<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
<channel>
<title>vllm-ascend: releases and problems | Archestack</title>
<link>https://archestack.com/best-of-ai/projects/vllm-ascend/</link>
<description>New releases (breaking changes and security fixes flagged), newly reported problems and fixed issues for vllm-ascend, from its public GitHub data.</description>
<language>en</language>
<lastBuildDate>Fri, 30 Oct 2026 00:00:00 GMT</lastBuildDate>
<atom:link href="https://archestack.com/best-of-ai/projects/vllm-ascend/feed.xml" rel="self" type="application/rss+xml"/>
<item>
<title>Fixed: [Bug]: ModelSlim MoE weight-loader wrapper retains discarded layers through reload metadata</title>
<link>https://github.com/vllm-project/vllm-ascend/issues/18679</link>
<guid isPermaLink="false">archestack-watch-fixed-best-of-ai/vllm-ascend-18679</guid>
<pubDate>Fri, 30 Oct 2026 00:00:00 GMT</pubDate>
<description>&lt;p&gt;&lt;a href=&quot;https://github.com/vllm-project/vllm-ascend/issues/18679&quot;&gt;[Bug]: ModelSlim MoE weight-loader wrapper retains discarded layers through reload metadata&lt;/a&gt; closed by &lt;a href=&quot;https://github.com/vllm-project/vllm-ascend/pull/17344&quot;&gt;[BugFix][Quantization] Release ModelSlim MoE reload metadata references&lt;/a&gt;.&lt;/p&gt;</description>
</item>
<item>
<title>vllm-ascend v0.30.0 (security fix)</title>
<link>https://github.com/vllm-project/vllm-ascend/releases/tag/v0.30.0</link>
<guid isPermaLink="false">archestack-watch-release-best-of-ai/vllm-ascend-v0.30.0</guid>
<pubDate>Thu, 29 Oct 2026 00:00:00 GMT</pubDate>
<description>&lt;p&gt;vllm-ascend v0.30.0 was released on 2026-10-29. Its notes flag a security fix.&lt;/p&gt;&lt;p&gt;CVE-2026-114053 “There is a known security issue CVE-2026-114053 about vLLM upstream between v0.7.3 and v0.31.0, and vLLM has resolved this issue in main branch by vllm-project/vllm#58830. Please pay attention to whe…”&lt;/p&gt;</description>
</item>
<item>
<title>New problem reported: [Bug]:  CPU binding silently skipped on Ascend 910B3: physical NPU ID used as key but lookup uses logic ID</title>
<link>https://github.com/vllm-project/vllm-ascend/issues/20049</link>
<guid isPermaLink="false">archestack-watch-problem-best-of-ai/vllm-ascend-20049</guid>
<pubDate>Wed, 28 Oct 2026 00:00:00 GMT</pubDate>
<description>&lt;p&gt;Open on GitHub, 1 reaction: &lt;a href=&quot;https://github.com/vllm-project/vllm-ascend/issues/20049&quot;&gt;[Bug]:  CPU binding silently skipped on Ascend 910B3: physical NPU ID used as key but lookup uses logic ID&lt;/a&gt;&lt;/p&gt;</description>
</item>
<item>
<title>Fixed: [Bug]: Benchmark report classification depends on parent directory names</title>
<link>https://github.com/vllm-project/vllm-ascend/issues/18237</link>
<guid isPermaLink="false">archestack-watch-fixed-best-of-ai/vllm-ascend-18237</guid>
<pubDate>Mon, 19 Oct 2026 00:00:00 GMT</pubDate>
<description>&lt;p&gt;&lt;a href=&quot;https://github.com/vllm-project/vllm-ascend/issues/18237&quot;&gt;[Bug]: Benchmark report classification depends on parent directory names&lt;/a&gt; closed by &lt;a href=&quot;https://github.com/vllm-project/vllm-ascend/pull/19200&quot;&gt;[Misc] Remove unused benchmark report converter script&lt;/a&gt;.&lt;/p&gt;</description>
</item>
<item>
<title>Fixed: [Bug]: Static kernel cache is not reused after restarting with the same model</title>
<link>https://github.com/vllm-project/vllm-ascend/issues/19176</link>
<guid isPermaLink="false">archestack-watch-fixed-best-of-ai/vllm-ascend-19176</guid>
<pubDate>Mon, 19 Oct 2026 00:00:00 GMT</pubDate>
<description>&lt;p&gt;&lt;a href=&quot;https://github.com/vllm-project/vllm-ascend/issues/19176&quot;&gt;[Bug]: Static kernel cache is not reused after restarting with the same model&lt;/a&gt; closed by &lt;a href=&quot;https://github.com/vllm-project/vllm-ascend/pull/19177&quot;&gt;[BugFix][Compilation] Initialize static kernel runtime before loading cached graphs&lt;/a&gt;.&lt;/p&gt;</description>
</item>
<item>
<title>Fixed: [Bug]: off-stride `max_num_seqs` silently disables ACL graph capture for the largest batches (~7.5x slower decode)</title>
<link>https://github.com/vllm-project/vllm-ascend/issues/16049</link>
<guid isPermaLink="false">archestack-watch-fixed-best-of-ai/vllm-ascend-16049</guid>
