Upload folder using huggingface_hub

.gitattributes

@@ -33,3 +33,9 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+Reference[[:space:]]Papers/classifier[[:space:]]free[[:space:]]diffusion[[:space:]]guidance[[:space:]]paper.pdf filter=lfs diff=lfs merge=lfs -text
+Reference[[:space:]]Papers/Denoising[[:space:]]Diffusion[[:space:]]Probabilistic[[:space:]]Models[[:space:]]paper.pdf filter=lfs diff=lfs merge=lfs -text
+Reference[[:space:]]Papers/High-Resolution[[:space:]]Image[[:space:]]Synthesis[[:space:]]with[[:space:]]Latent[[:space:]]Diffusion[[:space:]]Models[[:space:]]paper.pdf filter=lfs diff=lfs merge=lfs -text
+Reference[[:space:]]Papers/Learning[[:space:]]Transferable[[:space:]]Visual[[:space:]]Models[[:space:]]From[[:space:]]Natural[[:space:]]Language[[:space:]]Supervision[[:space:]]paper.pdf filter=lfs diff=lfs merge=lfs -text
+Reference[[:space:]]Papers/Photorealistic[[:space:]]Text-to-Image[[:space:]]Diffusion[[:space:]]Models[[:space:]]paper.pdf filter=lfs diff=lfs merge=lfs -text
+Stable_Diffusion_Diagrams_V2.pdf filter=lfs diff=lfs merge=lfs -text

Hollie_Mengert.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2c4c9a75f6045b861b3f9252f51442dc4880c70fb792b78446940abc232bdbb7
+size 2132903713

README.md

@@ -0,0 +1,207 @@
+---
+license: creativeml-openrail-m
+tags:
+- stable-diffusion
+- stable-diffusion-diffusers
+- text-to-image
+inference: true
+extra_gated_prompt: |-
+  This model is open access and available to all, with a CreativeML OpenRAIL-M license further specifying rights and usage.
+  The CreativeML OpenRAIL License specifies: 
+
+  1. You can't use the model to deliberately produce nor share illegal or harmful outputs or content 
+  2. CompVis claims no rights on the outputs you generate, you are free to use them and are accountable for their use which must not go against the provisions set in the license
+  3. You may re-distribute the weights and use the model commercially and/or as a service. If you do, please be aware you have to include the same use restrictions as the ones in the license and share a copy of the CreativeML OpenRAIL-M to all your users (please read the license entirely and carefully)
+  Please read the full license carefully here: https://huggingface.co/spaces/CompVis/stable-diffusion-license
+      
+extra_gated_heading: Please read the LICENSE to access this model
+---
+
+# Stable Diffusion v1-5 Model Card
+
+Stable Diffusion is a latent text-to-image diffusion model capable of generating photo-realistic images given any text input.
+For more information about how Stable Diffusion functions, please have a look at [🤗's Stable Diffusion blog](https://huggingface.co/blog/stable_diffusion).
+
+The **Stable-Diffusion-v1-5** checkpoint was initialized with the weights of the [Stable-Diffusion-v1-2](https:/steps/huggingface.co/CompVis/stable-diffusion-v1-2) 
+checkpoint and subsequently fine-tuned on 595k steps at resolution 512x512 on "laion-aesthetics v2 5+" and 10% dropping of the text-conditioning to improve [classifier-free guidance sampling](https://arxiv.org/abs/2207.12598).
+
+You can use this both with the [🧨Diffusers library](https://github.com/huggingface/diffusers) and the [RunwayML GitHub repository](https://github.com/runwayml/stable-diffusion).
+
+### Diffusers
+```py
+from diffusers import StableDiffusionPipeline
+import torch
+
+model_id = "runwayml/stable-diffusion-v1-5"
+pipe = StableDiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float16)
+pipe = pipe.to("cuda")
+
+prompt = "a photo of an astronaut riding a horse on mars"
+image = pipe(prompt).images[0]  
+    
+image.save("astronaut_rides_horse.png")
+```
+For more detailed instructions, use-cases and examples in JAX follow the instructions [here](https://github.com/huggingface/diffusers#text-to-image-generation-with-stable-diffusion)
+
+### Original GitHub Repository
+
+1. Download the weights 
+   - [v1-5-pruned-emaonly.ckpt](https://huggingface.co/runwayml/stable-diffusion-v1-5/resolve/main/v1-5-pruned-emaonly.ckpt) - 4.27GB, ema-only weight. uses less VRAM - suitable for inference
+   - [v1-5-pruned.ckpt](https://huggingface.co/runwayml/stable-diffusion-v1-5/resolve/main/v1-5-pruned.ckpt) - 7.7GB, ema+non-ema weights. uses more VRAM - suitable for fine-tuning
+
+2. Follow instructions [here](https://github.com/runwayml/stable-diffusion).
+
+## Model Details
+- **Developed by:** Robin Rombach, Patrick Esser
+- **Model type:** Diffusion-based text-to-image generation model
+- **Language(s):** English
+- **License:** [The CreativeML OpenRAIL M license](https://huggingface.co/spaces/CompVis/stable-diffusion-license) is an [Open RAIL M license](https://www.licenses.ai/blog/2022/8/18/naming-convention-of-responsible-ai-licenses), adapted from the work that [BigScience](https://bigscience.huggingface.co/) and [the RAIL Initiative](https://www.licenses.ai/) are jointly carrying in the area of responsible AI licensing. See also [the article about the BLOOM Open RAIL license](https://bigscience.huggingface.co/blog/the-bigscience-rail-license) on which our license is based.
+- **Model Description:** This is a model that can be used to generate and modify images based on text prompts. It is a [Latent Diffusion Model](https://arxiv.org/abs/2112.10752) that uses a fixed, pretrained text encoder ([CLIP ViT-L/14](https://arxiv.org/abs/2103.00020)) as suggested in the [Imagen paper](https://arxiv.org/abs/2205.11487).
+- **Resources for more information:** [GitHub Repository](https://github.com/CompVis/stable-diffusion), [Paper](https://arxiv.org/abs/2112.10752).
+- **Cite as:**
+
+      @InProceedings{Rombach_2022_CVPR,
+          author    = {Rombach, Robin and Blattmann, Andreas and Lorenz, Dominik and Esser, Patrick and Ommer, Bj\"orn},
+          title     = {High-Resolution Image Synthesis With Latent Diffusion Models},
+          booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
+          month     = {June},
+          year      = {2022},
+          pages     = {10684-10695}
+      }
+
+# Uses
+
+## Direct Use 
+The model is intended for research purposes only. Possible research areas and
+tasks include
+
+- Safe deployment of models which have the potential to generate harmful content.
+- Probing and understanding the limitations and biases of generative models.
+- Generation of artworks and use in design and other artistic processes.
+- Applications in educational or creative tools.
+- Research on generative models.
+
+Excluded uses are described below.
+
+ ### Misuse, Malicious Use, and Out-of-Scope Use
+_Note: This section is taken from the [DALLE-MINI model card](https://huggingface.co/dalle-mini/dalle-mini), but applies in the same way to Stable Diffusion v1_.
+
+
+The model should not be used to intentionally create or disseminate images that create hostile or alienating environments for people. This includes generating images that people would foreseeably find disturbing, distressing, or offensive; or content that propagates historical or current stereotypes.
+
+#### Out-of-Scope Use
+The model was not trained to be factual or true representations of people or events, and therefore using the model to generate such content is out-of-scope for the abilities of this model.
+
+#### Misuse and Malicious Use
+Using the model to generate content that is cruel to individuals is a misuse of this model. This includes, but is not limited to:
+
+- Generating demeaning, dehumanizing, or otherwise harmful representations of people or their environments, cultures, religions, etc.
+- Intentionally promoting or propagating discriminatory content or harmful stereotypes.
+- Impersonating individuals without their consent.
+- Sexual content without consent of the people who might see it.
+- Mis- and disinformation
+- Representations of egregious violence and gore
+- Sharing of copyrighted or licensed material in violation of its terms of use.
+- Sharing content that is an alteration of copyrighted or licensed material in violation of its terms of use.
+
+## Limitations and Bias
+
+### Limitations
+
+- The model does not achieve perfect photorealism
+- The model cannot render legible text
+- The model does not perform well on more difficult tasks which involve compositionality, such as rendering an image corresponding to “A red cube on top of a blue sphere”
+- Faces and people in general may not be generated properly.
+- The model was trained mainly with English captions and will not work as well in other languages.
+- The autoencoding part of the model is lossy
+- The model was trained on a large-scale dataset
+  [LAION-5B](https://laion.ai/blog/laion-5b/) which contains adult material
+  and is not fit for product use without additional safety mechanisms and
+  considerations.
+- No additional measures were used to deduplicate the dataset. As a result, we observe some degree of memorization for images that are duplicated in the training data.
+  The training data can be searched at [https://rom1504.github.io/clip-retrieval/](https://rom1504.github.io/clip-retrieval/) to possibly assist in the detection of memorized images.
+
+### Bias
+
+While the capabilities of image generation models are impressive, they can also reinforce or exacerbate social biases. 
+Stable Diffusion v1 was trained on subsets of [LAION-2B(en)](https://laion.ai/blog/laion-5b/), 
+which consists of images that are primarily limited to English descriptions. 
+Texts and images from communities and cultures that use other languages are likely to be insufficiently accounted for. 
+This affects the overall output of the model, as white and western cultures are often set as the default. Further, the 
+ability of the model to generate content with non-English prompts is significantly worse than with English-language prompts.
+
+### Safety Module
+
+The intended use of this model is with the [Safety Checker](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion/safety_checker.py) in Diffusers. 
+This checker works by checking model outputs against known hard-coded NSFW concepts.
+The concepts are intentionally hidden to reduce the likelihood of reverse-engineering this filter.
+Specifically, the checker compares the class probability of harmful concepts in the embedding space of the `CLIPTextModel` *after generation* of the images. 
+The concepts are passed into the model with the generated image and compared to a hand-engineered weight for each NSFW concept.
+
+
+## Training
+
+**Training Data**
+The model developers used the following dataset for training the model:
+
+- LAION-2B (en) and subsets thereof (see next section)
+
+**Training Procedure**
+Stable Diffusion v1-5 is a latent diffusion model which combines an autoencoder with a diffusion model that is trained in the latent space of the autoencoder. During training, 
+
+- Images are encoded through an encoder, which turns images into latent representations. The autoencoder uses a relative downsampling factor of 8 and maps images of shape H x W x 3 to latents of shape H/f x W/f x 4
+- Text prompts are encoded through a ViT-L/14 text-encoder.
+- The non-pooled output of the text encoder is fed into the UNet backbone of the latent diffusion model via cross-attention.
+- The loss is a reconstruction objective between the noise that was added to the latent and the prediction made by the UNet.
+
+Currently six Stable Diffusion checkpoints are provided, which were trained as follows.
+- [`stable-diffusion-v1-1`](https://huggingface.co/CompVis/stable-diffusion-v1-1): 237,000 steps at resolution `256x256` on [laion2B-en](https://huggingface.co/datasets/laion/laion2B-en).
+  194,000 steps at resolution `512x512` on [laion-high-resolution](https://huggingface.co/datasets/laion/laion-high-resolution) (170M examples from LAION-5B with resolution `>= 1024x1024`).
+- [`stable-diffusion-v1-2`](https://huggingface.co/CompVis/stable-diffusion-v1-2): Resumed from `stable-diffusion-v1-1`.
+  515,000 steps at resolution `512x512` on "laion-improved-aesthetics" (a subset of laion2B-en,
+filtered to images with an original size `>= 512x512`, estimated aesthetics score `> 5.0`, and an estimated watermark probability `< 0.5`. The watermark estimate is from the LAION-5B metadata, the aesthetics score is estimated using an [improved aesthetics estimator](https://github.com/christophschuhmann/improved-aesthetic-predictor)).
+- [`stable-diffusion-v1-3`](https://huggingface.co/CompVis/stable-diffusion-v1-3): Resumed from `stable-diffusion-v1-2` - 195,000 steps at resolution `512x512` on "laion-improved-aesthetics" and 10 % dropping of the text-conditioning to improve [classifier-free guidance sampling](https://arxiv.org/abs/2207.12598).
+- [`stable-diffusion-v1-4`](https://huggingface.co/CompVis/stable-diffusion-v1-4) Resumed from `stable-diffusion-v1-2` - 225,000 steps at resolution `512x512` on "laion-aesthetics v2 5+" and 10 % dropping of the text-conditioning to improve [classifier-free guidance sampling](https://arxiv.org/abs/2207.12598).
+- [`stable-diffusion-v1-5`](https://huggingface.co/runwayml/stable-diffusion-v1-5) Resumed from `stable-diffusion-v1-2` - 595,000 steps at resolution `512x512` on "laion-aesthetics v2 5+" and 10 % dropping of the text-conditioning to improve [classifier-free guidance sampling](https://arxiv.org/abs/2207.12598).
+- [`stable-diffusion-inpainting`](https://huggingface.co/runwayml/stable-diffusion-inpainting) Resumed from `stable-diffusion-v1-5` - then 440,000 steps of inpainting training at resolution 512x512 on “laion-aesthetics v2 5+” and 10% dropping of the text-conditioning. For inpainting, the UNet has 5 additional input channels (4 for the encoded masked-image and 1 for the mask itself) whose weights were zero-initialized after restoring the non-inpainting checkpoint. During training, we generate synthetic masks and in 25% mask everything.
+
+- **Hardware:** 32 x 8 x A100 GPUs
+- **Optimizer:** AdamW
+- **Gradient Accumulations**: 2
+- **Batch:** 32 x 8 x 2 x 4 = 2048
+- **Learning rate:** warmup to 0.0001 for 10,000 steps and then kept constant
+
+## Evaluation Results 
+Evaluations with different classifier-free guidance scales (1.5, 2.0, 3.0, 4.0,
+5.0, 6.0, 7.0, 8.0) and 50 PNDM/PLMS sampling
+steps show the relative improvements of the checkpoints:
+
+![pareto](https://huggingface.co/CompVis/stable-diffusion/resolve/main/v1-1-to-v1-5.png)
+
+Evaluated using 50 PLMS steps and 10000 random prompts from the COCO2017 validation set, evaluated at 512x512 resolution.  Not optimized for FID scores.
+## Environmental Impact
+
+**Stable Diffusion v1** **Estimated Emissions**
+Based on that information, we estimate the following CO2 emissions using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700). The hardware, runtime, cloud provider, and compute region were utilized to estimate the carbon impact.
+
+- **Hardware Type:** A100 PCIe 40GB
+- **Hours used:** 150000
+- **Cloud Provider:** AWS
+- **Compute Region:** US-east
+- **Carbon Emitted (Power consumption x Time x Carbon produced based on location of power grid):** 11250 kg CO2 eq.
+
+
+## Citation
+
+```bibtex
+    @InProceedings{Rombach_2022_CVPR,
+        author    = {Rombach, Robin and Blattmann, Andreas and Lorenz, Dominik and Esser, Patrick and Ommer, Bj\"orn},
+        title     = {High-Resolution Image Synthesis With Latent Diffusion Models},
+        booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
+        month     = {June},
+        year      = {2022},
+        pages     = {10684-10695}
+    }
+```
+
+*This model card was written by: Robin Rombach and Patrick Esser and is based on the [DALL-E Mini model card](https://huggingface.co/dalle-mini/dalle-mini).*

