update: 13K steps

Signed-off-by: Ming Yang <vivym@live.com>

README.md

@@ -36,7 +36,7 @@ extra_gated_heading: Please read the LICENSE to access this model
 
 # v-prediction variant of BK-SDM-Tiny
 
-Trained on text-image pairs from Midjourney v5.2, 2000 steps with a batch size of 2048.
+Trained on text-image pairs from Midjourney v5.2, 13000 steps with a batch size of 2048.
 
 # BK-SDM Model Card
 Block-removed Knowledge-distilled Stable Diffusion Model (BK-SDM) is an architecturally compressed SDM for efficient general-purpose text-to-image synthesis. This model is bulit with (i) removing several residual and attention blocks from the U-Net of [Stable Diffusion v1.4]( https://huggingface.co/CompVis/stable-diffusion-v1-4) and (ii) distillation pretraining on only 0.22M LAION pairs (fewer than 0.1% of the full training set). Despite being trained with very limited resources, our compact model can imitate the original SDM by benefiting from transferred knowledge.

unet/config.json

@@ -1,7 +1,7 @@
 {
   "_class_name": "UNet2DConditionModel",
   "_diffusers_version": "0.20.2",
-  "_name_or_path": "vivym/bk-sdm-tiny-vpred",
+  "_name_or_path": "nota-ai/bk-sdm-tiny",
   "act_fn": "silu",
   "addition_embed_type": null,
   "addition_embed_type_num_heads": 64,
@@ -44,7 +44,7 @@
   "num_attention_heads": null,
   "num_class_embeds": null,
   "only_cross_attention": false,
-  "optimization_step": 2000,
+  "optimization_step": 8000,
   "out_channels": 4,
   "power": 0.6666666666666666,
   "projection_class_embeddings_input_dim": null,

unet/diffusion_pytorch_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:55ff76d0c19263022293ef5248d89f5d8b3a70a7bc25b92fef3e76a0fd9bd93b
+oid sha256:828fbf7e98271e151c1b23b1c5a780871119ddc2a79443382dd49cd12089b72e
 size 1293583616