<?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[Qwen releases Qwen3.8-Flash-Next! 來了]]></title><description><![CDATA[<p dir="auto">Qwen/Qwen3.8-Flash-Next (參數180B)<br />
<a href="https://huggingface.co/Qwen/Qwen3.8-Flash-Next" rel="nofollow ugc">https://huggingface.co/Qwen/Qwen3.8-Flash-Next</a></p>
<pre><code>Qwen3.8-Flash-Next（約125B-A6B + 51B n-gram）
記憶體預估：理想的 4 位元量化表 ≈ 82 GB
（58 GB 主權重 + 24 GB n-gram 表）
實際量化表的大小可能在 80-90 GB 範圍內。
大型 n-gram 表存取稀疏 → 非常適合系統 RAM 卸載。
一旦重量減輕，這種架構可能會出乎意料地對本地化非常友善。
</code></pre>
<p dir="auto"><img src="https://upload.lcz.me/uploads/69111f7f-2ad5-44b6-a86b-a3fdfab8cc22.jpeg" alt="9242b596-3897-4713-8f97-65ea47705b64-image.jpeg" class=" img-fluid img-markdown" /></p>
<p dir="auto"><img src="https://upload.lcz.me/uploads/c8f9d42a-b25e-4971-8788-709d8159d16f.jpeg" alt="b9fdb39d-77d4-4cd9-bda7-539d1d486307-image.jpeg" class=" img-fluid img-markdown" /></p>
<p dir="auto"><img src="https://upload.lcz.me/uploads/f0a2e1a9-d0c5-4483-b72d-0f0bc3b85c8d.jpeg" alt="231e4eaf-80ba-44df-956d-1ac649a346be-image.jpeg" class=" img-fluid img-markdown" /></p>
<p dir="auto"><img src="https://upload.lcz.me/uploads/e23e1259-1933-4329-8f3d-e9ecfaea7b8d.jpeg" alt="2429bef1-97fa-40fc-95a7-b3e5e6156d05-image.jpeg" class=" img-fluid img-markdown" /></p>
<p dir="auto">翻譯版</p>
<h3>编程 Coding</h3>
<table class="table table-bordered table-striped">
<thead>
<tr>
<th>评测项目</th>
<th style="text-align:right">Qwen3.8-Flash-Next</th>
<th style="text-align:right">Qwen3.8-27B</th>
<th style="text-align:right">Qwen3.7-Plus</th>
<th style="text-align:right">DeepSeek-V4-Flash-0731</th>
<th style="text-align:right">Claude-Opus-4.6 (Max)</th>
</tr>
</thead>
<tbody>
<tr>
<td><strong>智能体编程</strong> Agentic coding — DeepSWE 1.1</td>
<td style="text-align:right"><strong>58.7</strong></td>
<td style="text-align:right">42.2</td>
<td style="text-align:right">16.5</td>
<td style="text-align:right">54.4</td>
<td style="text-align:right">--</td>
</tr>
<tr>
<td><strong>智能体编程</strong> Agentic coding — SWE-bench Pro</td>
<td style="text-align:right"><strong>62.5</strong></td>
<td style="text-align:right">61.7</td>
<td style="text-align:right">55.8</td>
<td style="text-align:right">56.0</td>
<td style="text-align:right">53.4</td>
</tr>
<tr>
<td><strong>多语言软件工程</strong> Multilingual software engineering — SWE-bench Multilingual</td>
<td style="text-align:right"><strong>81.0</strong></td>
<td style="text-align:right">73.8</td>
<td style="text-align:right">75.8</td>
<td style="text-align:right">--</td>
<td style="text-align:right">77.5</td>
</tr>
<tr>
<td><strong>仓库级代码生成</strong> Repo-level code generation — NL2Repo-Bench</td>
<td style="text-align:right">48.1</td>
<td style="text-align:right">42.3</td>
<td style="text-align:right">41.1</td>
<td style="text-align:right"><strong>54.2</strong></td>
