Upload 8 files

t2i/Fluently-XL-v2.py

@@ -0,0 +1,47 @@
+# https://huggingface.co/spaces/ehristoforu/dalle-3-xl-lora-v2
+# pip install diffusers transformers accelerate safetensors
+# HF_ENDPOINT=https://hf-mirror.com python Fluently-XL-v2.py
+
+import torch
+from diffusers import StableDiffusionXLPipeline, EulerAncestralDiscreteScheduler
+
+model_id = "fluently/Fluently-XL-v2"
+negative_prompt = "(deformed, distorted, disfigured:1.3), poorly drawn, bad anatomy, wrong anatomy, extra limb, missing limb, floating limbs, (mutated hands and fingers:1.4), disconnected limbs, mutation, mutated, ugly, disgusting, blurry, amputation, (NSFW:1.25)"
+width, height = 1024, 1024
+guidance_scale = 6
+
+def save_image(img):
+    import uuid
+    unique_name = str(uuid.uuid4()) + ".png"
+    img.save(unique_name)
+    return unique_name
+
+def t2i(prompt):
+        
+    pipe = StableDiffusionXLPipeline.from_pretrained(model_id, torch_dtype=torch.float16, use_safetensors=True,)
+    pipe.scheduler = EulerAncestralDiscreteScheduler.from_config(pipe.scheduler.config)
+
+    ## lora
+    # pipe.load_lora_weights("ehristoforu/dalle-3-xl-v2", weight_name="dalle-3-xl-lora-v2.safetensors", adapter_name="dalle")
+    # pipe.set_adapters("dalle")
+
+    pipe.to("cuda")
+
+    image = pipe(
+        prompt=prompt,
+        negative_prompt=negative_prompt,
+        width=width,
+        height=height,
+        guidance_scale=guidance_scale,
+        num_inference_steps=25,
+        num_images_per_prompt=1,
+        cross_attention_kwargs={"scale": 0.65},
+        output_type="pil",
+    ).images[0]
+
+    return image
+
+if __name__ == "__main__":
+    prompt = "a girl in beijing"
+    image = t2i(prompt)
+    image.save("fluently-xl-v2_output.png")

t2i/SDXL.py

@@ -0,0 +1,32 @@
+# https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0
+# pip install diffusers transformers accelerate safetensors
+# HF_ENDPOINT=https://hf-mirror.com python SDXL.py
+
+## download
+# git clone https://hf-mirror.com/stabilityai/stable-diffusion-xl-base-1.0
+# cd unet && wget -c https://hf-mirror.com/stabilityai/stable-diffusion-xl-base-1.0/resolve/main/unet/diffusion_pytorch_model.fp16.safetensors
+# cd vae && wget -c https://hf-mirror.com/stabilityai/stable-diffusion-xl-base-1.0/resolve/main/vae/diffusion_pytorch_model.fp16.safetensors
+# wget -c https://hf-mirror.com/stabilityai/stable-diffusion-xl-base-1.0/resolve/main/sd_xl_base_1.0_0.9vae.safetensors
+
+from diffusers import DiffusionPipeline
+import torch
+
+model_id = "stabilityai/stable-diffusion-xl-base-1.0"
+
+def t2i(prompt):
+    pipe = DiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float16, use_safetensors=True, variant="fp16")
+    pipe.to("cuda")
+
+    # 使用 torch >= 2.0 时，通过 torch.compile 可以将推理速度提高 20-30%。在运行管道之前，使用 torchcompile 简单地包装unet：
+    # pipe.unet = torch.compile(pipe.unet, mode="reduce-overhead", fullgraph=True)
+    
+    # if using torch < 2.0
+    # pipe.enable_xformers_memory_efficient_attention()
+
+    image = pipe(prompt=prompt).images[0]
+    return image
+
+if __name__ == "__main__":
+    prompt = "a girl in beijing"
+    image = t2i(prompt)
+    image.save("sdxl_output.png")

