Spaces:
Sleeping
Sleeping
1st commit!! :D
Browse files- app.py +136 -0
- requirements.txt +2 -0
app.py
ADDED
@@ -0,0 +1,136 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from typing import Dict, Tuple
|
2 |
+
from tqdm import tqdm
|
3 |
+
import torch
|
4 |
+
import torch.nn as nn
|
5 |
+
import torch.nn.functional as F
|
6 |
+
from torch.utils.data import DataLoader
|
7 |
+
from torchvision import models, transforms
|
8 |
+
from torchvision.utils import save_image, make_grid
|
9 |
+
import matplotlib.pyplot as plt
|
10 |
+
from matplotlib.animation import FuncAnimation, PillowWriter
|
11 |
+
import numpy as np
|
12 |
+
from IPython.display import HTML
|
13 |
+
from diffusion_utilities import *
|
14 |
+
|
15 |
+
openai.api_key = os.getenv('OPENAI_API_KEY')
|
16 |
+
|
17 |
+
class ContextUnet(nn.Module):
|
18 |
+
def __init__(self, in_channels, n_feat=256, n_cfeat=10, height=28): # cfeat - context features
|
19 |
+
super(ContextUnet, self).__init__()
|
20 |
+
|
21 |
+
# number of input channels, number of intermediate feature maps and number of classes
|
22 |
+
self.in_channels = in_channels
|
23 |
+
self.n_feat = n_feat
|
24 |
+
self.n_cfeat = n_cfeat
|
25 |
+
self.h = height #assume h == w. must be divisible by 4, so 28,24,20,16...
|
26 |
+
|
27 |
+
# Initialize the initial convolutional layer
|
28 |
+
self.init_conv = ResidualConvBlock(in_channels, n_feat, is_res=True)
|
29 |
+
|
30 |
+
# Initialize the down-sampling path of the U-Net with two levels
|
31 |
+
self.down1 = UnetDown(n_feat, n_feat) # down1 #[10, 256, 8, 8]
|
32 |
+
self.down2 = UnetDown(n_feat, 2 * n_feat) # down2 #[10, 256, 4, 4]
|
33 |
+
|
34 |
+
# original: self.to_vec = nn.Sequential(nn.AvgPool2d(7), nn.GELU())
|
35 |
+
self.to_vec = nn.Sequential(nn.AvgPool2d((4)), nn.GELU())
|
36 |
+
|
37 |
+
# Embed the timestep and context labels with a one-layer fully connected neural network
|
38 |
+
self.timeembed1 = EmbedFC(1, 2*n_feat)
|
39 |
+
self.timeembed2 = EmbedFC(1, 1*n_feat)
|
40 |
+
self.contextembed1 = EmbedFC(n_cfeat, 2*n_feat)
|
41 |
+
self.contextembed2 = EmbedFC(n_cfeat, 1*n_feat)
|
42 |
+
|
43 |
+
# Initialize the up-sampling path of the U-Net with three levels
|
44 |
+
self.up0 = nn.Sequential(
|
45 |
+
nn.ConvTranspose2d(2 * n_feat, 2 * n_feat, self.h//4, self.h//4),
|
46 |
+
nn.GroupNorm(8, 2 * n_feat), # normalize
|
47 |
+
nn.ReLU(),
|
48 |
+
)
|
49 |
+
self.up1 = UnetUp(4 * n_feat, n_feat)
|
50 |
+
self.up2 = UnetUp(2 * n_feat, n_feat)
|
51 |
+
|
52 |
+
# Initialize the final convolutional layers to map to the same number of channels as the input image
|
53 |
+
self.out = nn.Sequential(
|
54 |
+
nn.Conv2d(2 * n_feat, n_feat, 3, 1, 1), # reduce number of feature maps #in_channels, out_channels, kernel_size, stride=1, padding=0
|
55 |
+
nn.GroupNorm(8, n_feat), # normalize
|
56 |
+
nn.ReLU(),
|
57 |
+
nn.Conv2d(n_feat, self.in_channels, 3, 1, 1), # map to same number of channels as input
|
58 |
+
)
|
59 |
+
|
60 |
+
def forward(self, x, t, c=None):
|
61 |
+
"""
|
62 |
+
x : (batch, n_feat, h, w) : input image
|
63 |
+
t : (batch, n_cfeat) : time step
|
64 |
+
c : (batch, n_classes) : context label
|
65 |
+
"""
|
66 |
+
# x is the input image, c is the context label, t is the timestep, context_mask says which samples to block the context on
|
67 |