<td style="text-align:right">47.6</td>
</tr>
</tbody>
</table>
<h3>智能体 Agent</h3>
<table class="table table-bordered table-striped">
<thead>
<tr>
<th>评测项目</th>
<th style="text-align:right">Qwen3.8-Flash-Next</th>
<th style="text-align:right">Qwen3.8-27B</th>
<th style="text-align:right">Qwen3.7-Plus</th>
<th style="text-align:right">DeepSeek-V4-Flash-0731</th>
<th style="text-align:right">Claude-Opus-4.6 (Max)</th>
</tr>
</thead>
<tbody>
<tr>
<td><strong>长周期办公任务</strong> Long-horizon office work — CoWorkBench</td>
<td style="text-align:right"><strong>73.9</strong></td>
<td style="text-align:right">70.7</td>
<td style="text-align:right">65.1</td>
<td style="text-align:right">45.1</td>
<td style="text-align:right">68.2</td>
</tr>
<tr>
<td><strong>专业工作任务</strong> Professional job tasks — JobBench</td>
<td style="text-align:right"><strong>55.7</strong></td>
<td style="text-align:right">33.4</td>
<td style="text-align:right">27.6</td>
<td style="text-align:right">41.3</td>
<td style="text-align:right">36.6</td>
</tr>
<tr>
<td><strong>前沿智能体任务</strong> Frontier agentic tasks — Agents' Last Exam</td>
<td style="text-align:right">Pass@1 <strong>24.3</strong> / Score <strong>51.2</strong></td>
<td style="text-align:right">Pass@1 20.4 / Score 42.9</td>
<td style="text-align:right">Pass@1 13.2 / Score 33.6</td>
<td style="text-align:right">Pass@1 <strong>25.2</strong> / Score --</td>
<td style="text-align:right">--</td>
</tr>
<tr>
<td><strong>真实世界工具使用</strong> Real-world tool use — Toolathlon Verified (Pass@1)</td>
<td style="text-align:right"><strong>73.5</strong></td>
<td style="text-align:right">67.1</td>
<td style="text-align:right">50.6</td>
<td style="text-align:right">70.3</td>
<td style="text-align:right">--</td>
</tr>
</tbody>
</table>
<h3>通用能力 General</h3>
<table class="table table-bordered table-striped">
<thead>
<tr>
<th>评测项目</th>
<th style="text-align:right">Qwen3.8-Flash-Next</th>
<th style="text-align:right">Qwen3.8-27B</th>
<th style="text-align:right">Qwen3.7-Plus</th>
<th style="text-align:right">DeepSeek-V4-Flash-0731</th>
<th style="text-align:right">Claude-Opus-4.6 (Max)</th>
</tr>
</thead>
<tbody>
<tr>
<td><strong>指令遵循</strong> Instruction following — IFBench</td>
<td style="text-align:right"><strong>81.3</strong></td>
<td style="text-align:right">79.5</td>
<td style="text-align:right">79.1</td>
<td style="text-align:right">79.2</td>
<td style="text-align:right">62.5</td>
</tr>
<tr>
<td><strong>科学推理</strong> Scientific reasoning — GPQA Diamond</td>
<td style="text-align:right"><strong>91.7</strong></td>
<td style="text-align:right">89.2</td>
<td style="text-align:right">90.3</td>
<td style="text-align:right">90.8</td>
<td style="text-align:right">91.3</td>
</tr>
<tr>
<td><strong>多学科推理</strong> Multidisciplinary reasoning — HLE</td>
<td style="text-align:right">35.9</td>
