Datasets:
sayakpaul
/
diffusers-qa-chatbot-artifacts

Dataset card Data Studio Files
xet
 Community
1
text
stringlengths
0
5.54k
pipe = StableDiffusionXLPipeline.from_pretrained(
"stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.bfloat16
).to("cuda")
# Run the attention ops without efficiency.
pipe.unet.set_default_attn_processor()
pipe.vae.set_default_attn_processor()
prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
image = pipe(prompt, num_inference_steps=30).images[0] bfloat16 reduces the latency from 7.36 seconds to 4.63 seconds: Why bfloat16? Using a reduced numerical precision (such as float16, bfloat16) to run inference doesn’t affect the generation quality but significantly improves latency. The benefits of using the bfloa...
import torch
pipe = StableDiffusionXLPipeline.from_pretrained(
"stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.bfloat16
).to("cuda")
prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
image = pipe(prompt, num_inference_steps=30).images[0] scaled_dot_product_attention improves the latency from 4.63 seconds to 3.31 seconds. Use faster kernels with torch.compile Compile the UNet and the VAE to benefit from the faster kernels. First, configure a few compiler flags: Copied from diffusers import Stabl...
import torch
torch._inductor.config.conv_1x1_as_mm = True
torch._inductor.config.coordinate_descent_tuning = True
torch._inductor.config.epilogue_fusion = False
torch._inductor.config.coordinate_descent_check_all_directions = True For the full list of compiler flags, refer to this file. It is also important to change the memory layout of the UNet and the VAE to “channels_last” when compiling them. This ensures maximum speed: Copied pipe.unet.to(memory_format=torch.channels_l...
pipe.vae.to(memory_format=torch.channels_last) Then, compile and perform inference: Copied # Compile the UNet and VAE.
pipe.unet = torch.compile(pipe.unet, mode="max-autotune", fullgraph=True)
pipe.vae.decode = torch.compile(pipe.vae.decode, mode="max-autotune", fullgraph=True)
prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
# First call to `pipe` will be slow, subsequent ones will be faster.
image = pipe(prompt, num_inference_steps=30).images[0] torch.compile offers different backends and modes. As we’re aiming for maximum inference speed, we opt for the inductor backend using the “max-autotune”. “max-autotune” uses CUDA graphs and optimizes the compilation graph specifically for latency. Specifying fullgr...
import torch
# Notice the two new flags at the end.
torch._inductor.config.conv_1x1_as_mm = True
torch._inductor.config.coordinate_descent_tuning = True
torch._inductor.config.epilogue_fusion = False
torch._inductor.config.coordinate_descent_check_all_directions = True
torch._inductor.config.force_fuse_int_mm_with_mul = True
torch._inductor.config.use_mixed_mm = True Define the filtering functions: Copied def dynamic_quant_filter_fn(mod, *args):
return (
isinstance(mod, torch.nn.Linear)
and mod.in_features > 16
and (mod.in_features, mod.out_features)
not in [
(1280, 640),
(1920, 1280),
(1920, 640),
(2048, 1280),
(2048, 2560),
(2560, 1280),
(256, 128),
(2816, 1280),
(320, 640),
(512, 1536),
(512, 256),
(512, 512),
(640, 1280),
(640, 1920),
(640, 320),
(640, 5120),
(640, 640),
(960, 320),
(960, 640),
]
)
def conv_filter_fn(mod, *args):
return (
isinstance(mod, torch.nn.Conv2d) and mod.kernel_size == (1, 1) and 128 in [mod.in_channels, mod.out_channels]
) Then apply all the optimizations discussed so far: Copied # SDPA + bfloat16.
pipe = StableDiffusionXLPipeline.from_pretrained(
"stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.bfloat16
).to("cuda")
# Combine attention projection matrices.
pipe.fuse_qkv_projections()
# Change the memory layout.
pipe.unet.to(memory_format=torch.channels_last)
pipe.vae.to(memory_format=torch.channels_last) Since this quantization support is limited to linear layers only, we also turn suitable pointwise convolution layers into linear layers to maximize the benefit. Copied from torchao import swap_conv2d_1x1_to_linear
swap_conv2d_1x1_to_linear(pipe.unet, conv_filter_fn)
swap_conv2d_1x1_to_linear(pipe.vae, conv_filter_fn) Apply dynamic quantization: Copied from torchao import apply_dynamic_quant
apply_dynamic_quant(pipe.unet, dynamic_quant_filter_fn)
apply_dynamic_quant(pipe.vae, dynamic_quant_filter_fn) Finally, compile and perform inference: Copied pipe.unet = torch.compile(pipe.unet, mode="max-autotune", fullgraph=True)
pipe.vae.decode = torch.compile(pipe.vae.decode, mode="max-autotune", fullgraph=True)
prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
image = pipe(prompt, num_inference_steps=30).images[0] Applying dynamic quantization improves the latency from 2.52 seconds to 2.43 seconds.
Custom Diffusion Custom Diffusion is a training technique for personalizing image generation models. Like Textual Inversion, DreamBooth, and LoRA, Custom Diffusion only requires a few (~4-5) example images. This technique works by only training weights in the cross-attention layers, and it uses a special word to repres...
cd diffusers
pip install . Navigate to the example folder with the training script and install the required dependencies: Copied cd examples/custom_diffusion
pip install -r requirements.txt
pip install clip-retrieval 🤗 Accelerate is a library for helping you train on multiple GPUs/TPUs or with mixed-precision. It’ll automatically configure your training setup based on your hardware and environment. Take a look at the 🤗 Accelerate Quick tour to learn more. Initialize an 🤗 Accelerate environment: Copie...
write_basic_config() Lastly, if you want to train a model on your own dataset, take a look at the Create a dataset for training guide to learn how to create a dataset that works with the training script. The following sections highlight parts of the training script that are important for understanding how to modify it,...