SD/attention.py

@@ -0,0 +1,122 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+import math
+
+class SelfAttention(nn.Module):
+    def __init__(self, n_heads, d_embed, in_proj_bias=True, out_proj_bias=True):
+        super().__init__()
+        # This combines the Wq, Wk and Wv matrices into one matrix
+        self.in_proj = nn.Linear(d_embed, 3 * d_embed, bias=in_proj_bias)
+        # This one represents the Wo matrix
+        self.out_proj = nn.Linear(d_embed, d_embed, bias=out_proj_bias)
+        self.n_heads = n_heads
+        self.d_head = d_embed // n_heads
+
+    def forward(self, x, causal_mask=False):
+        # x: # (Batch_Size, Seq_Len, Dim)
+
+        # (Batch_Size, Seq_Len, Dim)
+        input_shape = x.shape 
+        
+        # (Batch_Size, Seq_Len, Dim)
+        batch_size, sequence_length, d_embed = input_shape 
+
+        # (Batch_Size, Seq_Len, H, Dim / H)
+        interim_shape = (batch_size, sequence_length, self.n_heads, self.d_head) 
+
+        # (Batch_Size, Seq_Len, Dim) -> (Batch_Size, Seq_Len, Dim * 3) -> 3 tensor of shape (Batch_Size, Seq_Len, Dim)
+        q, k, v = self.in_proj(x).chunk(3, dim=-1)
+        
+        # (Batch_Size, Seq_Len, Dim) -> (Batch_Size, Seq_Len, H, Dim / H) -> (Batch_Size, H, Seq_Len, Dim / H)
+        q = q.view(interim_shape).transpose(1, 2)
+        k = k.view(interim_shape).transpose(1, 2)
+        v = v.view(interim_shape).transpose(1, 2)
+
+        # (Batch_Size, H, Seq_Len, Dim) @ (Batch_Size, H, Dim, Seq_Len) -> (Batch_Size, H, Seq_Len, Seq_Len)
+        weight = q @ k.transpose(-1, -2)
+        
+        if causal_mask:
+            # Mask where the upper triangle (above the principal diagonal) is 1
+            mask = torch.ones_like(weight, dtype=torch.bool).triu(1) 
+            # Fill the upper triangle with -inf
+            weight.masked_fill_(mask, -torch.inf) 
+        
+        # Divide by d_k (Dim / H). 
+        # (Batch_Size, H, Seq_Len, Seq_Len) -> (Batch_Size, H, Seq_Len, Seq_Len)
+        weight /= math.sqrt(self.d_head) 
+
+        # (Batch_Size, H, Seq_Len, Seq_Len) -> (Batch_Size, H, Seq_Len, Seq_Len)
+        weight = F.softmax(weight, dim=-1) 
+
+        # (Batch_Size, H, Seq_Len, Seq_Len) @ (Batch_Size, H, Seq_Len, Dim / H) -> (Batch_Size, H, Seq_Len, Dim / H)
+        output = weight @ v
+
+        # (Batch_Size, H, Seq_Len, Dim / H) -> (Batch_Size, Seq_Len, H, Dim / H)
+        output = output.transpose(1, 2) 
+
+        # (Batch_Size, Seq_Len, H, Dim / H) -> (Batch_Size, Seq_Len, Dim)
+        output = output.reshape(input_shape) 
+
+        # (Batch_Size, Seq_Len, Dim) -> (Batch_Size, Seq_Len, Dim)
+        output = self.out_proj(output) 
+        
+        # (Batch_Size, Seq_Len, Dim)
+        return output
+
+class CrossAttention(nn.Module):
+    def __init__(self, n_heads, d_embed, d_cross, in_proj_bias=True, out_proj_bias=True):
+        super().__init__()
+        self.q_proj   = nn.Linear(d_embed, d_embed, bias=in_proj_bias)
+        self.k_proj   = nn.Linear(d_cross, d_embed, bias=in_proj_bias)
+        self.v_proj   = nn.Linear(d_cross, d_embed, bias=in_proj_bias)
+        self.out_proj = nn.Linear(d_embed, d_embed, bias=out_proj_bias)
+        self.n_heads = n_heads
+        self.d_head = d_embed // n_heads
+    
+    def forward(self, x, y):
+        # x (latent): # (Batch_Size, Seq_Len_Q, Dim_Q)
+        # y (context): # (Batch_Size, Seq_Len_KV, Dim_KV) = (Batch_Size, 77, 768)
+
+        input_shape = x.shape
+        batch_size, sequence_length, d_embed = input_shape
+        # Divide each embedding of Q into multiple heads such that d_heads * n_heads = Dim_Q
+        interim_shape = (batch_size, -1, self.n_heads, self.d_head)
+        
+        # (Batch_Size, Seq_Len_Q, Dim_Q) -> (Batch_Size, Seq_Len_Q, Dim_Q)
+        q = self.q_proj(x)
+        # (Batch_Size, Seq_Len_KV, Dim_KV) -> (Batch_Size, Seq_Len_KV, Dim_Q)
+        k = self.k_proj(y)
+        # (Batch_Size, Seq_Len_KV, Dim_KV) -> (Batch_Size, Seq_Len_KV, Dim_Q)
+        v = self.v_proj(y)
+
+        # (Batch_Size, Seq_Len_Q, Dim_Q) -> (Batch_Size, Seq_Len_Q, H, Dim_Q / H) -> (Batch_Size, H, Seq_Len_Q, Dim_Q / H)
+        q = q.view(interim_shape).transpose(1, 2) 
+        # (Batch_Size, Seq_Len_KV, Dim_Q) -> (Batch_Size, Seq_Len_KV, H, Dim_Q / H) -> (Batch_Size, H, Seq_Len_KV, Dim_Q / H)
+        k = k.view(interim_shape).transpose(1, 2) 
+        # (Batch_Size, Seq_Len_KV, Dim_Q) -> (Batch_Size, Seq_Len_KV, H, Dim_Q / H) -> (Batch_Size, H, Seq_Len_KV, Dim_Q / H)
+        v = v.view(interim_shape).transpose(1, 2) 
+        
+        # (Batch_Size, H, Seq_Len_Q, Dim_Q / H) @ (Batch_Size, H, Dim_Q / H, Seq_Len_KV) -> (Batch_Size, H, Seq_Len_Q, Seq_Len_KV)
+        weight = q @ k.transpose(-1, -2)
+        
+        # (Batch_Size, H, Seq_Len_Q, Seq_Len_KV)
+        weight /= math.sqrt(self.d_head)
+        
+        # (Batch_Size, H, Seq_Len_Q, Seq_Len_KV)
+        weight = F.softmax(weight, dim=-1)
+        
+        # (Batch_Size, H, Seq_Len_Q, Seq_Len_KV) @ (Batch_Size, H, Seq_Len_KV, Dim_Q / H) -> (Batch_Size, H, Seq_Len_Q, Dim_Q / H)
+        output = weight @ v
+        
+        # (Batch_Size, H, Seq_Len_Q, Dim_Q / H) -> (Batch_Size, Seq_Len_Q, H, Dim_Q / H)
+        output = output.transpose(1, 2).contiguous()
+        
+        # (Batch_Size, Seq_Len_Q, H, Dim_Q / H) -> (Batch_Size, Seq_Len_Q, Dim_Q)
+        output = output.view(input_shape)
+        
+        # (Batch_Size, Seq_Len_Q, Dim_Q) -> (Batch_Size, Seq_Len_Q, Dim_Q)
+        output = self.out_proj(output)
+
+        # (Batch_Size, Seq_Len_Q, Dim_Q)
+        return output

