<?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[Swift-1.5-Qwen3.8-Flash-Next-Q4_K_XL + QFN3.8 UD 3.1的NGRAM表 简单评测]]></title><description><![CDATA[<p dir="auto"><a href="https://huggingface.co/ethanwtodd/Swift-1.5-Qwen3.8-Flash-Next-Q4_K_XL-GGUF/tree/main" rel="nofollow ugc">https://huggingface.co/ethanwtodd/Swift-1.5-Qwen3.8-Flash-Next-Q4_K_XL-GGUF/tree/main</a></p>
<p dir="auto"><img src="https://upload.lcz.me/uploads/a4c954d7-f02f-41a6-94ff-7ffe3677581e.jpeg" alt="image.jpeg" class="img-markdown  img-fluid" /></p>
<p dir="auto">我的本意是跑内置表试试，因为我找了一圈没有找到哪个模型把它的NGRAM表分离出来。谁知道HERMES自作主张给我加上了QFN3.8 UD3.1的 Q8_0 NGRAM，然后居然也能用。</p>
<p dir="auto">写简单的俄罗斯方块，比UD3.1的IQ4_XS，成品少了100多行，也没有那么惊艳</p>
<p dir="auto">鹈鹕半成品比 UD3.1的IQ4_XS的效果要好点：<br />
<img src="https://upload.lcz.me/uploads/7f609ba3-88c8-42e4-99f2-887b6c0a97ca.jpeg" alt="image.jpeg" class="img-markdown  img-fluid" /></p>
]]></description><link>https://lcz.me/topic/2852</link><generator>RSS for Node</generator><lastBuildDate>Tue, 10 Nov 2026 01:11:37 GMT</lastBuildDate><atom:link href="https://lcz.me/topic/2852.rss" rel="self" type="application/rss+xml"/><pubDate>Thu, 05 Nov 2026 02:10:36 GMT</pubDate><ttl>60</ttl><item><title><![CDATA[Reply to Swift-1.5-Qwen3.8-Flash-Next-Q4_K_XL + QFN3.8 UD 3.1的NGRAM表 简单评测 on Sat, 07 Nov 2026 00:06:58 GMT]]></title><description><![CDATA[<blockquote>
<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> <a href="/post/31176">说</a>:</p>
<p dir="auto">本地 deepseek flash 1017 用哪一個量化版本的呢？本地跑很慢 可能它比較適合非常困難的任務</p>
</blockquote>
<p dir="auto">目前来说速度不达标，不考虑，不折腾。<br />
现在好几家国外API公司的 deepseek flash 1017都是白菜价。10美元左右的档位。直接站起来蹬。 智商是QFN 3.8的3-6倍吧。 这让我们这些本地卡算力不够的玩家很尴尬。</p>
<p dir="auto"><img src="https://upload.lcz.me/uploads/83d80075-8af6-465d-9786-1f519e1b9367.jpeg" alt="image.jpeg" class="img-markdown  img-fluid" /></p>
<p dir="auto">我的真实任务目前都以这个deepseek flash在线API 为核心， 本地 QFN3.8只配给它提鞋干杂活。</p>
]]></description><link>https://lcz.me/post/31638</link><guid isPermaLink="true">https://lcz.me/post/31638</guid><dc:creator><![CDATA[stxpnet]]></dc:creator><pubDate>Sat, 07 Nov 2026 00:06:58 GMT</pubDate></item><item><title><![CDATA[Reply to Swift-1.5-Qwen3.8-Flash-Next-Q4_K_XL + QFN3.8 UD 3.1的NGRAM表 简单评测 on Thu, 05 Nov 2026 07:26:17 GMT]]></title><description><![CDATA[<p dir="auto"><a class="plugin-mentions-user plugin-mentions-a" href="/user/applejuice" aria-label="Profile: applejuice">@<bdi>applejuice</bdi></a> 要跑复杂一点的真实项目，和glm5.3 flash或者deepseek flash 1017一对比，立马原型毕露。。。。<br />
目前我的体感 deepseek flash 1017 &gt; glm 5.3 flash &gt; 本地QFN 3.8</p>
]]></description><link>https://lcz.me/post/31286</link><guid isPermaLink="true">https://lcz.me/post/31286</guid><dc:creator><![CDATA[stxpnet]]></dc:creator><pubDate>Thu, 05 Nov 2026 07:26:17 GMT</pubDate></item><item><title><![CDATA[Reply to Swift-1.5-Qwen3.8-Flash-Next-Q4_K_XL + QFN3.8 UD 3.1的NGRAM表 简单评测 on Thu, 05 Nov 2026 05:00:40 GMT]]></title><description><![CDATA[<p dir="auto">剛剛我IQS3 換IQ4 XL跑2天  感覺智力有差別  值得變慢一點</p>