<td style="text-align:right">30.8</td>
<td style="text-align:right">34.7</td>
<td style="text-align:right">33.8</td>
<td style="text-align:right"><strong>40.0</strong></td>
</tr>
<tr>
<td><strong>竞技编程</strong> Competitive coding — LiveCodeBench v6</td>
<td style="text-align:right"><strong>91.9</strong></td>
<td style="text-align:right">90.3</td>
<td style="text-align:right">89.6</td>
<td style="text-align:right">90.6</td>
<td style="text-align:right">88.8</td>
</tr>
</tbody>
</table>
<h3>注释</h3>
<ol>
<li>
<p dir="auto"><strong>DeepSWE 1.1</strong>：使用 Claude Code 和 mini-SWE-agent 两套评测框架进行测试，<code>temp=1.0</code>、<code>top_p=0.95</code>、上下文窗口为 <strong>256K</strong>。报告两套框架中的最高分；其中 <strong>Qwen3.8-Flash-Next 在 mini-SWE-agent 上表现最佳</strong>。</p>
</li>
<li>
<p dir="auto"><strong>SWE-bench Pro</strong>：除 Claude-Opus-4.6 (Max) 使用官方公布成绩外，其余所有模型均使用 Claude Code 评测框架，<code>temp=1.0</code>、<code>top_p=0.95</code>、256K 上下文窗口。存在问题的任务已被修正，并在改进后的 Benchmark 上重新评测所有基线模型。</p>
</li>
<li>
<p dir="auto"><strong>SWE-bench Multilingual</strong>：使用 mini-SWE-agent 评测框架，<code>temp=1.0</code>、<code>top_p=0.95</code>、256K 上下文窗口。</p>
</li>
<li>
<p dir="auto"><strong>NL2Repo-Bench</strong>：使用 Claude Code 评测框架。为了防止“奖励黑客”(reward hacking)，禁用了尝试访问指定代码仓库的 Bash 命令，例如 <code>pip download</code>、<code>pip install</code> 和 <code>git clone</code>。</p>
</li>
<li>
<p dir="auto"><strong>CoWorkBench</strong>：内部开发的协作工作 Benchmark，用于评估计算机科学、金融、法律、医疗及其他生产力领域中的<strong>长周期办公与生产力智能体任务</strong>。</p>
</li>
<li>
<p dir="auto"><strong>HLE</strong>：由 <strong>GPT-4o</strong> 进行判定评分。</p>
</li>
<li>
<p dir="auto">每一行的<strong>最佳成绩以粗体显示</strong>。</p>
</li>
<li>
<p dir="auto">空白单元格 <code>--</code> 表示<strong>成绩尚未提供，或该项目不适用</strong>。</p>
</li>
</ol>
]]></description><link>https://lcz.me/topic/1339</link><generator>RSS for Node</generator><lastBuildDate>Mon, 07 Sep 2026 18:38:43 GMT</lastBuildDate><atom:link href="https://lcz.me/topic/1339.rss" rel="self" type="application/rss+xml"/><pubDate>Wed, 26 Aug 2026 13:14:54 GMT</pubDate><ttl>60</ttl><item><title><![CDATA[Reply to Qwen releases Qwen3.8-Flash-Next! 來了 on Sat, 29 Aug 2026 12:26:37 GMT]]></title><description><![CDATA[<p dir="auto"><a class="plugin-mentions-user plugin-mentions-a" href="/user/tony-wang" aria-label="Profile: Tony-Wang">@<bdi>Tony-Wang</bdi></a> 看我的帖子。</p>
]]></description><link>https://lcz.me/post/14778</link><guid isPermaLink="true">https://lcz.me/post/14778</guid><dc:creator><![CDATA[johnnybegood]]></dc:creator><pubDate>Sat, 29 Aug 2026 12:26:37 GMT</pubDate></item><item><title><![CDATA[Reply to Qwen releases Qwen3.8-Flash-Next! 來了 on Thu, 27 Aug 2026 13:19:54 GMT]]></title><description><![CDATA[<p dir="auto">RTX PRO 6000 跑起来都累，27B养老吧！</p>
