<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0"><channel><title><![CDATA[安装运行strata 遇到问题，请教一下大家]]></title><description><![CDATA[<p dir="auto">这个是运行的样子<br />
出错的时候RuntimeError: the engine exited before it was ready (see I:\Strata-main\strata-iq3_xxs.log)</p>
<p dir="auto">Strata - Qwen3.8-Flash-Next on a normal PC (a GPU + system RAM + CPU)<br />
[ok] GPUs: GPU 0 (NVIDIA GeForce RTX 2080 Ti, 22 GB) + GPU 1 (NVIDIA GeForce RTX 2080 Ti, 22 GB) together (layers split auto)</p>
<hr />
<h2>Starting qwen3.8-flash-next-iq3_xxs: it loads about 47 GB into RAM and locks part of it for the GPU.<br />
While it does, YOUR PC CAN BE SLOW OR STOP RESPONDING FOR 1-3 MINUTES (longer the first time after a<br />
restart). That is normal: please wait and don't close this window - the browser opens when it is ready.<br />
Later, closing this window stops the model.</h2>
<p dir="auto">Settings (strata-iq3_xxs.json): --expert-cache auto --prefill auto --spec 4 --spec-min-p 0.5<br />
--max-context 131072 --kv int8 --kv-resident 32768; server 127.0.0.1:8080, gpu 0,1, layer_split auto<br />
loading the model (the first start takes a minute or two) ...<br />
[strata] layer split across GPUs [0, 1] (auto)<br />
[strata] starting the engine: reading the model's weights ...<br />
[strata] still starting (20 s) - please wait ...<br />
[strata] loading the experts into RAM (about 47 GB) and locking part of them for the GPU.<br />
YOUR PC CAN BE SLOW OR STOP RESPONDING FOR 1-3 MINUTES NOW - this is normal.<br />
Please wait and don't close this window; the browser opens when it is ready.<br />
[strata] still starting (49 s) - please wait ...<br />
[strata] still starting (69 s) - please wait ...<br />
[strata] still starting (89 s) - please wait ...<br />
[strata] still starting (109 s) - please wait ...<br />
[strata] still starting (129 s) - please wait ...<br />
[strata] still starting (149 s) - please wait ...<br />
[strata] still starting (169 s) - please wait ...<br />
[strata] still starting (189 s) - please wait ...<br />
[strata] still starting (209 s) - please wait ...<br />
[strata] still starting (229 s) - please wait ...<br />
[strata] still starting (249 s) - please wait ...<br />
[strata] still starting (269 s) - please wait ...<br />
[strata] still starting (289 s) - please wait ...<br />
[strata] still starting (309 s) - please wait ...<br />
[strata] still starting (329 s) - please wait ...<br />
[strata] still starting (349 s) - please wait ...<br />
[strata] still starting (369 s) - please wait ...<br />
[strata] still starting (389 s) - please wait ...<br />
[strata] still starting (409 s) - please wait ...<br />
[strata] still starting (429 s) - please wait ...<br />
[strata] still starting (449 s) - please wait ...<br />
[strata] still starting (469 s) - please wait ...<br />
[strata] still starting (489 s) - please wait ...<br />
[strata] still starting (509 s) - please wait ...<br />
[strata] experts loaded: 39.97 GiB at 0.08 GiB/s (522 s so far)<br />
[strata] filling the GPU's expert cache (11260 experts, 16.34 GiB of VRAM) ...<br />
[strata] almost ready ...<br />
Traceback (most recent call last):<br />
File "I:\Strata-main\serve\server.py", line 5895, in &lt;module&gt;<br />
sys.exit(main())<br />
^^^^^^<br />
File "I:\Strata-main\serve\server.py", line 5720, in main<br />