SD/clip.py

@@ -0,0 +1,96 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from attention import SelfAttention
+
+class CLIPEmbedding(nn.Module):
+    def __init__(self, n_vocab: int, n_embd: int, n_token: int):
+        super().__init__()
+        
+        self.token_embedding = nn.Embedding(n_vocab, n_embd)
+        # A learnable weight matrix encodes the position information for each token
+        self.position_embedding = nn.Parameter(torch.zeros((n_token, n_embd)))
+    
+    def forward(self, tokens):
+        # (Batch_Size, Seq_Len) -> (Batch_Size, Seq_Len, Dim) 
+        x = self.token_embedding(tokens)
+        # (Batch_Size, Seq_Len) -> (Batch_Size, Seq_Len, Dim)
+        x += self.position_embedding
+        
+        return x
+
+class CLIPLayer(nn.Module):
+    def __init__(self, n_head: int, n_embd: int):
+        super().__init__()
+        
+        # Pre-attention norm
+        self.layernorm_1 = nn.LayerNorm(n_embd)
+        # Self attention
+        self.attention = SelfAttention(n_head, n_embd)
+        # Pre-FNN norm
+        self.layernorm_2 = nn.LayerNorm(n_embd)
+        # Feedforward layer
+        self.linear_1 = nn.Linear(n_embd, 4 * n_embd)
+        self.linear_2 = nn.Linear(4 * n_embd, n_embd)
+
+    def forward(self, x):
+        # (Batch_Size, Seq_Len, Dim)
+        residue = x
+        
+        ### SELF ATTENTION ###
+
+        # (Batch_Size, Seq_Len, Dim) -> (Batch_Size, Seq_Len, Dim)
+        x = self.layernorm_1(x)
+        
+        # (Batch_Size, Seq_Len, Dim) -> (Batch_Size, Seq_Len, Dim)
+        x = self.attention(x, causal_mask=True)
+        
+        # (Batch_Size, Seq_Len, Dim) + (Batch_Size, Seq_Len, Dim) -> (Batch_Size, Seq_Len, Dim)
+        x += residue
+
+        ### FEEDFORWARD LAYER ###
+        # Apply a feedforward layer where the hidden dimension is 4 times the embedding dimension. 
+
+        residue = x
+        # (Batch_Size, Seq_Len, Dim) -> (Batch_Size, Seq_Len, Dim)
+        x = self.layernorm_2(x)
+        
+        # (Batch_Size, Seq_Len, Dim) -> (Batch_Size, Seq_Len, 4 * Dim)
+        x = self.linear_1(x)
+        
+        # (Batch_Size, Seq_Len, 4 * Dim) -> (Batch_Size, Seq_Len, 4 * Dim)
+        x = x * torch.sigmoid(1.702 * x)   # QuickGELU activation function
+        
+        # (Batch_Size, Seq_Len, 4 * Dim) -> (Batch_Size, Seq_Len, Dim)
+        x = self.linear_2(x)
+        
+        # (Batch_Size, Seq_Len, Dim) + (Batch_Size, Seq_Len, Dim) -> (Batch_Size, Seq_Len, Dim)
+        x += residue
+
+        return x
+
+class CLIP(nn.Module):
+    def __init__(self):
+        super().__init__()
+        self.embedding = CLIPEmbedding(49408, 768, 77)
+
+        self.layers = nn.ModuleList([
+            CLIPLayer(12, 768) for i in range(12)
+        ])
+
+        self.layernorm = nn.LayerNorm(768)
+    
+    def forward(self, tokens: torch.LongTensor) -> torch.FloatTensor:
+        tokens = tokens.type(torch.long)
+        
+        # (Batch_Size, Seq_Len) -> (Batch_Size, Seq_Len, Dim)
+        state = self.embedding(tokens)
+
+        # Apply encoder layers similar to the Transformer's encoder.
+        for layer in self.layers: 
+            # (Batch_Size, Seq_Len, Dim) -> (Batch_Size, Seq_Len, Dim)
+            state = layer(state)
+        # (Batch_Size, Seq_Len, Dim) -> (Batch_Size, Seq_Len, Dim)
+        output = self.layernorm(state)
+        
+        return output

SD/ddpm.py

@@ -0,0 +1,123 @@
+import torch
+import numpy as np
+
+class DDPMSampler:
+
+    def __init__(self, generator: torch.Generator, num_training_steps=1000, beta_start: float = 0.00085, beta_end: float = 0.0120):
+        # Params "beta_start" and "beta_end" taken from: https://github.com/CompVis/stable-diffusion/blob/21f890f9da3cfbeaba8e2ac3c425ee9e998d5229/configs/stable-diffusion/v1-inference.yaml#L5C8-L5C8
+        # For the naming conventions, refer to the DDPM paper (https://arxiv.org/pdf/2006.11239.pdf)
+        self.betas = torch.linspace(beta_start ** 0.5, beta_end ** 0.5, num_training_steps, dtype=torch.float32) ** 2
+        self.alphas = 1.0 - self.betas
+        self.alphas_cumprod = torch.cumprod(self.alphas, dim=0)
+        self.one = torch.tensor(1.0)
+
+        self.generator = generator
+
+        self.num_train_timesteps = num_training_steps
+        self.timesteps = torch.from_numpy(np.arange(0, num_training_steps)[::-1].copy())
+
+    def set_inference_timesteps(self, num_inference_steps=50):
+        self.num_inference_steps = num_inference_steps
+        step_ratio = self.num_train_timesteps // self.num_inference_steps
+        timesteps = (np.arange(0, num_inference_steps) * step_ratio).round()[::-1].copy().astype(np.int64)
+        self.timesteps = torch.from_numpy(timesteps)
+
+    def _get_previous_timestep(self, timestep: int) -> int:
+        prev_t = timestep - self.num_train_timesteps // self.num_inference_steps
+        return prev_t
+    
+    def _get_variance(self, timestep: int) -> torch.Tensor:
+        prev_t = self._get_previous_timestep(timestep)
+
+        alpha_prod_t = self.alphas_cumprod[timestep]
+        alpha_prod_t_prev = self.alphas_cumprod[prev_t] if prev_t >= 0 else self.one
+        current_beta_t = 1 - alpha_prod_t / alpha_prod_t_prev
+
+        # For t > 0, compute predicted variance βt (see formula (6) and (7) from https://arxiv.org/pdf/2006.11239.pdf)
+        # and sample from it to get previous sample
+        # x_{t-1} ~ N(pred_prev_sample, variance) == add variance to pred_sample
+        variance = (1 - alpha_prod_t_prev) / (1 - alpha_prod_t) * current_beta_t
+
+        # we always take the log of variance, so clamp it to ensure it's not 0
+        variance = torch.clamp(variance, min=1e-20)
+
+        return variance
+    
+    def set_strength(self, strength=1):
+        """
+            Set how much noise to add to the input image. 
+            More noise (strength ~ 1) means that the output will be further from the input image.
+            Less noise (strength ~ 0) means that the output will be closer to the input image.
+        """
+        # start_step is the number of noise levels to skip
+        start_step = self.num_inference_steps - int(self.num_inference_steps * strength)
+        self.timesteps = self.timesteps[start_step:]
+        self.start_step = start_step
+
+    def step(self, timestep: int, latents: torch.Tensor, model_output: torch.Tensor):
+        t = timestep
+        prev_t = self._get_previous_timestep(t)
+
+        # 1. compute alphas, betas
+        alpha_prod_t = self.alphas_cumprod[t]
+        alpha_prod_t_prev = self.alphas_cumprod[prev_t] if prev_t >= 0 else self.one
+        beta_prod_t = 1 - alpha_prod_t
+        beta_prod_t_prev = 1 - alpha_prod_t_prev
+        current_alpha_t = alpha_prod_t / alpha_prod_t_prev
+        current_beta_t = 1 - current_alpha_t
+
+        # 2. compute predicted original sample from predicted noise also called
+        # "predicted x_0" of formula (15) from https://arxiv.org/pdf/2006.11239.pdf
+        pred_original_sample = (latents - beta_prod_t ** (0.5) * model_output) / alpha_prod_t ** (0.5)
+
+        # 4. Compute coefficients for pred_original_sample x_0 and current sample x_t
+        # See formula (7) from https://arxiv.org/pdf/2006.11239.pdf
+        pred_original_sample_coeff = (alpha_prod_t_prev ** (0.5) * current_beta_t) / beta_prod_t
+        current_sample_coeff = current_alpha_t ** (0.5) * beta_prod_t_prev / beta_prod_t
+
+        # 5. Compute predicted previous sample µ_t
+        # See formula (7) from https://arxiv.org/pdf/2006.11239.pdf
+        pred_prev_sample = pred_original_sample_coeff * pred_original_sample + current_sample_coeff * latents
+
+        # 6. Add noise
+        variance = 0
+        if t > 0:
+            device = model_output.device
+            noise = torch.randn(model_output.shape, generator=self.generator, device=device, dtype=model_output.dtype)
+            # Compute the variance as per formula (7) from https://arxiv.org/pdf/2006.11239.pdf
+            variance = (self._get_variance(t) ** 0.5) * noise
+        
+        # sample from N(mu, sigma) = X can be obtained by X = mu + sigma * N(0, 1)
+        # the variable "variance" is already multiplied by the noise N(0, 1)
+        pred_prev_sample = pred_prev_sample + variance
+
+        return pred_prev_sample
+    
+    def add_noise(
+        self,
+        original_samples: torch.FloatTensor,
+        timesteps: torch.IntTensor,
+    ) -> torch.FloatTensor:
+        alphas_cumprod = self.alphas_cumprod.to(device=original_samples.device, dtype=original_samples.dtype)
+        timesteps = timesteps.to(original_samples.device)
+
+        sqrt_alpha_prod = alphas_cumprod[timesteps] ** 0.5
+        sqrt_alpha_prod = sqrt_alpha_prod.flatten()
+        while len(sqrt_alpha_prod.shape) < len(original_samples.shape):
+            sqrt_alpha_prod = sqrt_alpha_prod.unsqueeze(-1)
+
+        sqrt_one_minus_alpha_prod = (1 - alphas_cumprod[timesteps]) ** 0.5
+        sqrt_one_minus_alpha_prod = sqrt_one_minus_alpha_prod.flatten()
+        while len(sqrt_one_minus_alpha_prod.shape) < len(original_samples.shape):
+            sqrt_one_minus_alpha_prod = sqrt_one_minus_alpha_prod.unsqueeze(-1)
+
+        # Sample from q(x_t | x_0) as in equation (4) of https://arxiv.org/pdf/2006.11239.pdf
+        # Because N(mu, sigma) = X can be obtained by X = mu + sigma * N(0, 1)
+        # here mu = sqrt_alpha_prod * original_samples and sigma = sqrt_one_minus_alpha_prod
+        noise = torch.randn(original_samples.shape, generator=self.generator, device=original_samples.device, dtype=original_samples.dtype)
+        noisy_samples = sqrt_alpha_prod * original_samples + sqrt_one_minus_alpha_prod * noise
+        return noisy_samples
+
+        
+
+    