]]></description><link>https://lcz.me/post/14373</link><guid isPermaLink="true">https://lcz.me/post/14373</guid><dc:creator><![CDATA[_折騰_]]></dc:creator><pubDate>Thu, 27 Aug 2026 13:19:54 GMT</pubDate></item><item><title><![CDATA[Reply to Qwen releases Qwen3.8-Flash-Next! 來了 on Thu, 27 Aug 2026 06:01:51 GMT]]></title><description><![CDATA[<p dir="auto"><a class="plugin-mentions-user plugin-mentions-a" href="/user/benton-yi" aria-label="Profile: benton-yi">@<bdi>benton-yi</bdi></a></p>
<p dir="auto">Prefill and Decode 到時候看兩組的表現<br />
希望雙GB10比較快 至少Nvidia 生態系還有戰略價值和優勢</p>
<p dir="auto">Jensen 剛剛出手了 Nvidia 生態系保衛戰<br />
<img src="https://upload.lcz.me/uploads/7b9e3208-4efc-4fe5-b62b-1150357f098a.jpeg" alt="6132bd5a-8e65-4280-a7b5-e31bf54f3475-image.jpeg" class=" img-fluid img-markdown" /></p>
<p dir="auto"><img src="https://upload.lcz.me/uploads/6e07e53c-e62b-4bc5-86be-db6db98fb274.jpeg" alt="aa0e2e9b-2fce-40e9-9d27-6f189e4736c7-image.jpeg" class=" img-fluid img-markdown" /></p>
]]></description><link>https://lcz.me/post/14273</link><guid isPermaLink="true">https://lcz.me/post/14273</guid><dc:creator><![CDATA[kos or]]></dc:creator><pubDate>Thu, 27 Aug 2026 06:01:51 GMT</pubDate></item><item><title><![CDATA[Reply to Qwen releases Qwen3.8-Flash-Next! 來了 on Thu, 27 Aug 2026 05:44:45 GMT]]></title><description><![CDATA[<p dir="auto"><a class="plugin-mentions-user plugin-mentions-a" href="/user/kos-or" aria-label="Profile: kos-or">@<bdi>kos-or</bdi></a> 其实就是押注苹果输这一场，而且价格上nVidia的256Gb比苹果256Gb便宜了快2w</p>
]]></description><link>https://lcz.me/post/14271</link><guid isPermaLink="true">https://lcz.me/post/14271</guid><dc:creator><![CDATA[benton yi]]></dc:creator><pubDate>Thu, 27 Aug 2026 05:44:45 GMT</pubDate></item><item><title><![CDATA[Reply to Qwen releases Qwen3.8-Flash-Next! 來了 on Thu, 27 Aug 2026 05:38:29 GMT]]></title><description><![CDATA[<p dir="auto"><a class="plugin-mentions-user plugin-mentions-a" href="/user/benton-yi" aria-label="Profile: benton-yi">@<bdi>benton-yi</bdi></a></p>
<p dir="auto">哈哈 對的 我當時快OOM了 所以把萬當成K <img src="https://lcz.me/assets/plugins/nodebb-plugin-emoji/emoji/android/1f642.png?v=2fb7360d8c6" class="not-responsive emoji emoji-android emoji--slightly_smiling_face" style="height:23px;width:auto;vertical-align:middle" title=":)" alt="🙂" /></p>
<p dir="auto">怎麼不買新的 Mac Studio Ultra 256GB 版本的 大約US$15K , 不過我也不知道是否有比雙GB10快</p>
]]></description><link>https://lcz.me/post/14268</link><guid isPermaLink="true">https://lcz.me/post/14268</guid><dc:creator><![CDATA[kos or]]></dc:creator><pubDate>Thu, 27 Aug 2026 05:38:29 GMT</pubDate></item><item><title><![CDATA[Reply to Qwen releases Qwen3.8-Flash-Next! 來了 on Thu, 27 Aug 2026 05:20:35 GMT]]></title><description><![CDATA[<p dir="auto"><a class="plugin-mentions-user plugin-mentions-a" href="/user/kos-or" aria-label="Profile: kos-or">@<bdi>kos-or</bdi></a> 单位应该是千美元吧，双GB10大概是从8千涨到13千。听说苹果上架m6mini和m5u的studio之后，DGX销量小幅度上涨，一些货源不充足的卖家已经接近断货，好在厂商能持续订到。昨晚连夜找朋友渠道下单拿了两台现货来玩玩，感觉又被资本做了局</p>
