[U] Update channel content
This commit is contained in:
@@ -2,7 +2,7 @@
|
||||
<feed xmlns="http://www.w3.org/2005/Atom" xml:lang="zh-cn">
|
||||
<id>https://aza.moe/life</id>
|
||||
<title>小桂桂的回忆录 📒</title>
|
||||
<updated>2024-04-12T22:09:17.088929+00:00</updated>
|
||||
<updated>2024-04-13T00:09:58.813952+00:00</updated>
|
||||
<link href="https://aza.moe/life" rel="alternate"/>
|
||||
<generator uri="https://lkiesow.github.io/python-feedgen" version="0.9.0">python-feedgen</generator>
|
||||
<logo>https://aza.moe/meru_256px.png</logo>
|
||||
|
||||
File diff suppressed because one or more lines are too long
@@ -48826,7 +48826,7 @@
|
||||
"id": 3575,
|
||||
"date": "2024-04-12T13:46:41",
|
||||
"text": "刚刚起床看到了 Fabrice Bellard 新发布的 <a href=\"https://bellard.org/tsac\">TSAC</a> 音频编码,是一个用二进制语言重写、为了实际应用优化过的 <a href=\"https://arxiv.org/abs/2306.06546\">DAC</a> neural codec,真的没想到去年六月刚 preprint 的论文今年就能见到实际应用了!\n\n我去年的研究里还用到了 DAC 来着,当时这个东西几乎是所有向量模型里最重的一个,即是双通道、每帧尺寸和帧频率也都很大... 导致这一个模型存下来的矢量 npy 几乎和所有其他模型加起来一样大了(还存了一张加载矢量的时候非常壮观的 btop)\n\n不过看上去 TSAC 解决了很多它的问题,转 8bit 把模型权重减轻了不少,也不需要依赖复杂的 python 环境运行了。好期待哪天 Bellard 能把它整合进 ffmpeg qwq",
|
||||
"views": 29,
|
||||
"views": 30,
|
||||
"forwards": 0,
|
||||
"media_group_id": 13703436811200701,
|
||||
"images": [
|
||||
|
||||
@@ -12,7 +12,7 @@
|
||||
<link>https://aza.moe/life</link>
|
||||
</image>
|
||||
<language>zh-cn</language>
|
||||
<lastBuildDate>Fri, 12 Apr 2024 22:09:17 +0000</lastBuildDate>
|
||||
<lastBuildDate>Sat, 13 Apr 2024 00:09:58 +0000</lastBuildDate>
|
||||
<item>
|
||||
<title>小桂桂的回忆录 📒 #3575</title>
|
||||
<link>https://aza.moe/life?post=3575</link>
|
||||
|
||||
Reference in New Issue
Block a user