SUR-adapter

SUR-adapter copied to clipboard

非常感谢作者的贡献，你们的研究给了我很大的启发，我想试试你们的模型。但是在中国无法使用谷歌网盘下载the dataset and vectors。请问有别的途径下载吗？

leon946

Can you please share the adapter checkpoint, `checkpoints/runwayml_fp16/test_llm13B_llml39_lr1e-05_llmw1e-05_promptw1e-05_adapterw0.1/adapter_checkpoint1000.pt` The one on Google drive (https://drive.google.com/drive/folders/1UyC9_AqTezmHXmj4dh0A-9RBKKx_JmJZ) is based on "SG161222/Realistic_Vision_V2.0"

statechular11

Hi, nice work! Would you please the 15 testing prompts per each type: counting, color, action? Thanks

statechular11

Colab？

1

您好，请问会计划放出colab测试吗。想看更多的case效果如何

rainsoulsrx

model open

1

Excuse me, will the model be open source? Looking forward to try the effect.

xiaominghero

Lexica 1, Civitai 2, and Stable Diffusion Online have a large number of images. May I ask what criteria or keywords did you use to select collect 114,148 image-text pairs...

menghuaa

关于蒸馏中LLM模型的语义特征

1

首先感谢你们的文章！看了很有启发性。 我有一点困惑，对于文中提到的取llama模型的语义特征，并且你们提到了下面这两行： for layer in self.layers: h = layer(h, start_pos, freqs_cis, mask) 我使用了你们的例子输入”a colorful animal with big eyes on a blue background，“但当我打印第40层layer的最终结果shape时，显示h的shape是【1， 12， 5120】，显然每个word都拥有一个[5120]长度的token，但你们的sur_data_small里面却是一个【5120】的token，这是怎么回事呢？我应该取LLM哪个位置的语义特征呢？ 期待你们的回复

wddwzwhhxx

没有获得预期的结果

3

使用的推理代码如下，权重使用https://drive.google.com/drive/folders/1UyC9_AqTezmHXmj4dh0A-9RBKKx_JmJZ import os os.environ['CUDA_VISIBLE_DEVICES']='0' from SUR_adapter_pipeline import SURStableDiffusionPipeline import torch from SUR_adapter import Adapter adapter_path = "adapter_checkpoint.pt" adapter=Adapter().to("cuda") adapter.load_state_dict(torch.load(adapter_path)) adapter.adapter_weight = 0.1 model_path = "runwayml/stable-diffusion-v1-5" pipe = SURStableDiffusionPipeline.from_pretrained(model_path, adapter=adapter) pipe.to("cuda")...

laoniandisko