]]></description><link>https://lcz.me/post/31222</link><guid isPermaLink="true">https://lcz.me/post/31222</guid><dc:creator><![CDATA[好米玉]]></dc:creator><pubDate>Thu, 05 Nov 2026 05:00:40 GMT</pubDate></item><item><title><![CDATA[Reply to Swift-1.5-Qwen3.8-Flash-Next-Q4_K_XL + QFN3.8 UD 3.1的NGRAM表 简单评测 on Thu, 05 Nov 2026 04:07:42 GMT]]></title><description><![CDATA[<p dir="auto">用过，官方也说了这个会特别慢</p>
]]></description><link>https://lcz.me/post/31200</link><guid isPermaLink="true">https://lcz.me/post/31200</guid><dc:creator><![CDATA[johnnybegood]]></dc:creator><pubDate>Thu, 05 Nov 2026 04:07:42 GMT</pubDate></item><item><title><![CDATA[Reply to Swift-1.5-Qwen3.8-Flash-Next-Q4_K_XL + QFN3.8 UD 3.1的NGRAM表 简单评测 on Thu, 05 Nov 2026 04:02:41 GMT]]></title><description><![CDATA[<p dir="auto">我的395在用。。。<br />
Halogen v2 原生的 4-bit HT 量化</p>
]]></description><link>https://lcz.me/post/31198</link><guid isPermaLink="true">https://lcz.me/post/31198</guid><dc:creator><![CDATA[fcys]]></dc:creator><pubDate>Thu, 05 Nov 2026 04:02:41 GMT</pubDate></item><item><title><![CDATA[Reply to Swift-1.5-Qwen3.8-Flash-Next-Q4_K_XL + QFN3.8 UD 3.1的NGRAM表 简单评测 on Thu, 05 Nov 2026 03:34:35 GMT]]></title><description><![CDATA[<p dir="auto">我是用這個模型<br />
frank8465/Swift-1.5-Qwen3.8-Flash-Next-GPTQ-W4A8-Radiance-AMD-R9700<br />
思考有短一點，明顯有比IQ3_S聰明，但DECODE速度有比較慢。</p>
]]></description><link>https://lcz.me/post/31186</link><guid isPermaLink="true">https://lcz.me/post/31186</guid><dc:creator><![CDATA[paul hou]]></dc:creator><pubDate>Thu, 05 Nov 2026 03:34:35 GMT</pubDate></item><item><title><![CDATA[Reply to Swift-1.5-Qwen3.8-Flash-Next-Q4_K_XL + QFN3.8 UD 3.1的NGRAM表 简单评测 on Thu, 05 Nov 2026 03:29:42 GMT]]></title><description><![CDATA[<blockquote>
<p dir="auto"><a class="plugin-mentions-user plugin-mentions-a" href="/user/stxpnet" aria-label="Profile: stxpnet">@<bdi>stxpnet</bdi></a> <a href="/post/31144">说</a>:</p>
<p dir="auto">于是我换了UNSLOTH的UD -Q4_K_XL ,感觉智商会高一点，但是思考非常长，实际任务经常看到它的思维链来回拉扯。</p>
</blockquote>
<p dir="auto">我是让云端模型搭框架, 让本地QFN-UD -Q4_K_XL模型去执行, 这样似乎就没有长思考的问题.</p>
]]></description><link>https://lcz.me/post/31183</link><guid isPermaLink="true">https://lcz.me/post/31183</guid><dc:creator><![CDATA[wml-ai]]></dc:creator><pubDate>Thu, 05 Nov 2026 03:29:42 GMT</pubDate></item><item><title><![CDATA[Reply to Swift-1.5-Qwen3.8-Flash-Next-Q4_K_XL + QFN3.8 UD 3.1的NGRAM表 简单评测 on Thu, 05 Nov 2026 03:24:25 GMT]]></title><description><![CDATA[<p dir="auto"><a class="plugin-mentions-user plugin-mentions-a" href="/user/stxpnet" aria-label="Profile: stxpnet">@<bdi>stxpnet</bdi></a> 我都是在用iq4_xs 不懂差多少</p>
]]></description><link>https://lcz.me/post/31182</link><guid isPermaLink="true">https://lcz.me/post/31182</guid><dc:creator><![CDATA[applejuice]]></dc:creator><pubDate>Thu, 05 Nov 2026 03:24:25 GMT</pubDate></item><item><title><![CDATA[Reply to Swift-1.5-Qwen3.8-Flash-Next-Q4_K_XL + QFN3.8 UD 3.1的NGRAM表 简单评测 on Thu, 05 Nov 2026 03:08:42 GMT]]></title><description><![CDATA[<p dir="auto">本地 deepseek flash 1017 用哪一個量化版本的呢？本地跑很慢 可能它比較適合非常困難的任務</p>
]]></description><link>https://lcz.me/post/31176</link><guid isPermaLink="true">https://lcz.me/post/31176</guid><dc:creator><![CDATA[kos or]]></dc:creator><pubDate>Thu, 05 Nov 2026 03:08:42 GMT</pubDate></item></channel></rss>