<pubDate>Wed, 23 Sep 2026 00:00:00 GMT</pubDate>
<description>&lt;p&gt;&lt;a href=&quot;https://github.com/vllm-project/vllm-ascend/issues/16049&quot;&gt;[Bug]: off-stride `max_num_seqs` silently disables ACL graph capture for the largest batches (~7.5x slower decode)&lt;/a&gt; closed by &lt;a href=&quot;https://github.com/vllm-project/vllm-ascend/pull/16970&quot;&gt;[BugFix] Add off-stride graph capture size ceilings&lt;/a&gt;.&lt;/p&gt;</description>
</item>
<item>
<title>vllm-ascend v0.23.0.post1</title>
<link>https://github.com/vllm-project/vllm-ascend/releases/tag/v0.23.0.post1</link>
<guid isPermaLink="false">archestack-watch-release-best-of-ai/vllm-ascend-v0.23.0.post1</guid>
<pubDate>Tue, 22 Sep 2026 00:00:00 GMT</pubDate>
<description>&lt;p&gt;vllm-ascend v0.23.0.post1 was released on 2026-09-22. Its notes do not flag a breaking change or a security fix; that does not mean the upgrade is safe.&lt;/p&gt;</description>
</item>
<item>
<title>New problem reported: [Bug]: DeepSeek-V4.1-Flash outputs gibberish on A3 dual-node mixed deployment with vllm-ascend 0.27.1</title>
<link>https://github.com/vllm-project/vllm-ascend/issues/16870</link>
<guid isPermaLink="false">archestack-watch-problem-best-of-ai/vllm-ascend-16870</guid>
<pubDate>Fri, 18 Sep 2026 00:00:00 GMT</pubDate>
<description>&lt;p&gt;Open on GitHub, 1 reaction: &lt;a href=&quot;https://github.com/vllm-project/vllm-ascend/issues/16870&quot;&gt;[Bug]: DeepSeek-V4.1-Flash outputs gibberish on A3 dual-node mixed deployment with vllm-ascend 0.27.1&lt;/a&gt;&lt;/p&gt;</description>
</item>
<item>
<title>New problem reported: [Bug]: glm-5.3-w8a8c8在vllm-ascend0.26.0rc版本推理启动成功，evalscope性能测试崩溃</title>
<link>https://github.com/vllm-project/vllm-ascend/issues/15906</link>
<guid isPermaLink="false">archestack-watch-problem-best-of-ai/vllm-ascend-15906</guid>
<pubDate>Mon, 07 Sep 2026 00:00:00 GMT</pubDate>
<description>&lt;p&gt;Open on GitHub, 1 reaction: &lt;a href=&quot;https://github.com/vllm-project/vllm-ascend/issues/15906&quot;&gt;[Bug]: glm-5.3-w8a8c8在vllm-ascend0.26.0rc版本推理启动成功，evalscope性能测试崩溃&lt;/a&gt;&lt;/p&gt;</description>
</item>
<item>
<title>New problem reported: [Misc]: 为什么2台服务器当1D</title>
<link>https://github.com/vllm-project/vllm-ascend/issues/15453</link>
<guid isPermaLink="false">archestack-watch-problem-best-of-ai/vllm-ascend-15453</guid>
<pubDate>Tue, 01 Sep 2026 00:00:00 GMT</pubDate>
<description>&lt;p&gt;Open on GitHub, 1 reaction: &lt;a href=&quot;https://github.com/vllm-project/vllm-ascend/issues/15453&quot;&gt;[Misc]: 为什么2台服务器当1D&lt;/a&gt;&lt;/p&gt;</description>
</item>
<item>
<title>New problem reported: [Bug]: Dsv4 speculative dsaprk error on 910C</title>
<link>https://github.com/vllm-project/vllm-ascend/issues/14802</link>
<guid isPermaLink="false">archestack-watch-problem-best-of-ai/vllm-ascend-14802</guid>
<pubDate>Mon, 24 Aug 2026 00:00:00 GMT</pubDate>
<description>&lt;p&gt;Open on GitHub, 1 reaction: &lt;a href=&quot;https://github.com/vllm-project/vllm-ascend/issues/14802&quot;&gt;[Bug]: Dsv4 speculative dsaprk error on 910C&lt;/a&gt;&lt;/p&gt;</description>
</item>
<item>
<title>vllm-ascend v0.23.0</title>
<link>https://github.com/vllm-project/vllm-ascend/releases/tag/v0.23.0</link>
<guid isPermaLink="false">archestack-watch-release-best-of-ai/vllm-ascend-v0.23.0</guid>
<pubDate>Sun, 16 Aug 2026 00:00:00 GMT</pubDate>
<description>&lt;p&gt;vllm-ascend v0.23.0 was released on 2026-08-16. Its notes do not flag a breaking change or a security fix; that does not mean the upgrade is safe.&lt;/p&gt;</description>
</item>
<item>
<title>Fixed: 【BUG】CPU binding may fail where the operating system language is Chinese</title>
<link>https://github.com/vllm-project/vllm-ascend/issues/6992</link>
<guid isPermaLink="false">archestack-watch-fixed-best-of-ai/vllm-ascend-6992</guid>
<pubDate>Thu, 16 Apr 2026 00:00:00 GMT</pubDate>
<description>&lt;p&gt;&lt;a href=&quot;https://github.com/vllm-project/vllm-ascend/issues/6992&quot;&gt;【BUG】CPU binding may fail where the operating system language is Chinese&lt;/a&gt; closed by &lt;a href=&quot;https://github.com/vllm-project/vllm-ascend/pull/8251&quot;&gt;[BugFix] Enforce C locale for CPU binding subprocess parsing&lt;/a&gt;.&lt;/p&gt;</description>
</item>
</channel>
</rss>
