Files
AMD-SHARK-Studio/shark/examples/shark_inference/stable_diffusion/main.py
Gaurav Shukla dfd6ba67b3 [SD] Update SD CLI to use model_db.json
Signed-Off-by: Gaurav Shukla <gaurav@nod-labs.com>
2022-12-22 02:13:04 +05:30

225 lines
7.1 KiB
Python

import os
os.environ["AMD_ENABLE_LLPC"] = "1"
from transformers import CLIPTextModel, CLIPTokenizer
import torch
from PIL import Image
import torchvision.transforms as T
from diffusers import (
LMSDiscreteScheduler,
PNDMScheduler,
DDIMScheduler,
DPMSolverMultistepScheduler,
EulerDiscreteScheduler,
)
from tqdm.auto import tqdm
import numpy as np
from random import randint
from stable_args import args
from utils import set_iree_runtime_flags, set_init_device_flags
from opt_params import get_unet, get_vae, get_clip
from schedulers import (
SharkEulerDiscreteScheduler,
)
import time
import sys
from shark.iree_utils.compile_utils import dump_isas
# Helper function to profile the vulkan device.
def start_profiling(file_path="foo.rdc", profiling_mode="queue"):
if args.vulkan_debug_utils and "vulkan" in args.device:
import iree
print(f"Profiling and saving to {file_path}.")
vulkan_device = iree.runtime.get_device(args.device)
vulkan_device.begin_profiling(mode=profiling_mode, file_path=file_path)
return vulkan_device
return None
def end_profiling(device):
if device:
return device.end_profiling()
if __name__ == "__main__":
dtype = torch.float32 if args.precision == "fp32" else torch.half
prompt = args.prompts
neg_prompt = args.negative_prompts
height = 512 # default height of Stable Diffusion
width = 512 # default width of Stable Diffusion
if args.version == "v2_1":
height = 768
width = 768
num_inference_steps = args.steps # Number of denoising steps
# Scale for classifier-free guidance
guidance_scale = torch.tensor(args.guidance_scale).to(torch.float32)
# Handle out of range seeds.
uint32_info = np.iinfo(np.uint32)
uint32_min, uint32_max = uint32_info.min, uint32_info.max
seed = args.seed
if seed < uint32_min or seed >= uint32_max:
seed = randint(uint32_min, uint32_max)
generator = torch.manual_seed(
seed
) # Seed generator to create the inital latent noise
# TODO: Add support for batch_size > 1.
batch_size = len(prompt)
if batch_size != 1:
sys.exit("More than one prompt is not supported yet.")
if batch_size != len(neg_prompt):
sys.exit("prompts and negative prompts must be of same length")
set_iree_runtime_flags()
set_init_device_flags()
clip = get_clip()
unet = get_unet()
vae = get_vae()
if args.dump_isa:
dump_isas(args.dispatch_benchmarks_dir)
tokenizer = CLIPTokenizer.from_pretrained("openai/clip-vit-large-patch14")
scheduler = DPMSolverMultistepScheduler.from_pretrained(
"CompVis/stable-diffusion-v1-4",
subfolder="scheduler",
)
cpu_scheduling = True
if args.version == "v2_1":
tokenizer = CLIPTokenizer.from_pretrained(
"stabilityai/stable-diffusion-2-1", subfolder="tokenizer"
)
scheduler = DPMSolverMultistepScheduler.from_pretrained(
"stabilityai/stable-diffusion-2-1",
subfolder="scheduler",
)
if args.version == "v2_1base":
tokenizer = CLIPTokenizer.from_pretrained(
"stabilityai/stable-diffusion-2-1-base", subfolder="tokenizer"
)
if args.use_compiled_scheduler:
scheduler = SharkEulerDiscreteScheduler.from_pretrained(
"stabilityai/stable-diffusion-2-1-base",
subfolder="scheduler",
)
scheduler.compile()
cpu_scheduling = False
else:
scheduler = EulerDiscreteScheduler.from_pretrained(
"stabilityai/stable-diffusion-2-1-base",
subfolder="scheduler",
)
# create a random initial latent.
latents = torch.randn(
(batch_size, 4, height // 8, width // 8),
generator=generator,
dtype=torch.float32,
).to(dtype)
# Warmup phase to improve performance.
if args.warmup_count >= 1:
vae_warmup_input = torch.clone(latents).detach().numpy()
clip_warmup_input = torch.randint(1, 2, (2, args.max_length))
for i in range(args.warmup_count):
vae.forward((vae_warmup_input,))
clip.forward((clip_warmup_input,))
start = time.time()
text_input = tokenizer(
prompt,
padding="max_length",
max_length=args.max_length,
truncation=True,
return_tensors="pt",
)
max_length = text_input.input_ids.shape[-1]
uncond_input = tokenizer(
neg_prompt,
padding="max_length",
max_length=max_length,
truncation=True,
return_tensors="pt",
)
text_input = torch.cat([uncond_input.input_ids, text_input.input_ids])
clip_inf_start = time.time()
text_embeddings = clip.forward((text_input,))
clip_inf_end = time.time()
text_embeddings = torch.from_numpy(text_embeddings).to(dtype)
text_embeddings_numpy = text_embeddings.detach().numpy()
scheduler.set_timesteps(num_inference_steps)
scheduler.is_scale_input_called = True
latents = latents * scheduler.init_noise_sigma
avg_ms = 0
for i, t in tqdm(enumerate(scheduler.timesteps), disable=args.hide_steps):
step_start = time.time()
if not args.hide_steps:
print(f"i = {i} t = {t}", end="")
timestep = torch.tensor([t]).to(dtype).detach().numpy()
latent_model_input = scheduler.scale_model_input(latents, t)
if cpu_scheduling:
latent_model_input = latent_model_input.detach().numpy()
profile_device = start_profiling(file_path="unet.rdc")
noise_pred = unet.forward(
(
latent_model_input,
timestep,
text_embeddings_numpy,
guidance_scale,
),
send_to_host=False,
)
end_profiling(profile_device)
if cpu_scheduling:
noise_pred = torch.from_numpy(noise_pred.to_host())
latents = scheduler.step(noise_pred, t, latents).prev_sample
else:
latents = scheduler.step(noise_pred, t, latents)
step_time = time.time() - step_start
avg_ms += step_time
step_ms = int((step_time) * 1000)
if not args.hide_steps:
print(f" ({step_ms}ms)")
latents_numpy = latents
if cpu_scheduling:
latents_numpy = latents.detach().numpy()
profile_device = start_profiling(file_path="vae.rdc")
vae_start = time.time()
images = vae.forward((latents_numpy,))
vae_end = time.time()
end_profiling(profile_device)
total_end = time.time()
avg_ms = 1000 * avg_ms / args.steps
clip_inf_time = (clip_inf_end - clip_inf_start) * 1000
vae_inf_time = (vae_end - vae_start) * 1000
print(f"Average step time: {avg_ms}ms/it")
print(f"Clip Inference time (ms) = {clip_inf_time:.3f}")
print(f"VAE Inference time (ms): {vae_inf_time:.3f}")
print(f"Total image generation runtime (s): {total_end - start:.4f}")
transform = T.ToPILImage()
pil_images = [
transform(image) for image in torch.from_numpy(images).to(torch.uint8)
]
for i in range(batch_size):
pil_images[i].save(f"{args.prompts[i]}_{i}.jpg")