t2i/SDXL_Lighting.py

@@ -0,0 +1,39 @@
+## download
+# git clone https://hf-mirror.com/stabilityai/stable-diffusion-xl-base-1.0
+# wget -c hf-mirror.com/stabilityai/stable-diffusion-xl-base-1.0/resolve/main/unet/diffusion_pytorch_model.fp16.safetensors
+# wget -c https://hf-mirror.com/ByteDance/SDXL-Lightning/resolve/main/sdxl_lightning_4step_unet.safetensors
+# base = xxx/stable-diffusion-xl-base-1.0
+# ckpt = SDXL-Lightning/sdxl_lightning_4step_unet.safetensors
+
+import torch
+from diffusers import StableDiffusionXLPipeline, UNet2DConditionModel, EulerDiscreteScheduler
+from huggingface_hub import hf_hub_download
+from safetensors.torch import load_file
+
+base = "stabilityai/stable-diffusion-xl-base-1.0"
+repo = "ByteDance/SDXL-Lightning"
+ckpt = "sdxl_lightning_4step_unet.safetensors" # Use the correct ckpt for your step setting!
+
+def t2i(prompt):
+
+    # Load model.
+    
+    # unet = UNet2DConditionModel.from_config(base, subfolder="unet").to("cuda", torch.float16)
+    # unet.load_state_dict(load_file(hf_hub_download(repo, ckpt), device="cuda"))
+    # pipe = StableDiffusionXLPipeline.from_pretrained(base, unet=unet, torch_dtype=torch.float16, variant="fp16").to("cuda")
+    
+    unet = UNet2DConditionModel.from_config(base, subfolder="unet").to("cuda", torch.float16)
+    unet.load_state_dict(load_file(ckpt), device="cuda"))
+    pipe = StableDiffusionXLPipeline.from_pretrained(base, unet=unet, torch_dtype=torch.float16, variant="fp16").to("cuda")
+
+    # Ensure sampler uses "trailing" timesteps.
+    pipe.scheduler = EulerDiscreteScheduler.from_config(pipe.scheduler.config, timestep_spacing="trailing")
+
+    # Ensure using the same inference steps as the loaded model and CFG set to 0.
+    image = pipe(prompt, num_inference_steps=4, guidance_scale=0).images[0]
+    return image
+
+if __name__ == "__main__":
+    prompt = "a girl in beijing"
+    image = t2i(prompt)
+    image.save("sdxl_lighting_output.png")

t2i/StableCascade.py

@@ -0,0 +1,61 @@
+# https://hf-mirror.com/stabilityai/stable-cascade
+# https://hf-mirror.com/stabilityai/stable-cascade-prior
+
+import torch
+from diffusers import StableCascadeDecoderPipeline, StableCascadePriorPipeline, StableCascadeCombinedPipeline
+
+cas = "stabilityai/stable-cascade"
+cas_prior = "stabilityai/stable-cascade-prior"
+
+def t2i_(prompt):
+
+    prior = StableCascadePriorPipeline.from_pretrained(cas_prior, variant="bf16", torch_dtype=torch.bfloat16)
+    decoder = StableCascadeDecoderPipeline.from_pretrained(cas, variant="bf16", torch_dtype=torch.float16)
+    
+    prior.to("cuda")
+    decoder.to("cuda")
+    # prior.enable_model_cpu_offload()
+    # decoder.enable_model_cpu_offload()
+
+    prior_output = prior(
+        prompt=prompt,
+        height=1024,
+        width=1024,
+        negative_prompt="",
+        guidance_scale=4.0,
+        num_images_per_prompt=1,
+        num_inference_steps=20
+    )
+    
+    image = decoder(
+        image_embeddings=prior_output.image_embeddings.to(torch.float16),
+        prompt=prompt,
+        negative_prompt="",
+        guidance_scale=0.0,
+        output_type="pil",
+        num_inference_steps=10
+    ).images[0]
+
+    return image
+
+def t2i(prompt):
+    pipe = StableCascadeCombinedPipeline.from_pretrained(cas, variant="bf16", torch_dtype=torch.bfloat16)
+    pipe.to("cuda")
+    
+    image = pipe(
+        prompt=prompt,
+        negative_prompt="",
+        num_inference_steps=10,
+        prior_num_inference_steps=20,
+        prior_guidance_scale=3.0,
+        width=1024,
+        height=1024,
+    ).images[0]
+
+    return image
+
+if __name__ == "__main__":
+    prompt = "a girl in beijing"
+    image = t2i(prompt)
+    # image = t2i_(prompt)
+    image.save("stablecascade_output.png")

t2i/Taiyi-XL.py

@@ -0,0 +1,20 @@
+# git clone https://github.com/IDEA-CCNL/Taiyi-Diffusion-XL.git
+# cd ./Taiyi-Diffusion-XL/
+# pip install -r requirements.txt
+# HF_ENDPOINT=https://hf-mirror.com python Taiyi-XL.py
+
+from diffusers import DiffusionPipeline
+import torch
+
+model_id = "IDEA-CCNL/Taiyi-Stable-Diffusion-XL-3.5B"
+
+def t2i(prompt):
+    pipe = DiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float16, use_safetensors=True, variant="fp16")
+    pipe.to("cuda")
+    image = pipe(prompt=prompt).images[0]
+    return image
+    
+if __name__ == "__main__":
+    prompt = "一个身穿汉服的美丽的猫女，有着黄色眼睛和黑色头发，简单妆容，眼罩和摩托车"
+    image = t2i(prompt)
+    image.save("taiyi_output.png")

