<?xml version="1.0" encoding="UTF-8"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Yestino - The Signal · FSDP2 Separate</title><link>https://yestino.com/entities/fsdp2-separate-dc5ff0</link><description>Every event involving FSDP2 Separate</description><language>en</language><atom:link href="https://yestino.com/entities/fsdp2-separate-dc5ff0/feed.xml" rel="self" type="application/rss+xml"/><item><title>trunk/38d3f6fd0d6e7ea5b68ee3854e1033ff08a5c92c: [FSDP2] Separate sharded and unsharded gradient dtypes (#194434)</title><link>https://yestino.com/events/trunk-38d3f6fd0d6e7ea5b68ee3854e1033ff08a5c92c-fsdp2-separat-1ab337</link><guid isPermaLink="true">https://yestino.com/events/trunk-38d3f6fd0d6e7ea5b68ee3854e1033ff08a5c92c-fsdp2-separat-1ab337</guid><pubDate>Fri, 02 Oct 2026 00:31:17 GMT</pubDate><description>Fixes #170648 feature 0 : when mp_policy.reduce_dtype is None, it follows orig model.parameters() .grad_dtype.
Sources: PyTorch Releases</description></item></channel></rss>