engine = StrataEngine(exe, engine_args(cfg) + (effort_end or []), cwd=cfg.get("cwd"), log=cfg.get("log"),<br />
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^<br />
File "I:\Strata-main\serve\server.py", line 693, in <strong>init</strong><br />
raise RuntimeError(why + (f" (see {log})" if log else "") +<br />
RuntimeError: the engine exited before it was ready (see I:\Strata-main\strata-iq3_xxs.log)<br />
the engine log's last lines:<br />
strata generate: 11 pool workers on logical processors 2,4,6,8,10,12,14,16,18,20,22, the host thread on 0 (draining too) (--host-core first)<br />
strata generate: expert cache auto: 17.03 GiB free, 700 MiB reserved (+0 MiB for the draft head) -&gt; 7533 slots<br />
strata generate: expert cache 11260 slots, 16.34 GiB of VRAM; policy is<br />
strata generate: the GPU computes the experts in the cache; it rounds differently from the CPU,<br />
so a reply can differ slightly from a run without the cache (same quality:<br />
bench/results/2026-09-27-cache-parity).<br />
PROFILE, ranked by routing frequency, no eviction.<br />
strata generate: pre-filled 11260 of 11260 slots from the profile in 3.4 s (5156 MB/s); slot 0 verified<br />
strata generate: layer split, CUDA1: 15.81 GiB free of 22.00, room for experts 15.04 GiB<br />
strata generate: layer split: CUDA1 runs layers 26-47, expert cache 8190 slots (15.04 GiB), 8190 of its 11264 profiled pairs; slot 0 verified<br />
strata generate: layer split: CUDA0 runs layers 0-25<br />
strata generate: layer split auto: L2 probe: the split is startable on the live free figures<br />
strata generate: R4 hit path ON - resident experts are computed on the GPU<br />
strata generate: 11 expert-pool workers + the host thread<br />
strata generate: session is up (engine 0.1.41)<br />
strata generate: token graph hit path: 19450 resident experts, decided on the device<br />
strata serve: layer split: 79% of the experts resident, the prompt path's streamed ring 96 slots<br />
strata serve: prompt chunk auto: 8192 tokens, a 96-slot ring<br />
strata serve: the prompt path borrows 2429 CUDA0 cache slots (3.50 GiB)<br />
strata serve:   CUDA1 prompt path borrows 1905 of its 8190 slots (3.50 GiB)<br />
请按任意键继续. . .</p>
]]></description><link>https://lcz.me/topic/2267</link><generator>RSS for Node</generator><lastBuildDate>Mon, 12 Oct 2026 12:24:29 GMT</lastBuildDate><atom:link href="https://lcz.me/topic/2267.rss" rel="self" type="application/rss+xml"/><pubDate>Sun, 11 Oct 2026 13:01:38 GMT</pubDate><ttl>60</ttl><item><title><![CDATA[Reply to 安装运行strata 遇到问题，请教一下大家 on Mon, 12 Oct 2026 04:25:58 GMT]]></title><description><![CDATA[<blockquote>
<p dir="auto"><a class="plugin-mentions-user plugin-mentions-a" href="/user/seabass" aria-label="Profile: seabass">@<bdi>seabass</bdi></a> <a href="/post/24140">说</a>:</p>
<p dir="auto">--kv-resident 32768</p>
</blockquote>
<p dir="auto">请问一下，你用什么版本的strata , 我改了参数 --kv-resident 32768 发现不支持这些参数</p>
]]></description><link>https://lcz.me/post/24179</link><guid isPermaLink="true">https://lcz.me/post/24179</guid><dc:creator><![CDATA[am Am]]></dc:creator><pubDate>Mon, 12 Oct 2026 04:25:58 GMT</pubDate></item><item><title><![CDATA[Reply to 安装运行strata 遇到问题，请教一下大家 on Mon, 12 Oct 2026 03:04:25 GMT]]></title><description><![CDATA[<blockquote>
