<?xml version="1.0" encoding="UTF-8"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>DeepSpeed — Lab Ledger Desk</title><link>https://labledgerdesk.pages.dev/lab/deepspeed/</link><description>Open-source DeepSpeed releases filed by Lab Ledger Desk.</description><atom:link href="https://labledgerdesk.pages.dev/lab/deepspeed/rss.xml" rel="self" type="application/rss+xml"/><item><title>DeepSpeed v0.20.0</title><link>https://labledgerdesk.pages.dev/b/2026/11/12/deepspeed-v0-20-0/</link><guid>https://labledgerdesk.pages.dev/b/2026/11/12/deepspeed-v0-20-0/</guid><category>Note</category><description>What's Changed Sum GPTBigCode's shared key/value gradient across tensor-parallel ranks by @Achyuthan-S in #8928 [AffineIR] Storage maps of AutoEP experts under EDP…</description><pubDate>Thu, 12 Nov 2026 23:02:58 GMT</pubDate></item><item><title>nightly-last-green: Fix unit tests failure on CPU with fp16 support (#9086)</title><link>https://labledgerdesk.pages.dev/b/2026/11/12/nightly-last-green-checkpoint-batch-replica-selection-for-affine/</link><guid>https://labledgerdesk.pages.dev/b/2026/11/12/nightly-last-green-checkpoint-batch-replica-selection-for-affine/</guid><category>Note</category><description>On runners whose CPU supports oneDNN fp16 (Sapphire Rapids or newer), fp16 tests that normally skip start running, and Muon's Newton-Schulz switches to fp16. …</description><pubDate>Thu, 12 Nov 2026 07:10:31 GMT</pubDate></item><item><title>v0.19.9 Patch Release</title><link>https://labledgerdesk.pages.dev/b/2026/10/29/v0-19-9-patch-release/</link><guid>https://labledgerdesk.pages.dev/b/2026/10/29/v0-19-9-patch-release/</guid><category>Launch</category><description>What's Changed Update version post 0.19.8 release by @loadams in #8871 Update to filter out version.txt from PR gates. …</description><pubDate>Thu, 29 Oct 2026 23:42:23 GMT</pubDate></item><item><title>nightly-cpu-multi-last-green: Fix multiple issues in Op Builder and refactor implementation (#9085)</title><link>https://labledgerdesk.pages.dev/b/2026/10/29/nightly-cpu-multi-last-green-add-a-v1-multi-rank-cpu-suite-manual/</link><guid>https://labledgerdesk.pages.dev/b/2026/10/29/nightly-cpu-multi-last-green-add-a-v1-multi-rank-cpu-suite-manual/</guid><category>Launch</category><description>Problems fixed Pre-built ops lose GPU architectures. Ops that filter their targets (e.g. …</description><pubDate>Thu, 29 Oct 2026 00:12:13 GMT</pubDate></item><item><title>v0.19.8 Patch Release</title><link>https://labledgerdesk.pages.dev/b/2026/10/20/v0-19-8-patch-release/</link><guid>https://labledgerdesk.pages.dev/b/2026/10/20/v0-19-8-patch-release/</guid><category>Launch</category><description>What's Changed Deprecate leftovers of compression library by @hwchen2017 in #8535 Update version.txt after 0.19.7 release by @loadams in #8555 Deprecate loco zero++…</description><pubDate>Tue, 20 Oct 2026 15:21:47 GMT</pubDate></item><item><title>nightly-bisect/c</title><link>https://labledgerdesk.pages.dev/b/2026/10/11/nightly-bisect-c/</link><guid>https://labledgerdesk.pages.dev/b/2026/10/11/nightly-bisect-c/</guid><category>Launch</category><description>DeepSpeed published release nightly-bisect/c on its official releases feed. …</description><pubDate>Sun, 11 Oct 2026 23:58:32 GMT</pubDate></item><item><title>nightly-bisect/b: [AutoSP] Honor seq_dim when AutoSP shards inputs (#8720)</title><link>https://labledgerdesk.pages.dev/b/2026/10/11/nightly-bisect-b-autosp-honor-seq-dim-when-autosp-shards-inputs-8720/</link><guid>https://labledgerdesk.pages.dev/b/2026/10/11/nightly-bisect-b-autosp-honor-seq-dim-when-autosp-shards-inputs-8720/</guid><category>Note</category><description>Split from #8457 . Problem prepare_autosp_inputs() takes a seq_dim argument, but only passes it to mark_dynamic . …</description><pubDate>Sun, 11 Oct 2026 22:46:09 GMT</pubDate></item><item><title>nightly-bisect/a: [Bug Fix] Fix CPUAdagrad updates with nonzero weight decay (#8781)</title><link>https://labledgerdesk.pages.dev/b/2026/10/11/nightly-bisect-a-bug-fix-fix-cpuadagrad-updates-with-nonzero-weight/</link><guid>https://labledgerdesk.pages.dev/b/2026/10/11/nightly-bisect-a-bug-fix-fix-cpuadagrad-updates-with-nonzero-weight/</guid><category>Note</category><description>Fix CPUAdagrad to use the weight-decay-adjusted gradient in both the accumulator and parameter update, matching torch.optim.Adagrad . …</description><pubDate>Sun, 11 Oct 2026 15:21:59 GMT</pubDate></item></channel></rss>