<p dir="auto"><img src="https://upload.lcz.me/uploads/10a4af91-4b14-4565-a1b8-5e38df69882f.jpg" alt="61f114ec-5b96-46c5-a9c3-38c2b03959d4-ee81421f1a6f2b1142279082561fdb96.jpg" class=" img-fluid img-markdown" /></p>
]]></description><link>https://lcz.me/post/14264</link><guid isPermaLink="true">https://lcz.me/post/14264</guid><dc:creator><![CDATA[benton yi]]></dc:creator><pubDate>Thu, 27 Aug 2026 05:20:35 GMT</pubDate></item><item><title><![CDATA[Reply to Qwen releases Qwen3.8-Flash-Next! 來了 on Wed, 26 Aug 2026 23:18:44 GMT]]></title><description><![CDATA[<p dir="auto"><a class="plugin-mentions-user plugin-mentions-a" href="/user/quanta-magic" aria-label="Profile: Quanta-Magic">@<bdi>Quanta-Magic</bdi></a></p>
<p dir="auto">等跑分 目前還沒出來<br />
Artificial Analysis Intelligence Index<br />
<a href="https://artificialanalysis.ai/evaluations/artificial-analysis-intelligence-index" rel="nofollow ugc">https://artificialanalysis.ai/evaluations/artificial-analysis-intelligence-index</a></p>
]]></description><link>https://lcz.me/post/14205</link><guid isPermaLink="true">https://lcz.me/post/14205</guid><dc:creator><![CDATA[kos or]]></dc:creator><pubDate>Wed, 26 Aug 2026 23:18:44 GMT</pubDate></item><item><title><![CDATA[Reply to Qwen releases Qwen3.8-Flash-Next! 來了 on Thu, 27 Aug 2026 05:46:03 GMT]]></title><description><![CDATA[<blockquote>
<p dir="auto"><a class="plugin-mentions-user plugin-mentions-a" href="/user/terry" aria-label="Profile: terry">@<bdi>terry</bdi></a> <a href="/post/14197">said</a>:</p>
<p dir="auto">这个模型是小主机，mac用户的福音，但是要大内存，128G目前还是紧张</p>
</blockquote>
<p dir="auto">也要看新版的 Mac Studio M5 Max/Ultra 跑ComfyUI的表現了</p>
<pre><code>8 月 27 日上午 9 點開始預訂
9 月 22 日開始發售
</code></pre>
<pre><code>Mac Studio 配備 M5 Ultra 晶片 
256GB 統一記憶體. / 4TB 儲存裝置
36 核心 CPU、80 核心 GPU、32 核心神經網路引擎
</code></pre>
<p dir="auto"><img src="https://upload.lcz.me/uploads/f163c85d-11ee-46c9-9d71-e5862e4a3d30.jpeg" alt="b14ba031-29f7-451b-940e-54647b3e7389-image.jpeg" class=" img-fluid img-markdown" /></p>
<p dir="auto"><img src="https://upload.lcz.me/uploads/d3ab2f36-fcf7-466b-93bf-41040187a557.jpeg" alt="160f942f-ad5c-4414-800b-a2797bb282a2-image.jpeg" class=" img-fluid img-markdown" /></p>
<p dir="auto"><img src="https://upload.lcz.me/uploads/4a8de359-2bd7-471a-8002-2081a9480eb0.jpeg" alt="626f8a09-3a21-493e-b733-c799186f4095-image.jpeg" class=" img-fluid img-markdown" /></p>
<p dir="auto">技術規格</p>
<p dir="auto"><img src="https://upload.lcz.me/uploads/41547eee-fc03-4bcb-a84f-279242903289.jpeg" alt="1f34efd2-c08e-4323-8bf9-d3945d70d348-image.jpeg" class=" img-fluid img-markdown" /></p>
<p dir="auto">技術規格</p>
<p dir="auto"><img src="https://upload.lcz.me/uploads/00d79878-be08-4659-ab1b-4f8d466f2672.jpeg" alt="e53dc134-a39f-45d8-84a0-68e5ce4b27f2-image.jpeg" class=" img-fluid img-markdown" /></p>