t2i/client.py

@@ -0,0 +1,26 @@
+import json
+import argparse
+from gradio_client import Client
+
+
+def main():
+    parser = argparse.ArgumentParser(description="x")
+    parser.add_argument('--model', '-m', type=str, default="red")
+    parser.add_argument('--prompt', '-p', type=str, default="a girl in beijing")
+    parser.add_argument('--api-url', type=str, default="http://127.0.0.1:7860/")
+    parser.add_argument('--api-name', type=str, default="/predict")
+
+    args = parser.parse_args()
+    client = Client(args.api_url)
+    output = client.predict(args.model, args.prompt, api_name=args.api_name)    
+
+    result = {
+        "model": args.model, 
+        "prompt": args.prompt, 
+        "output": output
+    }
+    
+    print(json.dumps(result, indent=2))
+
+if __name__ == "__main__":
+    main()

t2i/server.py

@@ -0,0 +1,26 @@
+# CUDA_VISIBLE_DEVICES=0 python server.py
+
+import importlib
+import gradio as gr
+
+
+def run(model_id, prompt):
+    print(f"{model_id}: {prompt}")
+    m = importlib.import_module(model_id)
+    print(m)
+    image = m.t2i(prompt)
+    return image
+
+def app():
+    model_id = gr.Textbox(label="model-id")
+    prompt = gr.Textbox(label="prompt")
+    image = gr.Image(label="output")
+    interface = gr.Interface(
+        fn=run,
+        inputs=[model_id, prompt],
+        outputs=image,
+    )
+    interface.launch()
+
+if __name__ == "__main__":
+    app()

vlm/IXC2-4KHD.py

@@ -0,0 +1,44 @@
+# https://huggingface.co/internlm/internlm-xcomposer2-4khd-7b
+# wget -c https://hf-mirror.com/internlm/internlm-xcomposer2-4khd-7b/resolve/main/pytorch_model-00001-of-00002.bin
+# wget -c https://hf-mirror.com/internlm/internlm-xcomposer2-4khd-7b/resolve/main/pytorch_model-00002-of-00002.bin
+
+
+import torch
+from transformers import AutoModel, AutoTokenizer
+
+torch.set_grad_enabled(False)
+
+# init model and tokenizer
+ckpt_path = "internlm/internlm-xcomposer2-4khd-7b"
+model = AutoModel.from_pretrained(ckpt_path, torch_dtype=torch.bfloat16, trust_remote_code=True).cuda().eval()
+tokenizer = AutoTokenizer.from_pretrained(ckpt_path, trust_remote_code=True)
+
+###############
+# First Round
+###############
+
+query1 = '<ImageHere>Illustrate the fine details present in the image'
+image = './example.webp'
+with torch.cuda.amp.autocast():
+  response, his = model.chat(tokenizer, query=query, image=image, hd_num=55, history=[], do_sample=False, num_beams=3)
+print(response)
+# The image is a vibrant and colorful infographic that showcases 7 graphic design trends that will dominate in 2021. The infographic is divided into 7 sections, each representing a different trend. 
+# Starting from the top, the first section focuses on "Muted Color Palettes", highlighting the use of muted colors in design.
+# The second section delves into "Simple Data Visualizations", emphasizing the importance of easy-to-understand data visualizations. 
+# The third section introduces "Geometric Shapes Everywhere", showcasing the use of geometric shapes in design. 
+# The fourth section discusses "Flat Icons and Illustrations", explaining how flat icons and illustrations are being used in design. 
+# The fifth section is dedicated to "Classic Serif Fonts", illustrating the resurgence of classic serif fonts in design.
+# The sixth section explores "Social Media Slide Decks", illustrating how slide decks are being used on social media. 
+# Finally, the seventh section focuses on "Text Heavy Videos", illustrating the trend of using text-heavy videos in design. 
+# Each section is filled with relevant images and text, providing a comprehensive overview of the 7 graphic design trends that will dominate in 2021.
+
+###############
+# Second Round
+###############
+query1 = 'what is the detailed explanation of the third part.'
+with torch.cuda.amp.autocast():
+  response, _ = model.chat(tokenizer, query=query1, image=image, hd_num=55, history=his, do_sample=False, num_beams=3)
+print(response)
+# The third part of the infographic is about "Geometric Shapes Everywhere". It explains that last year, designers used a lot of
+# flowing and abstract shapes in their designs. However, this year, they have been replaced with rigid, hard-edged geometric
+# shapes and patterns. The hard edges of a geometric shape create a great contrast against muted colors.