SD/decoder.py

@@ -0,0 +1,177 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from attention import SelfAttention
+
+class VAE_AttentionBlock(nn.Module):
+    def __init__(self, channels):
+        super().__init__()
+        self.groupnorm = nn.GroupNorm(32, channels)
+        self.attention = SelfAttention(1, channels)
+    
+    def forward(self, x):
+        # x: (Batch_Size, Features, Height, Width)
+
+        residue = x 
+
+        # (Batch_Size, Features, Height, Width) -> (Batch_Size, Features, Height, Width)
+        x = self.groupnorm(x)
+
+        n, c, h, w = x.shape
+        
+        # (Batch_Size, Features, Height, Width) -> (Batch_Size, Features, Height * Width)
+        x = x.view((n, c, h * w))
+        
+        # (Batch_Size, Features, Height * Width) -> (Batch_Size, Height * Width, Features). Each pixel becomes a feature of size "Features", the sequence length is "Height * Width".
+        x = x.transpose(-1, -2)
+        
+        # Perform self-attention WITHOUT mask
+        # (Batch_Size, Height * Width, Features) -> (Batch_Size, Height * Width, Features)
+        x = self.attention(x)
+        
+        # (Batch_Size, Height * Width, Features) -> (Batch_Size, Features, Height * Width)
+        x = x.transpose(-1, -2)
+        
+        # (Batch_Size, Features, Height * Width) -> (Batch_Size, Features, Height, Width)
+        x = x.view((n, c, h, w))
+        
+        # (Batch_Size, Features, Height, Width) + (Batch_Size, Features, Height, Width) -> (Batch_Size, Features, Height, Width) 
+        x += residue
+
+        # (Batch_Size, Features, Height, Width)
+        return x 
+
+class VAE_ResidualBlock(nn.Module):
+    def __init__(self, in_channels, out_channels):
+        super().__init__()
+        self.groupnorm_1 = nn.GroupNorm(32, in_channels)
+        self.conv_1 = nn.Conv2d(in_channels, out_channels, kernel_size=3, padding=1)
+
+        self.groupnorm_2 = nn.GroupNorm(32, out_channels)
+        self.conv_2 = nn.Conv2d(out_channels, out_channels, kernel_size=3, padding=1)
+
+        if in_channels == out_channels:
+            self.residual_layer = nn.Identity()
+        else:
+            self.residual_layer = nn.Conv2d(in_channels, out_channels, kernel_size=1, padding=0)
+    
+    def forward(self, x):
+        # x: (Batch_Size, In_Channels, Height, Width)
+
+        residue = x
+
+        # (Batch_Size, In_Channels, Height, Width) -> (Batch_Size, In_Channels, Height, Width)
+        x = self.groupnorm_1(x)
+        
+        # (Batch_Size, In_Channels, Height, Width) -> (Batch_Size, In_Channels, Height, Width)
+        x = F.silu(x)
+        
+        # (Batch_Size, In_Channels, Height, Width) -> (Batch_Size, Out_Channels, Height, Width)
+        x = self.conv_1(x)
+        
+        # (Batch_Size, Out_Channels, Height, Width) -> (Batch_Size, Out_Channels, Height, Width)
+        x = self.groupnorm_2(x)
+        
+        # (Batch_Size, Out_Channels, Height, Width) -> (Batch_Size, Out_Channels, Height, Width)
+        x = F.silu(x)
+        
+        # (Batch_Size, Out_Channels, Height, Width) -> (Batch_Size, Out_Channels, Height, Width)
+        x = self.conv_2(x)
+        
+        # (Batch_Size, Out_Channels, Height, Width) -> (Batch_Size, Out_Channels, Height, Width)
+        return x + self.residual_layer(residue)
+
+class VAE_Decoder(nn.Sequential):
+    def __init__(self):
+        super().__init__(
+            # (Batch_Size, 4, Height / 8, Width / 8) -> (Batch_Size, 4, Height / 8, Width / 8)
+            nn.Conv2d(4, 4, kernel_size=1, padding=0),
+
+            # (Batch_Size, 4, Height / 8, Width / 8) -> (Batch_Size, 512, Height / 8, Width / 8)
+            nn.Conv2d(4, 512, kernel_size=3, padding=1),
+            
+            # (Batch_Size, 512, Height / 8, Width / 8) -> (Batch_Size, 512, Height / 8, Width / 8)
+            VAE_ResidualBlock(512, 512), 
+            
+            # (Batch_Size, 512, Height / 8, Width / 8) -> (Batch_Size, 512, Height / 8, Width / 8)
+            VAE_AttentionBlock(512), 
+            
+            # (Batch_Size, 512, Height / 8, Width / 8) -> (Batch_Size, 512, Height / 8, Width / 8)
+            VAE_ResidualBlock(512, 512), 
+            
+            # (Batch_Size, 512, Height / 8, Width / 8) -> (Batch_Size, 512, Height / 8, Width / 8)
+            VAE_ResidualBlock(512, 512), 
+            
+            # (Batch_Size, 512, Height / 8, Width / 8) -> (Batch_Size, 512, Height / 8, Width / 8)
+            VAE_ResidualBlock(512, 512), 
+            
+            # (Batch_Size, 512, Height / 8, Width / 8) -> (Batch_Size, 512, Height / 8, Width / 8)
+            VAE_ResidualBlock(512, 512), 
+            
+            # Repeats the rows and columns of the data by scale_factor (like when you resize an image by doubling its size).
+            # (Batch_Size, 512, Height / 8, Width / 8) -> (Batch_Size, 512, Height / 4, Width / 4)
+            nn.Upsample(scale_factor=2),
+            
+            # (Batch_Size, 512, Height / 4, Width / 4) -> (Batch_Size, 512, Height / 4, Width / 4)
+            nn.Conv2d(512, 512, kernel_size=3, padding=1), 
+            
+            # (Batch_Size, 512, Height / 4, Width / 4) -> (Batch_Size, 512, Height / 4, Width / 4)
+            VAE_ResidualBlock(512, 512), 
+            
+            # (Batch_Size, 512, Height / 4, Width / 4) -> (Batch_Size, 512, Height / 4, Width / 4)
+            VAE_ResidualBlock(512, 512), 
+            
+            # (Batch_Size, 512, Height / 4, Width / 4) -> (Batch_Size, 512, Height / 4, Width / 4)
+            VAE_ResidualBlock(512, 512), 
+            
+            # (Batch_Size, 512, Height / 4, Width / 4) -> (Batch_Size, 512, Height / 2, Width / 2)
+            nn.Upsample(scale_factor=2), 
+            
+            # (Batch_Size, 512, Height / 2, Width / 2) -> (Batch_Size, 512, Height / 2, Width / 2)
+            nn.Conv2d(512, 512, kernel_size=3, padding=1), 
+            
+            # (Batch_Size, 512, Height / 2, Width / 2) -> (Batch_Size, 256, Height / 2, Width / 2)
+            VAE_ResidualBlock(512, 256), 
+            
+            # (Batch_Size, 256, Height / 2, Width / 2) -> (Batch_Size, 256, Height / 2, Width / 2)
+            VAE_ResidualBlock(256, 256), 
+            
+            # (Batch_Size, 256, Height / 2, Width / 2) -> (Batch_Size, 256, Height / 2, Width / 2)
+            VAE_ResidualBlock(256, 256), 
+            
+            # (Batch_Size, 256, Height / 2, Width / 2) -> (Batch_Size, 256, Height, Width)
+            nn.Upsample(scale_factor=2), 
+            
+            # (Batch_Size, 256, Height, Width) -> (Batch_Size, 256, Height, Width)
+            nn.Conv2d(256, 256, kernel_size=3, padding=1), 
+            
+            # (Batch_Size, 256, Height, Width) -> (Batch_Size, 128, Height, Width)
+            VAE_ResidualBlock(256, 128), 
+            
+            # (Batch_Size, 128, Height, Width) -> (Batch_Size, 128, Height, Width)
+            VAE_ResidualBlock(128, 128), 
+            
+            # (Batch_Size, 128, Height, Width) -> (Batch_Size, 128, Height, Width)
+            VAE_ResidualBlock(128, 128), 
+            
+            # (Batch_Size, 128, Height, Width) -> (Batch_Size, 128, Height, Width)
+            nn.GroupNorm(32, 128), 
+            
+            # (Batch_Size, 128, Height, Width) -> (Batch_Size, 128, Height, Width)
+            nn.SiLU(), 
+            
+            # (Batch_Size, 128, Height, Width) -> (Batch_Size, 3, Height, Width)
+            nn.Conv2d(128, 3, kernel_size=3, padding=1), 
+        )
+
+    def forward(self, x):
+        # x: (Batch_Size, 4, Height / 8, Width / 8)
+        
+        # Remove the scaling added by the Encoder.
+        x /= 0.18215
+
+        for module in self:
+            x = module(x)
+
+        # (Batch_Size, 3, Height, Width)
+        return x