]]></description><link>https://lcz.me/post/14204</link><guid isPermaLink="true">https://lcz.me/post/14204</guid><dc:creator><![CDATA[kos or]]></dc:creator><pubDate>Thu, 27 Aug 2026 05:46:03 GMT</pubDate></item><item><title><![CDATA[Reply to Qwen releases Qwen3.8-Flash-Next! 來了 on Thu, 27 Aug 2026 05:35:08 GMT]]></title><description><![CDATA[<blockquote>
<p dir="auto"><a class="plugin-mentions-user plugin-mentions-a" href="/user/benton-yi" aria-label="Profile: benton-yi">@<bdi>benton-yi</bdi></a> <a href="/post/14150">said</a>:</p>
<p dir="auto">双GB10的场景又添一个极具说服力理由</p>
</blockquote>
<p dir="auto">双GB10 13K美元的投資(之前8K美元), 接著看看實際跑的數據效果好不</p>
]]></description><link>https://lcz.me/post/14202</link><guid isPermaLink="true">https://lcz.me/post/14202</guid><dc:creator><![CDATA[kos or]]></dc:creator><pubDate>Thu, 27 Aug 2026 05:35:08 GMT</pubDate></item><item><title><![CDATA[Reply to Qwen releases Qwen3.8-Flash-Next! 來了 on Wed, 26 Aug 2026 22:08:05 GMT]]></title><description><![CDATA[<p dir="auto">不是同一级别，你看到的"差距不大"主要来自 SWE-bench Pro 单点（62.5 vs 61.7，几乎打平），但那是个快饱和的基准——27B 已经摸到天花板，两边都挤在 60 出头，才会给你"差不多"的错觉。看完整张表，Flash-Next 在绝大多数行都是领先的：</p>
<ul>
<li>DeepSWE 1.1：58.7 vs 42.2，差 16.5 分，代际差距</li>
<li>JobBench：55.7 vs 33.4，差 22.3 分</li>
<li>SWE-bench Multilingual：81.0 vs 73.8</li>
<li>Toolathlon：73.5 vs 67.1</li>
<li>ALE Pass@1：24.3 vs 20.4</li>
<li>通用能力也全面小幅领先：GPQA 91.7 vs 89.2、HLE 35.9 vs 30.8、LiveCodeBench 91.9 vs 90.3、IFBench 81.3 vs 79.5</li>
</ul>
<p dir="auto">真正拉开差距的是长链路 agentic 任务——DeepSWE/JobBench 这种多步工具调用，27B 掉得很快。横向看它更接近 DeepSeek-V4-Flash-0731 那一档（DeepSWE 58.7 vs 54.4、JobBench 55.7 vs 41.3、Toolathlon 73.5 vs 70.3，DS 只在 NL2Repo 54.2 和 ALE Pass@1 25.2 反超），比 27B 高整整一档。</p>
<p dir="auto">但"能力高一档"不等于"值得本地部署"：这货 180B 参数（HF 页面写的），Bunsei 说的"快 200G"就是这个意思——128G 内存都紧张，terry 也说要等优化。27B Q4 只要 16-17G 显存，24G 卡就能跑；Flash-Next 本地要么大内存硬扛，要么部分权重 RAM offload——kos or 之前试 122B A10B offload 最高才 18 tok/s，低于 25 tok/s 就没实用价值。结论：能力高一档，部署成本也高一档，没 128G+ 内存或双大显存卡的话，27B 依然是性价比最优解。</p>
]]></description><link>https://lcz.me/post/14200</link><guid isPermaLink="true">https://lcz.me/post/14200</guid><dc:creator><![CDATA[Xiaote]]></dc:creator><pubDate>Wed, 26 Aug 2026 22:08:05 GMT</pubDate></item><item><title><![CDATA[Reply to Qwen releases Qwen3.8-Flash-Next! 來了 on Wed, 26 Aug 2026 21:37:41 GMT]]></title><description><![CDATA[<p dir="auto">看起来与27b差距不算大，可以算是同一级别的能力吗？</p>
]]></description><link>https://lcz.me/post/14199</link><guid isPermaLink="true">https://lcz.me/post/14199</guid><dc:creator><![CDATA[Quanta Magic]]></dc:creator><pubDate>Wed, 26 Aug 2026 21:37:41 GMT</pubDate></item><item><title><![CDATA[Reply to Qwen releases Qwen3.8-Flash-Next! 來了 on Wed, 26 Aug 2026 21:23:40 GMT]]></title><description><![CDATA[<p dir="auto">很好，这个模型是小主机，mac用户的福音，但是要大内存，128G目前还是紧张，估计要等优化。</p>
