_name_or_path": "/root/.cache/huggingface/hub/models--stabilityai--stable-diffusion-2-1/snapshots/845609e6cf0a060d8cd837297e5c169df5bff72c/vae",
   "act_fn": "silu",
   "block_out_channels": [
     128,