SD/diffusion.py

@@ -0,0 +1,349 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from attention import SelfAttention, CrossAttention
+
+class TimeEmbedding(nn.Module):
+    def __init__(self, n_embd):
+        super().__init__()
+        self.linear_1 = nn.Linear(n_embd, 4 * n_embd)
+        self.linear_2 = nn.Linear(4 * n_embd, 4 * n_embd)
+
+    def forward(self, x):
+        # x: (1, 320)
+
+        # (1, 320) -> (1, 1280)
+        x = self.linear_1(x)
+        
+        # (1, 1280) -> (1, 1280)
+        x = F.silu(x) 
+        
+        # (1, 1280) -> (1, 1280)
+        x = self.linear_2(x)
+
+        return x
+
+class UNET_ResidualBlock(nn.Module):
+    def __init__(self, in_channels, out_channels, n_time=1280):
+        super().__init__()
+        self.groupnorm_feature = nn.GroupNorm(32, in_channels)
+        self.conv_feature = nn.Conv2d(in_channels, out_channels, kernel_size=3, padding=1)
+        self.linear_time = nn.Linear(n_time, out_channels)
+
+        self.groupnorm_merged = nn.GroupNorm(32, out_channels)
+        self.conv_merged = nn.Conv2d(out_channels, out_channels, kernel_size=3, padding=1)
+
+        if in_channels == out_channels:
+            self.residual_layer = nn.Identity()
+        else:
+            self.residual_layer = nn.Conv2d(in_channels, out_channels, kernel_size=1, padding=0)
+    
+    def forward(self, feature, time):
+        # feature: (Batch_Size, In_Channels, Height, Width)
+        # time: (1, 1280)
+
+        residue = feature
+        
+        # (Batch_Size, In_Channels, Height, Width) -> (Batch_Size, In_Channels, Height, Width)
+        feature = self.groupnorm_feature(feature)
+        
+        # (Batch_Size, In_Channels, Height, Width) -> (Batch_Size, In_Channels, Height, Width)
+        feature = F.silu(feature)
+        
+        # (Batch_Size, In_Channels, Height, Width) -> (Batch_Size, Out_Channels, Height, Width)
+        feature = self.conv_feature(feature)
+        
+        # (1, 1280) -> (1, 1280)
+        time = F.silu(time)
+
+        # (1, 1280) -> (1, Out_Channels)
+        time = self.linear_time(time)
+        
+        # Add width and height dimension to time. 
+        # (Batch_Size, Out_Channels, Height, Width) + (1, Out_Channels, 1, 1) -> (Batch_Size, Out_Channels, Height, Width)
+        merged = feature + time.unsqueeze(-1).unsqueeze(-1)
+        
+        # (Batch_Size, Out_Channels, Height, Width) -> (Batch_Size, Out_Channels, Height, Width)
+        merged = self.groupnorm_merged(merged)
+        
+        # (Batch_Size, Out_Channels, Height, Width) -> (Batch_Size, Out_Channels, Height, Width)
+        merged = F.silu(merged)
+        
+        # (Batch_Size, Out_Channels, Height, Width) -> (Batch_Size, Out_Channels, Height, Width)
+        merged = self.conv_merged(merged)
+        
+        # (Batch_Size, Out_Channels, Height, Width) + (Batch_Size, Out_Channels, Height, Width) -> (Batch_Size, Out_Channels, Height, Width)
+        return merged + self.residual_layer(residue)
+
+class UNET_AttentionBlock(nn.Module):
+    def __init__(self, n_head: int, n_embd: int, d_context=768):
+        super().__init__()
+        channels = n_head * n_embd
+        
+        self.groupnorm = nn.GroupNorm(32, channels, eps=1e-6)
+        self.conv_input = nn.Conv2d(channels, channels, kernel_size=1, padding=0)
+
+        self.layernorm_1 = nn.LayerNorm(channels)
+        self.attention_1 = SelfAttention(n_head, channels, in_proj_bias=False)
+        self.layernorm_2 = nn.LayerNorm(channels)
+        self.attention_2 = CrossAttention(n_head, channels, d_context, in_proj_bias=False)
+        self.layernorm_3 = nn.LayerNorm(channels)
+        self.linear_geglu_1  = nn.Linear(channels, 4 * channels * 2)
+        self.linear_geglu_2 = nn.Linear(4 * channels, channels)
+
+        self.conv_output = nn.Conv2d(channels, channels, kernel_size=1, padding=0)
+    
+    def forward(self, x, context):
+        # x: (Batch_Size, Features, Height, Width)
+        # context: (Batch_Size, Seq_Len, Dim)
+
+        residue_long = x
+
+        # (Batch_Size, Features, Height, Width) -> (Batch_Size, Features, Height, Width)
+        x = self.groupnorm(x)
+        
+        # (Batch_Size, Features, Height, Width) -> (Batch_Size, Features, Height, Width)
+        x = self.conv_input(x)
+        
+        n, c, h, w = x.shape
+        
+        # (Batch_Size, Features, Height, Width) -> (Batch_Size, Features, Height * Width)
+        x = x.view((n, c, h * w))
+        
+        # (Batch_Size, Features, Height * Width) -> (Batch_Size, Height * Width, Features)
+        x = x.transpose(-1, -2)
+        
+        # Normalization + Self-Attention with skip connection
+
+        # (Batch_Size, Height * Width, Features)
+        residue_short = x
+        
+        # (Batch_Size, Height * Width, Features) -> (Batch_Size, Height * Width, Features)
+        x = self.layernorm_1(x)
+        
+        # (Batch_Size, Height * Width, Features) -> (Batch_Size, Height * Width, Features)
+        x = self.attention_1(x)
+        
+        # (Batch_Size, Height * Width, Features) + (Batch_Size, Height * Width, Features) -> (Batch_Size, Height * Width, Features)
+        x += residue_short
+        
+        # (Batch_Size, Height * Width, Features)
+        residue_short = x
+
+        # Normalization + Cross-Attention with skip connection
+        
+        # (Batch_Size, Height * Width, Features) -> (Batch_Size, Height * Width, Features)
+        x = self.layernorm_2(x)
+        
+        # (Batch_Size, Height * Width, Features) -> (Batch_Size, Height * Width, Features)
+        x = self.attention_2(x, context)
+        
+        # (Batch_Size, Height * Width, Features) + (Batch_Size, Height * Width, Features) -> (Batch_Size, Height * Width, Features)
+        x += residue_short
+        
+        # (Batch_Size, Height * Width, Features)
+        residue_short = x
+
+        # Normalization + FFN with GeGLU and skip connection
+        
+        # (Batch_Size, Height * Width, Features) -> (Batch_Size, Height * Width, Features)
+        x = self.layernorm_3(x)
+        
+        # GeGLU as implemented in the original code: https://github.com/CompVis/stable-diffusion/blob/21f890f9da3cfbeaba8e2ac3c425ee9e998d5229/ldm/modules/attention.py#L37C10-L37C10
+        # (Batch_Size, Height * Width, Features) -> two tensors of shape (Batch_Size, Height * Width, Features * 4)
+        x, gate = self.linear_geglu_1(x).chunk(2, dim=-1) 
+        
+        # Element-wise product: (Batch_Size, Height * Width, Features * 4) * (Batch_Size, Height * Width, Features * 4) -> (Batch_Size, Height * Width, Features * 4)
+        x = x * F.gelu(gate)
+        
+        # (Batch_Size, Height * Width, Features * 4) -> (Batch_Size, Height * Width, Features)
+        x = self.linear_geglu_2(x)
+        
+        # (Batch_Size, Height * Width, Features) + (Batch_Size, Height * Width, Features) -> (Batch_Size, Height * Width, Features)
+        x += residue_short
+        
+        # (Batch_Size, Height * Width, Features) -> (Batch_Size, Features, Height * Width)
+        x = x.transpose(-1, -2)
+        
+        # (Batch_Size, Features, Height * Width) -> (Batch_Size, Features, Height, Width)
+        x = x.view((n, c, h, w))
+
+        # Final skip connection between initial input and output of the block
+        # (Batch_Size, Features, Height, Width) + (Batch_Size, Features, Height, Width) -> (Batch_Size, Features, Height, Width)
+        return self.conv_output(x) + residue_long
+
+class Upsample(nn.Module):
+    def __init__(self, channels):
+        super().__init__()
+        self.conv = nn.Conv2d(channels, channels, kernel_size=3, padding=1)
+    
+    def forward(self, x):
+        # (Batch_Size, Features, Height, Width) -> (Batch_Size, Features, Height * 2, Width * 2)
+        x = F.interpolate(x, scale_factor=2, mode='nearest') 
+        return self.conv(x)
+
+class SwitchSequential(nn.Sequential):
+    def forward(self, x, context, time):
+        for layer in self:
+            if isinstance(layer, UNET_AttentionBlock):
+                x = layer(x, context)
+            elif isinstance(layer, UNET_ResidualBlock):
+                x = layer(x, time)
+            else:
+                x = layer(x)
+        return x
+
+class UNET(nn.Module):
+    def __init__(self):
+        super().__init__()
+        self.encoders = nn.ModuleList([
+            # (Batch_Size, 4, Height / 8, Width / 8) -> (Batch_Size, 320, Height / 8, Width / 8)
+            SwitchSequential(nn.Conv2d(4, 320, kernel_size=3, padding=1)),
+            
+            # (Batch_Size, 320, Height / 8, Width / 8) -> # (Batch_Size, 320, Height / 8, Width / 8) -> (Batch_Size, 320, Height / 8, Width / 8)
+            SwitchSequential(UNET_ResidualBlock(320, 320), UNET_AttentionBlock(8, 40)),
+            
+            # (Batch_Size, 320, Height / 8, Width / 8) -> # (Batch_Size, 320, Height / 8, Width / 8) -> (Batch_Size, 320, Height / 8, Width / 8)
+            SwitchSequential(UNET_ResidualBlock(320, 320), UNET_AttentionBlock(8, 40)),
+            
+            # (Batch_Size, 320, Height / 8, Width / 8) -> (Batch_Size, 320, Height / 16, Width / 16)
+            SwitchSequential(nn.Conv2d(320, 320, kernel_size=3, stride=2, padding=1)),
+            
+            # (Batch_Size, 320, Height / 16, Width / 16) -> (Batch_Size, 640, Height / 16, Width / 16) -> (Batch_Size, 640, Height / 16, Width / 16)
+            SwitchSequential(UNET_ResidualBlock(320, 640), UNET_AttentionBlock(8, 80)),
+            
+            # (Batch_Size, 640, Height / 16, Width / 16) -> (Batch_Size, 640, Height / 16, Width / 16) -> (Batch_Size, 640, Height / 16, Width / 16)
+            SwitchSequential(UNET_ResidualBlock(640, 640), UNET_AttentionBlock(8, 80)),
+            
+            # (Batch_Size, 640, Height / 16, Width / 16) -> (Batch_Size, 640, Height / 32, Width / 32)
+            SwitchSequential(nn.Conv2d(640, 640, kernel_size=3, stride=2, padding=1)),
+            
+            # (Batch_Size, 640, Height / 32, Width / 32) -> (Batch_Size, 1280, Height / 32, Width / 32) -> (Batch_Size, 1280, Height / 32, Width / 32)
+            SwitchSequential(UNET_ResidualBlock(640, 1280), UNET_AttentionBlock(8, 160)),
+            
+            # (Batch_Size, 1280, Height / 32, Width / 32) -> (Batch_Size, 1280, Height / 32, Width / 32) -> (Batch_Size, 1280, Height / 32, Width / 32)
+            SwitchSequential(UNET_ResidualBlock(1280, 1280), UNET_AttentionBlock(8, 160)),
+            
+            # (Batch_Size, 1280, Height / 32, Width / 32) -> (Batch_Size, 1280, Height / 64, Width / 64)
+            SwitchSequential(nn.Conv2d(1280, 1280, kernel_size=3, stride=2, padding=1)),
+            
+            # (Batch_Size, 1280, Height / 64, Width / 64) -> (Batch_Size, 1280, Height / 64, Width / 64)
+            SwitchSequential(UNET_ResidualBlock(1280, 1280)),
+            
+            # (Batch_Size, 1280, Height / 64, Width / 64) -> (Batch_Size, 1280, Height / 64, Width / 64)
+            SwitchSequential(UNET_ResidualBlock(1280, 1280)),
+        ])
+
+        self.bottleneck = SwitchSequential(
+            # (Batch_Size, 1280, Height / 64, Width / 64) -> (Batch_Size, 1280, Height / 64, Width / 64)
+            UNET_ResidualBlock(1280, 1280), 
+            
+            # (Batch_Size, 1280, Height / 64, Width / 64) -> (Batch_Size, 1280, Height / 64, Width / 64)
+            UNET_AttentionBlock(8, 160), 
+            
+            # (Batch_Size, 1280, Height / 64, Width / 64) -> (Batch_Size, 1280, Height / 64, Width / 64)
+            UNET_ResidualBlock(1280, 1280), 
+        )
+        
+        self.decoders = nn.ModuleList([
+            # (Batch_Size, 2560, Height / 64, Width / 64) -> (Batch_Size, 1280, Height / 64, Width / 64)
+            SwitchSequential(UNET_ResidualBlock(2560, 1280)),
+            
+            # (Batch_Size, 2560, Height / 64, Width / 64) -> (Batch_Size, 1280, Height / 64, Width / 64)
+            SwitchSequential(UNET_ResidualBlock(2560, 1280)),
+            
+            # (Batch_Size, 2560, Height / 64, Width / 64) -> (Batch_Size, 1280, Height / 64, Width / 64) -> (Batch_Size, 1280, Height / 32, Width / 32) 
+            SwitchSequential(UNET_ResidualBlock(2560, 1280), Upsample(1280)),
+            
+            # (Batch_Size, 2560, Height / 32, Width / 32) -> (Batch_Size, 1280, Height / 32, Width / 32) -> (Batch_Size, 1280, Height / 32, Width / 32)
+            SwitchSequential(UNET_ResidualBlock(2560, 1280), UNET_AttentionBlock(8, 160)),
+            
+            # (Batch_Size, 2560, Height / 32, Width / 32) -> (Batch_Size, 1280, Height / 32, Width / 32) -> (Batch_Size, 1280, Height / 32, Width / 32)
+            SwitchSequential(UNET_ResidualBlock(2560, 1280), UNET_AttentionBlock(8, 160)),
+            
+            # (Batch_Size, 1920, Height / 32, Width / 32) -> (Batch_Size, 1280, Height / 32, Width / 32) -> (Batch_Size, 1280, Height / 32, Width / 32) -> (Batch_Size, 1280, Height / 16, Width / 16)
+            SwitchSequential(UNET_ResidualBlock(1920, 1280), UNET_AttentionBlock(8, 160), Upsample(1280)),
+            
+            # (Batch_Size, 1920, Height / 16, Width / 16) -> (Batch_Size, 640, Height / 16, Width / 16) -> (Batch_Size, 640, Height / 16, Width / 16)
+            SwitchSequential(UNET_ResidualBlock(1920, 640), UNET_AttentionBlock(8, 80)),
+            
+            # (Batch_Size, 1280, Height / 16, Width / 16) -> (Batch_Size, 640, Height / 16, Width / 16) -> (Batch_Size, 640, Height / 16, Width / 16)
+            SwitchSequential(UNET_ResidualBlock(1280, 640), UNET_AttentionBlock(8, 80)),
+            
+            # (Batch_Size, 960, Height / 16, Width / 16) -> (Batch_Size, 640, Height / 16, Width / 16) -> (Batch_Size, 640, Height / 16, Width / 16) -> (Batch_Size, 640, Height / 8, Width / 8)
+            SwitchSequential(UNET_ResidualBlock(960, 640), UNET_AttentionBlock(8, 80), Upsample(640)),
+            
+            # (Batch_Size, 960, Height / 8, Width / 8) -> (Batch_Size, 320, Height / 8, Width / 8) -> (Batch_Size, 320, Height / 8, Width / 8)
+            SwitchSequential(UNET_ResidualBlock(960, 320), UNET_AttentionBlock(8, 40)),
+            
+            # (Batch_Size, 640, Height / 8, Width / 8) -> (Batch_Size, 320, Height / 8, Width / 8) -> (Batch_Size, 320, Height / 8, Width / 8)
+            SwitchSequential(UNET_ResidualBlock(640, 320), UNET_AttentionBlock(8, 40)),
+            
+            # (Batch_Size, 640, Height / 8, Width / 8) -> (Batch_Size, 320, Height / 8, Width / 8) -> (Batch_Size, 320, Height / 8, Width / 8)
+            SwitchSequential(UNET_ResidualBlock(640, 320), UNET_AttentionBlock(8, 40)),
+        ])
+
+    def forward(self, x, context, time):
+        # x: (Batch_Size, 4, Height / 8, Width / 8)
+        # context: (Batch_Size, Seq_Len, Dim) 
+        # time: (1, 1280)
+
+        skip_connections = []
+        for layers in self.encoders:
+            x = layers(x, context, time)
+            skip_connections.append(x)
+
+        x = self.bottleneck(x, context, time)
+
+        for layers in self.decoders:
+            # Since we always concat with the skip connection of the encoder, the number of features increases before being sent to the decoder's layer
+            x = torch.cat((x, skip_connections.pop()), dim=1) 
+            x = layers(x, context, time)
+        
+        return x
+
+
+class UNET_OutputLayer(nn.Module):
+    def __init__(self, in_channels, out_channels):
+        super().__init__()
+        self.groupnorm = nn.GroupNorm(32, in_channels)
+        self.conv = nn.Conv2d(in_channels, out_channels, kernel_size=3, padding=1)
+    
+    def forward(self, x):
+        # x: (Batch_Size, 320, Height / 8, Width / 8)
+
+        # (Batch_Size, 320, Height / 8, Width / 8) -> (Batch_Size, 320, Height / 8, Width / 8)
+        x = self.groupnorm(x)
+        
+        # (Batch_Size, 320, Height / 8, Width / 8) -> (Batch_Size, 320, Height / 8, Width / 8)
+        x = F.silu(x)
+        
+        # (Batch_Size, 320, Height / 8, Width / 8) -> (Batch_Size, 4, Height / 8, Width / 8)
+        x = self.conv(x)
+        
+        # (Batch_Size, 4, Height / 8, Width / 8) 
+        return x
+
+class Diffusion(nn.Module):
+    def __init__(self):
+        super().__init__()
+        self.time_embedding = TimeEmbedding(320)
+        self.unet = UNET()
+        self.final = UNET_OutputLayer(320, 4)
+    
+    def forward(self, latent, context, time):
+        # latent: (Batch_Size, 4, Height / 8, Width / 8)
+        # context: (Batch_Size, Seq_Len, Dim)
+        # time: (1, 320)
+
+        # (1, 320) -> (1, 1280)
+        time = self.time_embedding(time)
+        
+        # (Batch, 4, Height / 8, Width / 8) -> (Batch, 320, Height / 8, Width / 8)
+        output = self.unet(latent, context, time)
+        
+        # (Batch, 320, Height / 8, Width / 8) -> (Batch, 4, Height / 8, Width / 8)
+        output = self.final(output)
+        
+        # (Batch, 4, Height / 8, Width / 8)
+        return output

