Program.cs

using System;
using System.Collections.Generic;
using System.IO;
using System.Linq;
using TorchSharp;

torchvision.io.DefaultImager = new torchvision.io.SkiaImager();
var device = TorchSharp.torch.device("cuda:0");
var clipEncoder = new ClipEncoder("clip_encoder.ckpt", device);
var start_token = 49406;
var end_token = 49407;
var dictionary = new Dictionary<string, long>(){
    {"cat", 2368},
    {"a", 320},
    {"cute", 2242},
    {"blue", 1746},
    {"wild", 3220},
    {"green", 1901},
};

var batch = 1;

var prompt = "a wild cute green cat";
var tokens = prompt.Split(' ').Select(x => dictionary[x]).ToList();
tokens = tokens.Prepend(start_token).ToList();
tokens = tokens.Append(end_token).ToList();
tokens = tokens.Concat(Enumerable.Repeat<long>(0, 77 - tokens.Count)).ToList();
var uncontional_tokens = new[]{start_token, end_token}.Concat(Enumerable.Repeat(0, 75)).ToList();
var tokenTensor = torch.tensor(tokens.ToArray(), dtype: torch.ScalarType.Int64, device: device);
tokenTensor = tokenTensor.repeat(batch, 1);
var unconditional_tokenTensor = torch.tensor(uncontional_tokens.ToArray(), dtype: torch.ScalarType.Int64, device: device);
unconditional_tokenTensor = unconditional_tokenTensor.repeat(batch, 1);
var img = torch.randn(batch, 4, 64, 64, dtype: torch.ScalarType.Float32, device: device);
var t = torch.full(new[]{batch, 1L}, value: batch, dtype: torch.ScalarType.Int32, device: device);
var condition = clipEncoder.Forward(tokenTensor);
var unconditional_condition = clipEncoder.Forward(unconditional_tokenTensor);

clipEncoder.Dispose();
var ddpm = new DDPM("ddim_v_sampler.ckpt", device);
var ddimSampler = new DDIMSampler(ddpm);
var ddim_steps = 50;
img = ddimSampler.Sample(img, condition, unconditional_condition, ddim_steps);
ddpm.Dispose();

var autoencoderKL = new AutoencoderKL("autoencoder_kl.ckpt", device);
var decoded_images = (torch.Tensor)autoencoderKL.Forward(img);
decoded_images = torch.clamp((decoded_images + 1.0) / 2.0, 0.0, 1.0);


for(int i = 0; i!= batch; ++i)
{
    var image = decoded_images[i];
    image = (image * 255.0).to(torch.ScalarType.Byte).cpu();
    torchvision.io.write_image(image, $"{i}.png", torchvision.ImageFormat.Png);
}