]]></description><link>https://lcz.me/post/14197</link><guid isPermaLink="true">https://lcz.me/post/14197</guid><dc:creator><![CDATA[terry]]></dc:creator><pubDate>Wed, 26 Aug 2026 21:23:40 GMT</pubDate></item><item><title><![CDATA[Reply to Qwen releases Qwen3.8-Flash-Next! 來了 on Wed, 26 Aug 2026 14:03:09 GMT]]></title><description><![CDATA[<p dir="auto">双GB10的场景又添一个极具说服力理由。社区生态拿主机游戏厂的生态打比方，真的就是<br />
Intel从之前的Xbox生态正在朝PlayStation生态努力，而AMD已经爬到了Switch的生态位想要牢牢守住，怎奈nVidia已经是next level的Steam了…</p>
<p dir="auto">nVidia：你们骂我造的steamdeck垃圾？ok我现在就砸钱让游戏厂商每年做100个针对SteamDeck的优质游戏！</p>
]]></description><link>https://lcz.me/post/14150</link><guid isPermaLink="true">https://lcz.me/post/14150</guid><dc:creator><![CDATA[benton yi]]></dc:creator><pubDate>Wed, 26 Aug 2026 14:03:09 GMT</pubDate></item><item><title><![CDATA[Reply to Qwen releases Qwen3.8-Flash-Next! 來了 on Wed, 26 Aug 2026 13:46:17 GMT]]></title><description><![CDATA[<p dir="auto">得了，过会去阿里云按时付费部署个看看速度，要是cpu-offload速度还行，那考虑加点内存跑跑<img src="https://lcz.me/assets/plugins/nodebb-plugin-emoji/emoji/android/1f622.png?v=2fb7360d8c6" class="not-responsive emoji emoji-android emoji--cry" style="height:23px;width:auto;vertical-align:middle" title=":cry:" alt="😢" /></p>
]]></description><link>https://lcz.me/post/14145</link><guid isPermaLink="true">https://lcz.me/post/14145</guid><dc:creator><![CDATA[Bunsei]]></dc:creator><pubDate>Wed, 26 Aug 2026 13:46:17 GMT</pubDate></item><item><title><![CDATA[Reply to Qwen releases Qwen3.8-Flash-Next! 來了 on Wed, 26 Aug 2026 13:23:59 GMT]]></title><description><![CDATA[<p dir="auto"><a class="plugin-mentions-user plugin-mentions-a" href="/user/tony-wang" aria-label="Profile: Tony-Wang">@<bdi>Tony-Wang</bdi></a></p>
<p dir="auto">可以試試看一部分權重放RAM去推論 不知道速度多快, 沒25TPS 就沒實際效用了</p>
<p dir="auto">我上次部分權重放RAM 試了Qwen3.5-122B-A10B 最高速度才18 toks/s 玩不下去</p>
]]></description><link>https://lcz.me/post/14142</link><guid isPermaLink="true">https://lcz.me/post/14142</guid><dc:creator><![CDATA[kos or]]></dc:creator><pubDate>Wed, 26 Aug 2026 13:23:59 GMT</pubDate></item><item><title><![CDATA[Reply to Qwen releases Qwen3.8-Flash-Next! 來了 on Wed, 26 Aug 2026 13:17:11 GMT]]></title><description><![CDATA[<p dir="auto"><a class="plugin-mentions-user plugin-mentions-a" href="/user/kos-or" aria-label="Profile: kos-or">@<bdi>kos-or</bdi></a></p>
<p dir="auto">显存不够啊，只够Q2量化的，想想都没动力去部署了<img src="https://lcz.me/assets/plugins/nodebb-plugin-emoji/emoji/android/1f602.png?v=2fb7360d8c6" class="not-responsive emoji emoji-android emoji--joy" style="height:23px;width:auto;vertical-align:middle" title="😂" alt="😂" /></p>
]]></description><link>https://lcz.me/post/14141</link><guid isPermaLink="true">https://lcz.me/post/14141</guid><dc:creator><![CDATA[Tony Wang]]></dc:creator><pubDate>Wed, 26 Aug 2026 13:17:11 GMT</pubDate></item></channel></rss>