【深度学习】PixArt-Sigma 实战【3】速度测试

css 复制代码
import time

import torch
from diffusers import Transformer2DModel, PixArtSigmaPipeline
from diffusers import ConsistencyDecoderVAE

device = torch.device("cuda:1" if torch.cuda.is_available() else "cpu")
weight_dtype = torch.float16

pipe = PixArtSigmaPipeline.from_pretrained(
    "./PixArt-Sigma-XL-2-1024-MS",
    torch_dtype=weight_dtype,
    use_safetensors=True,
)
pipe.to(device)

# transformer = Transformer2DModel.from_pretrained(
#     # "PixArt-alpha/PixArt-Sigma-XL-2-1024-MS",
#     # "/ssd/xiedong/PixArt/PixArt-Sigma-XL-2-2K-MS",
#     "/ssd/xiedong/PixArt/PixArt-Sigma-XL-2-2K-MS",
#     subfolder='transformer',
#     torch_dtype=weight_dtype,
# )
# pipe = PixArtSigmaPipeline.from_pretrained(
#     # "PixArt-alpha/pixart_sigma_sdxlvae_T5_diffusers",
#     "/ssd/xiedong/PixArt/PixArt-sigma/output/pixart_sigma_sdxlvae_T5_diffusers",
#     transformer=transformer,
#     torch_dtype=weight_dtype,
#     use_safetensors=True,
# )
# pipe.vae = ConsistencyDecoderVAE.from_pretrained("/ssd/xiedong/PixArt/consistency-decoder", torch_dtype=torch.float16)
# pipe.to(device)

# Enable memory optimizations.
# pipe.enable_model_cpu_offload()

time1 = time.time()
prompt = "A small cactus with a happy face in the Sahara desert."
image = pipe(prompt).images[0]
time2 = time.time()
print(f"time use:{time2 - time1}")
image.save("./catcus.png")

time1 = time.time()
prompt = "A small cactus with a happy face in the Sahara desert."
image = pipe(prompt).images[0]
time2 = time.time()
print(f"time use:{time2 - time1}")
image.save("./catcus.png")

A100速度 20轮4.4秒。

Loading pipeline components...: 0%| | 0/5 [00:00<?, ?it/s]You are using the default legacy behaviour of the <class 'transformers.models.t5.tokenization_t5.T5Tokenizer'>. This is expected, and simply means that the legacy (previous) behavior will be used so nothing changes for you. If you want to use the new behaviour, set legacy=False. This should only be set if you understand what it means, and thouroughly read the reason why this was added as explained in https://github.com/huggingface/transformers/pull/24565

Loading pipeline components...: 60%|██████ | 3/5 [00:01<00:01, 1.65it/s]

Loading checkpoint shards: 0%| | 0/2 [00:00<?, ?it/s]

Loading checkpoint shards: 50%|█████ | 1/2 [00:01<00:01, 1.83s/it]

Loading checkpoint shards: 100%|██████████| 2/2 [00:03<00:00, 1.70s/it]

Loading pipeline components...: 100%|██████████| 5/5 [00:11<00:00, 2.29s/it]

100%|██████████| 20/20 [00:05<00:00, 3.89it/s]

time use:6.027105093002319

100%|██████████| 20/20 [00:04<00:00, 4.94it/s]

time use:4.406545162200928

相关推荐
CV-杨帆19 分钟前
论文阅读:arxiv 2025 Scaling Laws for Differentially Private Language Models
论文阅读·人工智能·语言模型
羊羊小栈22 分钟前
基于「多模态大模型 + BGE向量检索增强RAG」的航空维修智能问答系统(vue+flask+AI算法)
vue.js·人工智能·python·语言模型·flask·毕业设计
viperrrrrrrrrr723 分钟前
GPT系列模型-详解
人工智能·gpt·llm
算家计算1 小时前
Wan2.2-Animate-14B 使用指南:从图片到动画的完整教程
人工智能·开源·aigc
西柚小萌新1 小时前
【深入浅出PyTorch】--4.PyTorch基础实战
人工智能·pytorch·python
渡我白衣1 小时前
深度学习入门(一)——从神经元到损失函数,一步步理解前向传播(下)
人工智能·深度学习·神经网络
算家计算1 小时前
快手新模型登顶开源编程模型榜首!超越Qwen3-Coder等模型
人工智能·开源·资讯
ManageEngineITSM2 小时前
IT 服务自动化的时代:让效率与体验共进
运维·数据库·人工智能·自动化·itsm·工单系统
总有刁民想爱朕ha2 小时前
AI大模型学习(17)python-flask AI大模型和图片处理工具的从一张图到多平台适配的简单方法
人工智能·python·学习·电商图片处理
302AI2 小时前
体验升级而非颠覆,API成本直降75%:DeepSeek-V3.2-Exp评测
人工智能·llm·deepseek