+
|
68 |
+
# pass the input image through the initial convolutional layer
|
69 |
+
x = self.init_conv(x)
|
70 |
+
# pass the result through the down-sampling path
|
71 |
+
down1 = self.down1(x) #[10, 256, 8, 8]
|
72 |
+
down2 = self.down2(down1) #[10, 256, 4, 4]
|
73 |
+
|
74 |
+
# convert the feature maps to a vector and apply an activation
|
75 |
+
hiddenvec = self.to_vec(down2)
|
76 |
+
|
77 |
+
# mask out context if context_mask == 1
|
78 |
+
if c is None:
|
79 |
+
c = torch.zeros(x.shape[0], self.n_cfeat).to(x)
|
80 |
+
|
81 |
+
# embed context and timestep
|
82 |
+
cemb1 = self.contextembed1(c).view(-1, self.n_feat * 2, 1, 1) # (batch, 2*n_feat, 1,1)
|
83 |
+
temb1 = self.timeembed1(t).view(-1, self.n_feat * 2, 1, 1)
|
84 |
+
cemb2 = self.contextembed2(c).view(-1, self.n_feat, 1, 1)
|
85 |
+
temb2 = self.timeembed2(t).view(-1, self.n_feat, 1, 1)
|
86 |
+
#print(f"uunet forward: cemb1 {cemb1.shape}. temb1 {temb1.shape}, cemb2 {cemb2.shape}. temb2 {temb2.shape}")
|
87 |
+
|
88 |
+
|
89 |
+
up1 = self.up0(hiddenvec)
|
90 |
+
up2 = self.up1(cemb1*up1 + temb1, down2) # add and multiply embeddings
|
91 |
+
up3 = self.up2(cemb2*up2 + temb2, down1)
|
92 |
+
out = self.out(torch.cat((up3, x), 1))
|
93 |
+
return out
|
94 |
+
|
95 |
+
# hyperparameters
|
96 |
+
|
97 |
+
# diffusion hyperparameters
|
98 |
+
timesteps = 500
|
99 |
+
beta1 = 1e-4
|
100 |
+
beta2 = 0.02
|
101 |
+
|
102 |
+
# network hyperparameters
|
103 |
+
device = torch.device("cuda:0" if torch.cuda.is_available() else torch.device('cpu'))
|
104 |
+
n_feat = 64 # 64 hidden dimension feature
|
105 |
+
n_cfeat = 5 # context vector is of size 5
|
106 |
+
height = 16 # 16x16 image
|
107 |
+
save_dir = './weights/'
|
108 |
+
|
109 |
+
# training hyperparameters
|
110 |
+
batch_size = 100
|
111 |
+
n_epoch = 32
|
112 |
+
lrate=1e-3
|
113 |
+
|
114 |
+
# construct DDPM noise schedule
|
115 |
+
b_t = (beta2 - beta1) * torch.linspace(0, 1, timesteps + 1, device=device) + beta1
|
116 |
+
a_t = 1 - b_t
|
117 |
+
ab_t = torch.cumsum(a_t.log(), dim=0).exp()
|
118 |
+
ab_t[0] = 1
|
119 |
+
|
120 |
+
# construct model
|
121 |
+
nn_model = ContextUnet(in_channels=3, n_feat=n_feat, n_cfeat=n_cfeat, height=height).to(device)
|
122 |
+
|
123 |
+
def greet(input):
|
124 |
+
prompt = f"""
|
125 |
+
Recommend complementary shop combinations which match well with the shop(s) described in the following text, which is delimited by triple backticks. Rank by synergy: \
|
126 |
+
Text: ```{input}```
|
127 |
+
"""
|
128 |
+
response = prompt
|
129 |
+
return response
|
130 |
+
|
131 |
+
#iface = gr.Interface(fn=greet, inputs="text", outputs="text")
|
132 |
+
#iface.launch()
|
133 |
+
|
134 |
+
#iface = gr.Interface(fn=greet, inputs=[gr.Textbox(label="Text to find entities", lines=2)], outputs=[gr.HighlightedText(label="Text with entities")], title="NER with dslim/bert-base-NER", description="Find entities using the `dslim/bert-base-NER` model under the hood!", allow_flagging="never", examples=["My name is Andrew and I live in California", "My name is Poli and work at HuggingFace"])
|
135 |
+
iface = gr.Interface(fn=greet, inputs=[gr.Textbox(label="Co-Retailing Business")], outputs="text")
|
136 |
+
iface.launch()
|
requirements.txt
ADDED
@@ -0,0 +1,2 @@
|
|
|
|
|
|
|
1 |
+
openai
|
2 |
+
python-dotenv
|