First version of the model

README.md

@@ -0,0 +1,4 @@
+This model is converted with https://github.com/huggingface/diffusers/blob/main/scripts/convert_original_stable_diffusion_to_diffusers.py.
+
+However, the tokenizer in the diffuser model is wrong, for proper usage, see description at https://medium.com/@enryu9000/TODO, and instructions/examples at TODO.
+Also, the original checkpoint in the Latent Diffusion format is available.

bert/config.json

@@ -0,0 +1,26 @@
+{
+  "activation_dropout": 0.0,
+  "activation_function": "gelu",
+  "architectures": [
+    "LDMBertModel"
+  ],
+  "attention_dropout": 0.0,
+  "classifier_dropout": 0.0,
+  "d_model": 768,
+  "dropout": 0.1,
+  "encoder_attention_heads": 8,
+  "encoder_ffn_dim": 3072,
+  "encoder_layerdrop": 0.0,
+  "encoder_layers": 12,
+  "head_dim": 64,
+  "init_std": 0.02,
+  "max_position_embeddings": 50,
+  "model_type": "ldmbert",
+  "num_hidden_layers": 12,
+  "pad_token_id": 0,
+  "scale_embedding": false,
+  "torch_dtype": "float32",
+  "transformers_version": "4.21.1",
+  "use_cache": true,
+  "vocab_size": 3000
+}

bert/pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:01259faf08d6dd977d933dca39c46d5a2887c1c0e6a5f812e412d9e6380ed0ed
+size 321011645

config.json

@@ -0,0 +1,37 @@
+{
+  "_class_name": "UNet2DConditionModel",
+  "_diffusers_version": "0.2.2",
+  "act_fn": "silu",
+  "attention_head_dim": 8,
+  "block_out_channels": [
+    224,
+    448,
+    672,
+    896
+  ],
+  "center_input_sample": false,
+  "cross_attention_dim": 768,
+  "down_block_types": [
+    "CrossAttnDownBlock2D",
+    "CrossAttnDownBlock2D",
+    "CrossAttnDownBlock2D",
+    "DownBlock2D"
+  ],
+  "downsample_padding": 1,
+  "flip_sin_to_cos": true,
+  "freq_shift": 0,
+  "in_channels": 3,
+  "layers_per_block": 2,
+  "mid_block_scale_factor": 1,
+  "norm_eps": 1e-05,
+  "norm_num_groups": 32,
+  "out_channels": 3,
+  "sample_size": 64,
+  "up_block_types": [
+    "UpBlock2D",
+    "CrossAttnUpBlock2D",
+    "CrossAttnUpBlock2D",
+    "CrossAttnUpBlock2D"
+  ]
+}
+

model_index.json

@@ -0,0 +1,24 @@
+{
+  "_class_name": "LDMTextToImagePipeline",
+  "_diffusers_version": "0.2.4",
+  "bert": [
+    "latent_diffusion",
+    "LDMBertModel"
+  ],
+  "scheduler": [
+    "diffusers",
+    "PNDMScheduler"
+  ],
+  "tokenizer": [
+    "transformers",
+    "BertTokenizerFast"
+  ],
+  "unet": [
+    "diffusers",
+    "UNet2DConditionModel"
+  ],
+  "vqvae": [
+    "diffusers",
+    "VQModel"
+  ]
+}

original_ckpt.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b330b009ff625ce41093a374b99dea4ab8bcc48769590843833c8f2c460a7b5e
+size 2231207298

scheduler/scheduler_config.json

@@ -0,0 +1,9 @@
+{
+  "_class_name": "PNDMScheduler",
+  "_diffusers_version": "0.2.4",
+  "beta_end": 0.0195,
+  "beta_schedule": "scaled_linear",
+  "beta_start": 0.0015,
+  "num_train_timesteps": 1000,
+  "skip_prk_steps": true
+}

tokenizer/special_tokens_map.json

@@ -0,0 +1,7 @@
+{
+  "cls_token": "[CLS]",
+  "mask_token": "[MASK]",
+  "pad_token": "[PAD]",
+  "sep_token": "[SEP]",
+  "unk_token": "[UNK]"
+}

tokenizer/tokenizer_config.json

@@ -0,0 +1,14 @@
+{
+  "cls_token": "[CLS]",
+  "do_lower_case": true,
+  "mask_token": "[MASK]",
+  "model_max_length": 512,
+  "name_or_path": "bert-base-uncased",
+  "pad_token": "[PAD]",
+  "sep_token": "[SEP]",
+  "special_tokens_map_file": null,
+  "strip_accents": null,
+  "tokenize_chinese_chars": true,
+  "tokenizer_class": "BertTokenizer",
+  "unk_token": "[UNK]"
+}

unet/config.json

@@ -0,0 +1,36 @@
+{
+  "_class_name": "UNet2DConditionModel",
+  "_diffusers_version": "0.2.4",
+  "act_fn": "silu",
+  "attention_head_dim": 8,
+  "block_out_channels": [
+    224,
+    448,
+    672,
+    896
+  ],
+  "center_input_sample": false,
+  "cross_attention_dim": 768,
+  "down_block_types": [
+    "DownBlock2D",
+    "CrossAttnDownBlock2D",
+    "CrossAttnDownBlock2D",
+    "CrossAttnDownBlock2D"
+  ],
+  "downsample_padding": 1,
+  "flip_sin_to_cos": true,
+  "freq_shift": 0,
+  "in_channels": 3,
+  "layers_per_block": 2,
+  "mid_block_scale_factor": 1,
+  "norm_eps": 1e-05,
+  "norm_num_groups": 32,
+  "out_channels": 3,
+  "sample_size": 64,
+  "up_block_types": [
+    "CrossAttnUpBlock2D",
+    "CrossAttnUpBlock2D",
+    "CrossAttnUpBlock2D",
+    "UpBlock2D"
+  ]
+}

unet/diffusion_pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:54511f99cf34e34ff51a72b04b9d964a53f97b797a91fc718442710a89736244
+size 1681604705

vqvae/config.json

@@ -0,0 +1,29 @@
+{
+  "_class_name": "VQModel",
+  "_diffusers_version": "0.2.4",
+  "act_fn": "silu",
+  "block_out_channels": [
+    128,
+    128,
+    256,
+    512
+  ],
+  "down_block_types": [
+    "DownEncoderBlock2D",
+    "DownEncoderBlock2D",
+    "DownEncoderBlock2D",
+    "DownEncoderBlock2D"
+  ],
+  "in_channels": 3,
+  "latent_channels": 3,
+  "layers_per_block": 2,
+  "num_vq_embeddings": 8192,
+  "out_channels": 3,
+  "sample_size": 512,
+  "up_block_types": [
+    "UpDecoderBlock2D",
+    "UpDecoderBlock2D",
+    "UpDecoderBlock2D",
+    "UpDecoderBlock2D"
+  ]
+}

vqvae/diffusion_pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:864e0bb2bab2be04cba7dd60ede7ec37e72ff548056969f5ed1f9c4c02bde5a0
+size 228477455