SD/encoder.py

@@ -0,0 +1,103 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from decoder import VAE_AttentionBlock, VAE_ResidualBlock
+
+class VAE_Encoder(nn.Sequential):
+    def __init__(self):
+        super().__init__(
+            # (Batch_Size, Channel, Height, Width) -> (Batch_Size, 128, Height, Width)
+            nn.Conv2d(3, 128, kernel_size=3, padding=1),
+            
+             # (Batch_Size, 128, Height, Width) -> (Batch_Size, 128, Height, Width)
+            VAE_ResidualBlock(128, 128),
+            
+            # (Batch_Size, 128, Height, Width) -> (Batch_Size, 128, Height, Width)
+            VAE_ResidualBlock(128, 128),
+            
+            # (Batch_Size, 128, Height, Width) -> (Batch_Size, 128, Height / 2, Width / 2)
+            nn.Conv2d(128, 128, kernel_size=3, stride=2, padding=0),
+            
+            # (Batch_Size, 128, Height / 2, Width / 2) -> (Batch_Size, 256, Height / 2, Width / 2)
+            VAE_ResidualBlock(128, 256), 
+            
+            # (Batch_Size, 256, Height / 2, Width / 2) -> (Batch_Size, 256, Height / 2, Width / 2)
+            VAE_ResidualBlock(256, 256), 
+            
+            # (Batch_Size, 256, Height / 2, Width / 2) -> (Batch_Size, 256, Height / 4, Width / 4)
+            nn.Conv2d(256, 256, kernel_size=3, stride=2, padding=0), 
+            
+            # (Batch_Size, 256, Height / 4, Width / 4) -> (Batch_Size, 512, Height / 4, Width / 4)
+            VAE_ResidualBlock(256, 512), 
+            
+            # (Batch_Size, 512, Height / 4, Width / 4) -> (Batch_Size, 512, Height / 4, Width / 4)
+            VAE_ResidualBlock(512, 512), 
+            
+            # (Batch_Size, 512, Height / 4, Width / 4) -> (Batch_Size, 512, Height / 8, Width / 8)
+            nn.Conv2d(512, 512, kernel_size=3, stride=2, padding=0), 
+            
+            # (Batch_Size, 512, Height / 8, Width / 8) -> (Batch_Size, 512, Height / 8, Width / 8)
+            VAE_ResidualBlock(512, 512), 
+            
+            # (Batch_Size, 512, Height / 8, Width / 8) -> (Batch_Size, 512, Height / 8, Width / 8)
+            VAE_ResidualBlock(512, 512), 
+            
+            # (Batch_Size, 512, Height / 8, Width / 8) -> (Batch_Size, 512, Height / 8, Width / 8)
+            VAE_ResidualBlock(512, 512), 
+            
+            # (Batch_Size, 512, Height / 8, Width / 8) -> (Batch_Size, 512, Height / 8, Width / 8)
+            VAE_AttentionBlock(512), 
+            
+            # (Batch_Size, 512, Height / 8, Width / 8) -> (Batch_Size, 512, Height / 8, Width / 8)
+            VAE_ResidualBlock(512, 512), 
+            
+            # (Batch_Size, 512, Height / 8, Width / 8) -> (Batch_Size, 512, Height / 8, Width / 8)
+            nn.GroupNorm(32, 512), 
+            
+            # (Batch_Size, 512, Height / 8, Width / 8) -> (Batch_Size, 512, Height / 8, Width / 8)
+            nn.SiLU(), 
+
+            # Because the padding=1, it means the width and height will increase by 2
+            # Out_Height = In_Height + Padding_Top + Padding_Bottom
+            # Out_Width = In_Width + Padding_Left + Padding_Right
+            # Since padding = 1 means Padding_Top = Padding_Bottom = Padding_Left = Padding_Right = 1,
+            # Since the Out_Width = In_Width + 2 (same for Out_Height), it will compensate for the Kernel size of 3
+            # (Batch_Size, 512, Height / 8, Width / 8) -> (Batch_Size, 8, Height / 8, Width / 8). 
+            nn.Conv2d(512, 8, kernel_size=3, padding=1), 
+
+            # (Batch_Size, 8, Height / 8, Width / 8) -> (Batch_Size, 8, Height / 8, Width / 8)
+            nn.Conv2d(8, 8, kernel_size=1, padding=0), 
+        )
+
+    def forward(self, x, noise):
+        # x: (Batch_Size, Channel, Height, Width)
+        # noise: (Batch_Size, 4, Height / 8, Width / 8)
+
+        for module in self:
+
+            if getattr(module, 'stride', None) == (2, 2):  # Padding at downsampling should be asymmetric (see #8)
+                # Pad: (Padding_Left, Padding_Right, Padding_Top, Padding_Bottom).
+                # Pad with zeros on the right and bottom.
+                # (Batch_Size, Channel, Height, Width) -> (Batch_Size, Channel, Height + Padding_Top + Padding_Bottom, Width + Padding_Left + Padding_Right) = (Batch_Size, Channel, Height + 1, Width + 1)
+                x = F.pad(x, (0, 1, 0, 1))
+            
+            x = module(x)
+        # (Batch_Size, 8, Height / 8, Width / 8) -> two tensors of shape (Batch_Size, 4, Height / 8, Width / 8)
+        mean, log_variance = torch.chunk(x, 2, dim=1)
+        # Clamp the log variance between -30 and 20, so that the variance is between (circa) 1e-14 and 1e8. 
+        # (Batch_Size, 4, Height / 8, Width / 8) -> (Batch_Size, 4, Height / 8, Width / 8)
+        log_variance = torch.clamp(log_variance, -30, 20)
+        # (Batch_Size, 4, Height / 8, Width / 8) -> (Batch_Size, 4, Height / 8, Width / 8)
+        variance = log_variance.exp()
+        # (Batch_Size, 4, Height / 8, Width / 8) -> (Batch_Size, 4, Height / 8, Width / 8)
+        stdev = variance.sqrt()
+        
+        # Transform N(0, 1) -> N(mean, stdev) 
+        # (Batch_Size, 4, Height / 8, Width / 8) -> (Batch_Size, 4, Height / 8, Width / 8)
+        x = mean + stdev * noise
+        
+        # Scale by a constant
+        # Constant taken from: https://github.com/CompVis/stable-diffusion/blob/21f890f9da3cfbeaba8e2ac3c425ee9e998d5229/configs/stable-diffusion/v1-inference.yaml#L17C1-L17C1
+        x *= 0.18215
+        
+        return x

