JosephusCheung's picture
Upload 121 files
8832500
raw
history blame
No virus
5.42 kB
<!DOCTYPE html>
<html lang="en">
<head>
<meta http-equiv="X-UA-Compatible" content="IE=Edge" />
<meta charset="utf-8" />
</head>
<body style="margin: 0;">
<div id="p11" style="overflow: hidden; position: relative; background-color: white; width: 2200px; height: 1237px;">
<!-- Begin shared CSS values -->
<style class="shared-css" type="text/css" >
.t {
transform-origin: bottom left;
z-index: 2;
position: absolute;
white-space: pre;
overflow: visible;
line-height: 1.5;
}
.text-container {
white-space: pre;
}
@supports (-webkit-touch-callout: none) {
.text-container {
white-space: normal;
}
}
</style>
<!-- End shared CSS values -->
<!-- Begin inline CSS -->
<style type="text/css" >
#t1_11{left:0px;bottom:1054px;letter-spacing:-0.23px;}
#t2_11{left:0px;bottom:917px;letter-spacing:-0.24px;}
#t3_11{left:0px;bottom:781px;letter-spacing:-0.24px;}
#t4_11{left:0px;bottom:645px;letter-spacing:-0.21px;}
#t5_11{left:539px;bottom:969px;letter-spacing:0.23px;}
#t6_11{left:539px;bottom:905px;letter-spacing:0.2px;}
#t7_11{left:539px;bottom:840px;letter-spacing:0.2px;}
#t8_11{left:539px;bottom:776px;letter-spacing:0.21px;}
#t9_11{left:539px;bottom:712px;letter-spacing:0.18px;}
#ta_11{left:539px;bottom:648px;letter-spacing:0.21px;}
#tb_11{left:539px;bottom:584px;letter-spacing:0.23px;}
#tc_11{left:573px;bottom:1080px;letter-spacing:0.16px;}
#td_11{left:28px;bottom:462px;letter-spacing:0.22px;}
#te_11{left:279px;bottom:462px;letter-spacing:0.22px;}
#tf_11{left:28px;bottom:398px;letter-spacing:0.21px;}
#tg_11{left:28px;bottom:334px;letter-spacing:0.21px;}
#th_11{left:28px;bottom:270px;letter-spacing:0.21px;}
#ti_11{left:28px;bottom:206px;letter-spacing:0.21px;}
#tj_11{left:28px;bottom:141px;letter-spacing:0.2px;}
#tk_11{left:28px;bottom:77px;letter-spacing:0.21px;}
#tl_11{left:28px;bottom:13px;letter-spacing:0.21px;}
#tm_11{left:729px;bottom:13px;letter-spacing:0.23px;}
#tn_11{left:1054px;bottom:13px;}
.s1_11{font-size:115px;font-family:IBMPlexSans_2t;color:#000;}
.s2_11{font-size:115px;font-family:IBMPlexSans-Bold_2s;color:#000;}
.s3_11{font-size:53px;font-family:IBMPlexSans-Italic_1q;color:#000;}
.s4_11{font-size:53px;font-family:IBMPlexSans_2t;color:#000;}
.s5_11{font-size:99px;font-family:IBMPlexSans-Bold_2s;color:#000;}
.s6_11{font-size:53px;font-family:IBMPlexSans-Bold_2s;color:#000;}
</style>
<!-- End inline CSS -->
<!-- Begin embedded font definitions -->
<style id="fonts11" type="text/css" >
@font-face {
font-family: IBMPlexSans-Bold_2s;
src: url("fonts/IBMPlexSans-Bold_2s.woff") format("woff");
}
@font-face {
font-family: IBMPlexSans-Italic_1q;
src: url("fonts/IBMPlexSans-Italic_1q.woff") format("woff");
}
@font-face {
font-family: IBMPlexSans_2t;
src: url("fonts/IBMPlexSans_2t.woff") format("woff");
}
</style>
<!-- End embedded font definitions -->
<!-- Begin page background -->
<div id="pg11Overlay" style="width:100%; height:100%; position:absolute; z-index:1; background-color:rgba(0,0,0,0); -webkit-user-select: none;"></div>
<div id="pg11" style="-webkit-user-select: none;"><object width="2200" height="1237" data="11/11.svg" type="image/svg+xml" id="pdf11" style="width:2200px; height:1237px; -moz-transform:scale(1); z-index: 0;"></object></div>
<!-- End page background -->
<!-- Begin text definitions (Positioned/styled in CSS) -->
<div class="text-container"><span id="t1_11" class="t s1_11">Problems </span>
<span id="t2_11" class="t s1_11">and </span>
<span id="t3_11" class="t s2_11">Proposed </span>
<span id="t4_11" class="t s2_11">Solutions </span>
<span id="t5_11" class="t s3_11">Some merged models have good performance, such as AnythingV3. </span>
<span id="t6_11" class="t s3_11">Should I continue to merge? </span>
<span id="t7_11" class="t s4_11">This is not scientifically sound and will ultimately result in a model </span>
<span id="t8_11" class="t s4_11">that is overfitted in some cases and normal in others. This model </span>
<span id="t9_11" class="t s4_11">looks good at first, but you will find that it is not faithful to the </span>
<span id="ta_11" class="t s4_11">input prompt and suffers from the problem of language drift </span>
<span id="tb_11" class="t s4_11">mentioned above. </span>
<span id="tc_11" class="t s5_11">3.Merged Models? </span>
<span id="td_11" class="t s4_11">Solution </span><span id="te_11" class="t s4_11">We can use the method mentioned above to train two models together using a </span>
<span id="tf_11" class="t s4_11">word frequency list with Dreambooth. We can add or replace the training data with the </span>
<span id="tg_11" class="t s4_11">images generated by the model we want to merge, according to the calculated ratio, and </span>
<span id="th_11" class="t s4_11">maintain a dynamic training dataset during training to prevent overfitting, as mentioned </span>
<span id="ti_11" class="t s4_11">above. We get a balanced model that does not overfit in certain directions. Then choose a </span>
<span id="tj_11" class="t s4_11">checkpoint that is about to be overfitted but not yet as the final version. This type of </span>
<span id="tk_11" class="t s4_11">model is popular in the community because it has good output even under poorly written </span>
<span id="tl_11" class="t s4_11">prompt inputs, such as the </span><span id="tm_11" class="t s6_11">CertainThing</span><span id="tn_11" class="t s4_11">. </span></div>
<!-- End text definitions -->
</div>
</body>
</html>