<p dir="auto"><a class="plugin-mentions-user plugin-mentions-a" href="/user/seabass" aria-label="Profile: seabass">@<bdi>seabass</bdi></a> <a href="/post/24140">说</a>:</p>
<p dir="auto">我的参数（基本全 auto）：<br />
--expert-cache auto --prefill auto --spec 4 --spec-min-p 0.5 --mtp ... --max-context 262144 --kv int8 --kv-resident 32768 --vram-reserve-mib 700 --vision</p>
</blockquote>
<p dir="auto">我再测试一下，谢谢大家，我在考虑要不要重新安装一遍系统<br />
再折腾一下，有了新的结果跟大家分享。</p>
]]></description><link>https://lcz.me/post/24146</link><guid isPermaLink="true">https://lcz.me/post/24146</guid><dc:creator><![CDATA[am Am]]></dc:creator><pubDate>Mon, 12 Oct 2026 03:04:25 GMT</pubDate></item><item><title><![CDATA[Reply to 安装运行strata 遇到问题，请教一下大家 on Mon, 12 Oct 2026 02:42:44 GMT]]></title><description><![CDATA[<p dir="auto">我用 Claude 做翻译和核对。</p>
<p dir="auto">我的配置和你很接近，可以做个对照：双 2080 Ti 22G，E5-2666 v3（和你的 2678 v3 同代，AVX2，没有 AVX-512），252G 内存，Ubuntu。<br />
模型是 Qwen3.8-Flash-Next IQ3_S，262K 上下文，解码 55–77 tok/s，29K 长 prompt 预填充约 1050 tok/s。所以 2080 Ti 和这代 CPU 本身没问题，7 tok/s 肯定不正常。</p>
<p dir="auto">我的参数（基本全 auto）：<br />
--expert-cache auto --prefill auto --spec 4 --spec-min-p 0.5 --mtp ... --max-context 262144 --kv int8 --kv-resident 32768 --vram-reserve-mib 700 --vision<br />
日志里：层自动切分 0–19 在卡0、20–47 在卡1；pcie_frac 自动探测为 0.23（PCIe 10.8 GB/s）；专家缓存每张卡约 15GB，加载完显存只剩约 470MB；专家缓存命中率加权平均约 97%。</p>
<p dir="auto">我觉得问题主要在内存和硬盘：</p>
<ol>
<li>启动时约 47GB 专家载入锁页内存，我这边运行时系统内存占用约 60GB。你 64G 内存还要跑 Windows，非常紧。如果专家放不下，解码时就要去读盘，速度会掉得很厉害。</li>
<li>盘速对比（直接读，绕过缓存）：我的 SATA SSD 546 MB/s，冷启动加载 47GB 约 90 秒；挪到 NVMe（1.5 GB/s）后加载速度从 0.52 提到 1.45 GiB/s，约 66<br />
秒就绪。解码速度两者一样，因为专家全在内存/显存里时根本不读盘。</li>
<li>你说 SSD 只有 300 多 MB/s，这正好是 SATA II（3Gb/s）的上限，老 X79/X99 板很常见。用 CrystalDiskInfo 看传输模式是 SATA/300 还是 SATA/600，能换就插到 SATA3 口。机械硬盘就更不行了。</li>
</ol>
<p dir="auto">建议自查：</p>
<ul>
<li>看日志里的 expert cache hit rate，正常应该在 90% 以上；KV streaming ... hit VRAM 也应该接近 100%。</li>
<li>生成时打开任务管理器，看磁盘占用是不是一直很高。如果是，就说明在从盘读专家。</li>
<li>上下文降到 8K 对显存帮助不大：KV 只有 --kv-resident 那部分（32K）常驻显存，其余在内存里。</li>
</ul>
]]></description><link>https://lcz.me/post/24140</link><guid isPermaLink="true">https://lcz.me/post/24140</guid><dc:creator><![CDATA[seabass]]></dc:creator><pubDate>Mon, 12 Oct 2026 02:42:44 GMT</pubDate></item><item><title><![CDATA[Reply to 安装运行strata 遇到问题，请教一下大家 on Mon, 12 Oct 2026 00:02:47 GMT]]></title><description><![CDATA[<p dir="auto">7 t/s 按下面顺序排，基本能定位：</p>
<ol>
<li>先确认 decode 每 token 要读的专家/PLE 是不是还在走 SSD。3843 那条是对的：先 <code>--ple-io ram</code> 把 PLE 表搬进内存，并把 <code>--pcie-frac</code> 调高，让瓶颈落在 PCIe 而不是磁盘。</li>
<li>2080Ti 是 Turing（sm_75），没有 BF16/FP8，只能走 INT8/FP16 路径。确认日志里 kernel 的选择；如果引擎默认按 FP8/BF16 编译，会退化成更慢的路径甚至回退 CPU。</li>
<li>用任务管理器看 decode 时谁忙：磁盘 100% → I/O 瓶颈；GPU 闲 → 卡在专家加载/PCIe；CPU 单核 100% → 调度/拆分问题。三种原因给的解法完全不同。</li>
<li>现设置 <code>--kv-resident 32768</code> 在 44G 卡上偏保守；上下文已降到 8K，可以先把 <code>--expert-cache</code> 从 <code>auto</code> 改成固定值，把热专家钉在显存，避免每 token 重读。</li>
</ol>
<p dir="auto">建议直接参考 topic/2224，那位机器和你同量级，先抄配置再微调。</p>
]]></description><link>https://lcz.me/post/24104</link><guid isPermaLink="true">https://lcz.me/post/24104</guid><dc:creator><![CDATA[Xiaote]]></dc:creator><pubDate>Mon, 12 Oct 2026 00:02:47 GMT</pubDate></item><item><title><![CDATA[Reply to 安装运行strata 遇到问题，请教一下大家 on Sun, 11 Oct 2026 23:17:28 GMT]]></title><description><![CDATA[<p dir="auto">机械硬盘太慢了，玩不了</p>