SD/model_loader.py

@@ -0,0 +1,28 @@
+from clip import CLIP
+from encoder import VAE_Encoder
+from decoder import VAE_Decoder
+from diffusion import Diffusion
+
+import model_converter
+
+def preload_models_from_standard_weights(ckpt_path, device):
+    state_dict = model_converter.load_from_standard_weights(ckpt_path, device)
+
+    encoder = VAE_Encoder().to(device)
+    encoder.load_state_dict(state_dict['encoder'], strict=True)
+
+    decoder = VAE_Decoder().to(device)
+    decoder.load_state_dict(state_dict['decoder'], strict=True)
+
+    diffusion = Diffusion().to(device)
+    diffusion.load_state_dict(state_dict['diffusion'], strict=True)
+
+    clip = CLIP().to(device)
+    clip.load_state_dict(state_dict['clip'], strict=True)
+
+    return {
+        'clip': clip,
+        'encoder': encoder,
+        'decoder': decoder,
+        'diffusion': diffusion,
+    }

SD/pipeline.py

@@ -0,0 +1,170 @@
+import torch
+import numpy as np
+from tqdm import tqdm
+from ddpm import DDPMSampler
+
+WIDTH = 512
+HEIGHT = 512
+LATENTS_WIDTH = WIDTH // 8
+LATENTS_HEIGHT = HEIGHT // 8
+
+def generate(
+    prompt,
+    uncond_prompt=None,
+    input_image=None,
+    strength=0.8,
+    do_cfg=True,
+    cfg_scale=7.5,
+    sampler_name="ddpm",
+    n_inference_steps=50,
+    models={},
+    seed=None,
+    device=None,
+    idle_device=None,
+    tokenizer=None,
+):
+    with torch.no_grad():
+        if not 0 < strength <= 1:
+            raise ValueError("strength must be between 0 and 1")
+
+        if idle_device:
+            to_idle = lambda x: x.to(idle_device)
+        else:
+            to_idle = lambda x: x
+
+        # Initialize random number generator according to the seed specified
+        generator = torch.Generator(device=device)
+        if seed is None:
+            generator.seed()
+        else:
+            generator.manual_seed(seed)
+
+        clip = models["clip"]
+        clip.to(device)
+        
+        if do_cfg:
+            # Convert into a list of length Seq_Len=77
+            cond_tokens = tokenizer.batch_encode_plus(
+                [prompt], padding="max_length", max_length=77
+            ).input_ids
+            # (Batch_Size, Seq_Len)
+            cond_tokens = torch.tensor(cond_tokens, dtype=torch.long, device=device)
+            # (Batch_Size, Seq_Len) -> (Batch_Size, Seq_Len, Dim)
+            cond_context = clip(cond_tokens)
+            # Convert into a list of length Seq_Len=77
+            uncond_tokens = tokenizer.batch_encode_plus(
+                [uncond_prompt], padding="max_length", max_length=77
+            ).input_ids
+            # (Batch_Size, Seq_Len)
+            uncond_tokens = torch.tensor(uncond_tokens, dtype=torch.long, device=device)
+            # (Batch_Size, Seq_Len) -> (Batch_Size, Seq_Len, Dim)
+            uncond_context = clip(uncond_tokens)
+            # (Batch_Size, Seq_Len, Dim) + (Batch_Size, Seq_Len, Dim) -> (2 * Batch_Size, Seq_Len, Dim)
+            context = torch.cat([cond_context, uncond_context])
+        else:
+            # Convert into a list of length Seq_Len=77
+            tokens = tokenizer.batch_encode_plus(
+                [prompt], padding="max_length", max_length=77
+            ).input_ids
+            # (Batch_Size, Seq_Len)
+            tokens = torch.tensor(tokens, dtype=torch.long, device=device)
+            # (Batch_Size, Seq_Len) -> (Batch_Size, Seq_Len, Dim)
+            context = clip(tokens)
+        to_idle(clip)
+
+        if sampler_name == "ddpm":
+            sampler = DDPMSampler(generator)
+            sampler.set_inference_timesteps(n_inference_steps)
+        else:
+            raise ValueError("Unknown sampler value %s. ")
+
+        latents_shape = (1, 4, LATENTS_HEIGHT, LATENTS_WIDTH)
+
+        if input_image:
+            encoder = models["encoder"]
+            encoder.to(device)
+
+            input_image_tensor = input_image.resize((WIDTH, HEIGHT))
+            # (Height, Width, Channel)
+            input_image_tensor = np.array(input_image_tensor)
+            # (Height, Width, Channel) -> (Height, Width, Channel)
+            input_image_tensor = torch.tensor(input_image_tensor, dtype=torch.float32, device=device)
+            # (Height, Width, Channel) -> (Height, Width, Channel)
+            input_image_tensor = rescale(input_image_tensor, (0, 255), (-1, 1))
+            # (Height, Width, Channel) -> (Batch_Size, Height, Width, Channel)
+            input_image_tensor = input_image_tensor.unsqueeze(0)
+            # (Batch_Size, Height, Width, Channel) -> (Batch_Size, Channel, Height, Width)
+            input_image_tensor = input_image_tensor.permute(0, 3, 1, 2)
+
+            # (Batch_Size, 4, Latents_Height, Latents_Width)
+            encoder_noise = torch.randn(latents_shape, generator=generator, device=device)
+            # (Batch_Size, 4, Latents_Height, Latents_Width)
+            latents = encoder(input_image_tensor, encoder_noise)
+
+            # Add noise to the latents (the encoded input image)
+            # (Batch_Size, 4, Latents_Height, Latents_Width)
+            sampler.set_strength(strength=strength)
+            latents = sampler.add_noise(latents, sampler.timesteps[0])
+
+            to_idle(encoder)
+        else:
+            # (Batch_Size, 4, Latents_Height, Latents_Width)
+            latents = torch.randn(latents_shape, generator=generator, device=device)
+
+        diffusion = models["diffusion"]
+        diffusion.to(device)
+
+        timesteps = tqdm(sampler.timesteps)
+        for i, timestep in enumerate(timesteps):
+            # (1, 320)
+            time_embedding = get_time_embedding(timestep).to(device)
+
+            # (Batch_Size, 4, Latents_Height, Latents_Width)
+            model_input = latents
+
+            if do_cfg:
+                # (Batch_Size, 4, Latents_Height, Latents_Width) -> (2 * Batch_Size, 4, Latents_Height, Latents_Width)
+                model_input = model_input.repeat(2, 1, 1, 1)
+
+            # model_output is the predicted noise
+            # (Batch_Size, 4, Latents_Height, Latents_Width) -> (Batch_Size, 4, Latents_Height, Latents_Width)
+            model_output = diffusion(model_input, context, time_embedding)
+
+            if do_cfg:
+                output_cond, output_uncond = model_output.chunk(2)
+                model_output = cfg_scale * (output_cond - output_uncond) + output_uncond
+
+            # (Batch_Size, 4, Latents_Height, Latents_Width) -> (Batch_Size, 4, Latents_Height, Latents_Width)
+            latents = sampler.step(timestep, latents, model_output)
+
+        to_idle(diffusion)
+
+        decoder = models["decoder"]
+        decoder.to(device)
+        # (Batch_Size, 4, Latents_Height, Latents_Width) -> (Batch_Size, 3, Height, Width)
+        images = decoder(latents)
+        to_idle(decoder)
+
+        images = rescale(images, (-1, 1), (0, 255), clamp=True)
+        # (Batch_Size, Channel, Height, Width) -> (Batch_Size, Height, Width, Channel)
+        images = images.permute(0, 2, 3, 1)
+        images = images.to("cpu", torch.uint8).numpy()
+        return images[0]
+    
+def rescale(x, old_range, new_range, clamp=False):
+    old_min, old_max = old_range
+    new_min, new_max = new_range
+    x -= old_min
+    x *= (new_max - new_min) / (old_max - old_min)
+    x += new_min
+    if clamp:
+        x = x.clamp(new_min, new_max)
+    return x
+
+def get_time_embedding(timestep):
+    # Shape: (160,)
+    freqs = torch.pow(10000, -torch.arange(start=0, end=160, dtype=torch.float32) / 160) 
+    # Shape: (1, 160)
+    x = torch.tensor([timestep], dtype=torch.float32)[:, None] * freqs[None]
+    # Shape: (1, 160 * 2)
+    return torch.cat([torch.cos(x), torch.sin(x)], dim=-1)

SD/run.py

@@ -0,0 +1,64 @@
+import model_loader
+import pipeline
+from PIL import Image
+from pathlib import Path
+from transformers import CLIPTokenizer
+import torch
+
+
+DEVICE = "cpu"
+
+ALLOW_CUDA = True
+ALLOW_MPS = False
+
+if torch.cuda.is_available() and ALLOW_CUDA:
+    DEVICE = "cuda"
+
+print(f"Using device: {DEVICE}")
+
+tokenizer = CLIPTokenizer("../data/tokenizer_vocab.json", merges_file="../data/tokenizer_merges.txt")
+model_file = "../data/v1-5-pruned-emaonly.ckpt"
+models = model_loader.preload_models_from_standard_weights(model_file, device=DEVICE)
+
+## TEXT TO IMAGE
+
+# prompt = "A dog with sunglasses, wearing comfy hat, looking at camera, highly detailed, ultra sharp, cinematic, 100mm lens, 8k resolution."
+prompt = "A cat stretching on the floor, highly detailed, ultra sharp, cinematic, 100mm lens, 8k resolution."
+uncond_prompt = ""  # Also known as negative prom pt
+do_cfg = True
+cfg_scale = 8  # min: 1, max: 14
+
+## IMAGE TO IMAGE
+
+input_image = None
+# Comment to disable image to image
+image_path = "../images/dog.jpg"
+# input_image = Image.open(image_path)
+# Higher values means more noise will be added to the input image, so the result will further from the input image.
+# Lower values means less noise is added to the input image, so output will be closer to the input image.
+strength = 0.9
+
+## SAMPLER
+
+sampler = "ddpm"
+num_inference_steps = 2
+seed = 42
+
+output_image = pipeline.generate(
+    prompt=prompt,
+    uncond_prompt=uncond_prompt,
+    input_image=input_image,
+    strength=strength,
+    do_cfg=do_cfg,
+    cfg_scale=cfg_scale,
+    sampler_name=sampler,
+    n_inference_steps=num_inference_steps,
+    seed=seed,
+    models=models,
+    device=DEVICE,
+    idle_device="cpu",
+    tokenizer=tokenizer,
+)
+
+# Combine the input image and the output image into a single image.
+Image.fromarray(output_image)

SD_Inkpunk_V1.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:629ddef95988fd88760808067c8b92625061937e153ab8eff99c933c1516f5d8
+size 2132856622