]]></description><link>https://lcz.me/post/24098</link><guid isPermaLink="true">https://lcz.me/post/24098</guid><dc:creator><![CDATA[marsempire]]></dc:creator><pubDate>Sun, 11 Oct 2026 23:17:28 GMT</pubDate></item><item><title><![CDATA[Reply to 安装运行strata 遇到问题，请教一下大家 on Sun, 11 Oct 2026 20:23:18 GMT]]></title><description><![CDATA[<blockquote>
<p dir="auto"><a class="plugin-mentions-user plugin-mentions-a" href="/user/j-z-0" aria-label="Profile: J-Z-0">@<bdi>J-Z-0</bdi></a> <a href="/post/24088">说</a>:</p>
<p dir="auto">什么硬盘？ 我的机器只有300多m/s读取的ssd，用 --ple-io ram 把PLE 表搬进内存才解决。<br />
还有是什么cpu，如果太弱，就把--pcie-frac 加高。<br />
你看我这个帖子。 <a href="https://lcz.me/topic/2224">https://lcz.me/topic/2224</a></p>
<p dir="auto">你显存大，用q2模型，64g内存我觉得也能用这个设置</p>
</blockquote>
<p dir="auto">我还没买到固态，还是机械硬盘，目前配置如下。<br />
系统，WIN10<br />
内存DDR3 4通道64G<br />
CPU (E5-2678V3,<br />
显卡 2080TI *2  44G 显存</p>
]]></description><link>https://lcz.me/post/24091</link><guid isPermaLink="true">https://lcz.me/post/24091</guid><dc:creator><![CDATA[am Am]]></dc:creator><pubDate>Sun, 11 Oct 2026 20:23:18 GMT</pubDate></item><item><title><![CDATA[Reply to 安装运行strata 遇到问题，请教一下大家 on Sun, 11 Oct 2026 19:57:19 GMT]]></title><description><![CDATA[<p dir="auto">什么硬盘？ 我的机器只有300多m/s读取的ssd，用 --ple-io ram 把PLE 表搬进内存才解决。<br />
还有是什么cpu，如果太弱，就把--pcie-frac 加高。<br />
你看我这个帖子。 <a href="https://lcz.me/topic/2224">https://lcz.me/topic/2224</a></p>
<p dir="auto">你显存大，用q2模型，64g内存我觉得也能用这个设置</p>
]]></description><link>https://lcz.me/post/24088</link><guid isPermaLink="true">https://lcz.me/post/24088</guid><dc:creator><![CDATA[J Z 0]]></dc:creator><pubDate>Sun, 11 Oct 2026 19:57:19 GMT</pubDate></item><item><title><![CDATA[Reply to 安装运行strata 遇到问题，请教一下大家 on Sun, 11 Oct 2026 19:27:10 GMT]]></title><description><![CDATA[<p dir="auto">谢谢大家，我发现之前的问题的确可能是内存或者显存不足，我把模型换到Q2_0,然后把上下文设置到8K,就可以打开了，但是运行的速度只有7token 每秒，我的电脑是44G显存，64G内存，不知道是哪里出了问题。</p>
]]></description><link>https://lcz.me/post/24087</link><guid isPermaLink="true">https://lcz.me/post/24087</guid><dc:creator><![CDATA[am Am]]></dc:creator><pubDate>Sun, 11 Oct 2026 19:27:10 GMT</pubDate></item><item><title><![CDATA[Reply to 安装运行strata 遇到问题，请教一下大家 on Sun, 11 Oct 2026 14:32:18 GMT]]></title><description><![CDATA[<p dir="auto">试一试Max content = 8k, 先跑起来再加大</p>
]]></description><link>https://lcz.me/post/24037</link><guid isPermaLink="true">https://lcz.me/post/24037</guid><dc:creator><![CDATA[imbiplaza ASUS]]></dc:creator><pubDate>Sun, 11 Oct 2026 14:32:18 GMT</pubDate></item><item><title><![CDATA[Reply to 安装运行strata 遇到问题，请教一下大家 on Sun, 11 Oct 2026 13:58:10 GMT]]></title><description><![CDATA[<p dir="auto">现在哪个还自己搞系统？让OPENCLAW或者HERMES搞定就行了</p>
]]></description><link>https://lcz.me/post/24025</link><guid isPermaLink="true">https://lcz.me/post/24025</guid><dc:creator><![CDATA[九门奇人]]></dc:creator><pubDate>Sun, 11 Oct 2026 13:58:10 GMT</pubDate></item><item><title><![CDATA[Reply to 安装运行strata 遇到问题，请教一下大家 on Sun, 11 Oct 2026 13:24:53 GMT]]></title><description><![CDATA[<p dir="auto">首先，问问题你应该先打出你基本的设备信息，其次，你这什么都没说明就丢一堆日志过来真的很不尊重人方便他人完全是方便自己，最起码得详细解释一下为什么会发生这种原因，最后这种问题不能问AI吗？这种软件小问题哪个智能体都能给一定解释，从日志看就是爆显存了，调配有问题，两块2080ti44g总容量是够的，但占太满了没留下足够缓冲区，自己下个dsh让AI解决就行了</p>
]]></description><link>https://lcz.me/post/24011</link><guid isPermaLink="true">https://lcz.me/post/24011</guid><dc:creator><![CDATA[ikum]]></dc:creator><pubDate>Sun, 11 Oct 2026 13:24:53 GMT</pubDate></item></channel></rss>