SD_Inkpunk_V2.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2182245415908822cbac065128a4c5144cc547d0701feb21241cb4e70bb5cf56
+size 2132856622

feature_extractor/preprocessor_config.json

@@ -0,0 +1,20 @@
+{
+  "crop_size": 224,
+  "do_center_crop": true,
+  "do_convert_rgb": true,
+  "do_normalize": true,
+  "do_resize": true,
+  "feature_extractor_type": "CLIPFeatureExtractor",
+  "image_mean": [
+    0.48145466,
+    0.4578275,
+    0.40821073
+  ],
+  "image_std": [
+    0.26862954,
+    0.26130258,
+    0.27577711
+  ],
+  "resample": 3,
+  "size": 224
+}

license.txt

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) [year] [fullname]
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

model_index.json

@@ -0,0 +1,32 @@
+{
+  "_class_name": "StableDiffusionPipeline",
+  "_diffusers_version": "0.6.0",
+  "feature_extractor": [
+    "transformers",
+    "CLIPImageProcessor"
+  ],
+  "safety_checker": [
+    "stable_diffusion",
+    "StableDiffusionSafetyChecker"
+  ],
+  "scheduler": [
+    "diffusers",
+    "PNDMScheduler"
+  ],
+  "text_encoder": [
+    "transformers",
+    "CLIPTextModel"
+  ],
+  "tokenizer": [
+    "transformers",
+    "CLIPTokenizer"
+  ],
+  "unet": [
+    "diffusers",
+    "UNet2DConditionModel"
+  ],
+  "vae": [
+    "diffusers",
+    "AutoencoderKL"
+  ]
+}

requirements.txt

@@ -0,0 +1,8 @@
+## Python version: 3.11.3
+
+torch==2.0.1
+numpy==1.25.0
+tqdm==4.65.0
+transformers==4.33.2
+lightning==2.0.9
+pillow==9.5.0

safety_checker/config.json

@@ -0,0 +1,175 @@
+{
+  "_commit_hash": "4bb648a606ef040e7685bde262611766a5fdd67b",
+  "_name_or_path": "CompVis/stable-diffusion-safety-checker",
+  "architectures": [
+    "StableDiffusionSafetyChecker"
+  ],
+  "initializer_factor": 1.0,
+  "logit_scale_init_value": 2.6592,
+  "model_type": "clip",
+  "projection_dim": 768,
+  "text_config": {
+    "_name_or_path": "",
+    "add_cross_attention": false,
+    "architectures": null,
+    "attention_dropout": 0.0,
+    "bad_words_ids": null,
+    "bos_token_id": 0,
+    "chunk_size_feed_forward": 0,
+    "cross_attention_hidden_size": null,
+    "decoder_start_token_id": null,
+    "diversity_penalty": 0.0,
+    "do_sample": false,
+    "dropout": 0.0,
+    "early_stopping": false,
+    "encoder_no_repeat_ngram_size": 0,
+    "eos_token_id": 2,
+    "exponential_decay_length_penalty": null,
+    "finetuning_task": null,
+    "forced_bos_token_id": null,
+    "forced_eos_token_id": null,
+    "hidden_act": "quick_gelu",
+    "hidden_size": 768,
+    "id2label": {
+      "0": "LABEL_0",
+      "1": "LABEL_1"
+    },
+    "initializer_factor": 1.0,
+    "initializer_range": 0.02,
+    "intermediate_size": 3072,
+    "is_decoder": false,
+    "is_encoder_decoder": false,
+    "label2id": {
+      "LABEL_0": 0,
+      "LABEL_1": 1
+    },
+    "layer_norm_eps": 1e-05,
+    "length_penalty": 1.0,
+    "max_length": 20,
+    "max_position_embeddings": 77,
+    "min_length": 0,
+    "model_type": "clip_text_model",
+    "no_repeat_ngram_size": 0,
+    "num_attention_heads": 12,
+    "num_beam_groups": 1,
+    "num_beams": 1,
+    "num_hidden_layers": 12,
+    "num_return_sequences": 1,
+    "output_attentions": false,
+    "output_hidden_states": false,
+    "output_scores": false,
+    "pad_token_id": 1,
+    "prefix": null,
+    "problem_type": null,
+    "pruned_heads": {},
+    "remove_invalid_values": false,
+    "repetition_penalty": 1.0,
+    "return_dict": true,
+    "return_dict_in_generate": false,
+    "sep_token_id": null,
+    "task_specific_params": null,
+    "temperature": 1.0,
+    "tf_legacy_loss": false,
+    "tie_encoder_decoder": false,
+    "tie_word_embeddings": true,
+    "tokenizer_class": null,
+    "top_k": 50,
+    "top_p": 1.0,
+    "torch_dtype": null,
+    "torchscript": false,
+    "transformers_version": "4.22.0.dev0",
+    "typical_p": 1.0,
+    "use_bfloat16": false,
+    "vocab_size": 49408
+  },
+  "text_config_dict": {
+    "hidden_size": 768,
+    "intermediate_size": 3072,
+    "num_attention_heads": 12,
+    "num_hidden_layers": 12
+  },
+  "torch_dtype": "float32",
+  "transformers_version": null,
+  "vision_config": {
+    "_name_or_path": "",
+    "add_cross_attention": false,
+    "architectures": null,
+    "attention_dropout": 0.0,
+    "bad_words_ids": null,
+    "bos_token_id": null,
+    "chunk_size_feed_forward": 0,
+    "cross_attention_hidden_size": null,
+    "decoder_start_token_id": null,
+    "diversity_penalty": 0.0,
+    "do_sample": false,
+    "dropout": 0.0,
+    "early_stopping": false,
+    "encoder_no_repeat_ngram_size": 0,
+    "eos_token_id": null,
+    "exponential_decay_length_penalty": null,
+    "finetuning_task": null,
+    "forced_bos_token_id": null,
+    "forced_eos_token_id": null,
+    "hidden_act": "quick_gelu",
+    "hidden_size": 1024,
+    "id2label": {
+      "0": "LABEL_0",
+      "1": "LABEL_1"
+    },
+    "image_size": 224,
+    "initializer_factor": 1.0,
+    "initializer_range": 0.02,
+    "intermediate_size": 4096,
+    "is_decoder": false,
+    "is_encoder_decoder": false,
+    "label2id": {
+      "LABEL_0": 0,
+      "LABEL_1": 1
+    },
+    "layer_norm_eps": 1e-05,
+    "length_penalty": 1.0,
+    "max_length": 20,
+    "min_length": 0,
+    "model_type": "clip_vision_model",
+    "no_repeat_ngram_size": 0,
+    "num_attention_heads": 16,
+    "num_beam_groups": 1,
+    "num_beams": 1,
+    "num_channels": 3,
+    "num_hidden_layers": 24,
+    "num_return_sequences": 1,
+    "output_attentions": false,
+    "output_hidden_states": false,
+    "output_scores": false,
+    "pad_token_id": null,
+    "patch_size": 14,
+    "prefix": null,
+    "problem_type": null,
+    "pruned_heads": {},
+    "remove_invalid_values": false,
+    "repetition_penalty": 1.0,
+    "return_dict": true,
+    "return_dict_in_generate": false,
+    "sep_token_id": null,
+    "task_specific_params": null,
+    "temperature": 1.0,
+    "tf_legacy_loss": false,
+    "tie_encoder_decoder": false,
+    "tie_word_embeddings": true,
+    "tokenizer_class": null,
+    "top_k": 50,
+    "top_p": 1.0,
+    "torch_dtype": null,
+    "torchscript": false,
+    "transformers_version": "4.22.0.dev0",
+    "typical_p": 1.0,
+    "use_bfloat16": false
+  },
+  "vision_config_dict": {
+    "hidden_size": 1024,
+    "intermediate_size": 4096,
+    "num_attention_heads": 16,
+    "num_hidden_layers": 24,
+    "patch_size": 14
+  }
+}

safety_checker/pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:193490b58ef62739077262e833bf091c66c29488058681ac25cf7df3d8190974
+size 1216061799

scheduler/scheduler_config.json

@@ -0,0 +1,13 @@
+{
+  "_class_name": "PNDMScheduler",
+  "_diffusers_version": "0.6.0",
+  "beta_end": 0.012,
+  "beta_schedule": "scaled_linear",
+  "beta_start": 0.00085,
+  "num_train_timesteps": 1000,
+  "set_alpha_to_one": false,
+  "skip_prk_steps": true,
+  "steps_offset": 1,
+  "trained_betas": null,
+  "clip_sample": false
+}

text_encoder/config.json

@@ -0,0 +1,25 @@
+{
+  "_name_or_path": "openai/clip-vit-large-patch14",
+  "architectures": [
+    "CLIPTextModel"
+  ],
+  "attention_dropout": 0.0,
+  "bos_token_id": 0,
+  "dropout": 0.0,
+  "eos_token_id": 2,
+  "hidden_act": "quick_gelu",
+  "hidden_size": 768,
+  "initializer_factor": 1.0,
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "layer_norm_eps": 1e-05,
+  "max_position_embeddings": 77,
+  "model_type": "clip_text_model",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "pad_token_id": 1,
+  "projection_dim": 768,
+  "torch_dtype": "float32",
+  "transformers_version": "4.22.0.dev0",
+  "vocab_size": 49408
+}

text_encoder/pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1bf5fb25270cfa3642bea72f6874f06a38d2475fbae1d944bbbead81b3187e1b
+size 336957440

tokenizer/special_tokens_map.json

@@ -0,0 +1,24 @@
+{
+  "bos_token": {
+    "content": "<|startoftext|>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": "<|endoftext|>",
+  "unk_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer/tokenizer_config.json

@@ -0,0 +1,34 @@
+{
+  "add_prefix_space": false,
+  "bos_token": {
+    "__type": "AddedToken",
+    "content": "<|startoftext|>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "do_lower_case": true,
+  "eos_token": {
+    "__type": "AddedToken",
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "errors": "replace",
+  "model_max_length": 77,
+  "name_or_path": "openai/clip-vit-large-patch14",
+  "pad_token": "<|endoftext|>",
+  "special_tokens_map_file": "./special_tokens_map.json",
+  "tokenizer_class": "CLIPTokenizer",
+  "unk_token": {
+    "__type": "AddedToken",
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  }
+}

unet/config.json

@@ -0,0 +1,36 @@
+{
+  "_class_name": "UNet2DConditionModel",
+  "_diffusers_version": "0.6.0",
+  "act_fn": "silu",
+  "attention_head_dim": 8,
+  "block_out_channels": [
+    320,
+    640,
+    1280,
+    1280
+  ],
+  "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": 4,
+  "layers_per_block": 2,
+  "mid_block_scale_factor": 1,
+  "norm_eps": 1e-05,
+  "norm_num_groups": 32,
+  "out_channels": 4,
+  "sample_size": 64,
+  "up_block_types": [
+    "UpBlock2D",
+    "CrossAttnUpBlock2D",
+    "CrossAttnUpBlock2D",
+    "CrossAttnUpBlock2D"
+  ]
+}

unet/diffusion_pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c7da0e21ba7ea50637bee26e81c220844defdf01aafca02b2c42ecdadb813de4
+size 3438354725

unet/diffusion_pytorch_model.fp16.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:30eb3dc47c90e4a55476332b284b2331774c530edbbb83b70cacdd9e7b91af92
+size 1719327893

unet/diffusion_pytorch_model.non_ema.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:42bc8b8f3af32866db3c7bb5bcf591ab04438296c2712246d7a640bde5a5ddc1
+size 3438366373

unet/diffusion_pytorch_model.non_ema.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cd1b6db09a81cb1d39fbd245a89c1e3db9da9fe8eba5e8f9098ea6c4994221d3
+size 3438167536

unet/diffusion_pytorch_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:19da7aaa4b880e59d56843f1fcb4dd9b599c28a1d9d9af7c1143057c8ffae9f1
+size 3438167540