2026-10-12T09:01:32Z https://njump.cloud.vinney.xyz Jazzilla https://njump.cloud.vinney.xyz/npub1zdyj8tth570ythwlg7jvazeku87pyu8smae47dq9vkv8txju30tqctfgxu https://static.noauthority.social/accounts/avatars/111/754/417/227/134/390/original/872f03501000127e.jpg https://static.noauthority.social/accounts/avatars/111/754/417/227/134/390/original/872f03501000127e.jpg https://njump.cloud.vinney.xyz/nevent1qqs0h2kjm4egyshdeeylurn7stvfewyr8k8g7hnkyp65cl8ldyrxgtgzyqf5jgadw7neu3wamar6fn5txmslcyns7r0hxhe5q4jesav6tj9avk4pqcn In reply to <a href='/nevent1qqs26pk45mf3ne5as0d9u4hh9wj79y69p2hraaktzzwfug9wfketqxgdy3ucq'>nevent1q…3ucq</a><br/>_________________________<br/><br/>A lot of ppl would say you can&#39;t do anything serious with only 262k context. <br/><br/>That&#39;s true. <br/><br/>But you can practice prompt discipline:<br/><br/>* Watch the message count on the server<br/>* Have it stop and create a session restore packet with a seed prompt when you get a little over half way to failure for your context. <br/>* /compact the session<br/>* Continue and repeat<br/><br/>This replicates big services practice for 1M contexts.<br/><br/><span itemprop="mentions" itemscope itemtype="https://schema.org/Person"><a itemprop="url" href="/npub17trshr98kk3p76hljadj29lf0e8ug5e3l09vw5pe9t8unqeydcjsyq9dsg" class="bg-lavender dark:prose:text-neutral-50 dark:text-neutral-50 dark:bg-garnet px-1"><span>SirCrocky of Cosmic Hi-Country</span> (<span class="italic">npub17tr…9dsg</span>)</a></span> 2026-09-25T22:33:12Z https://njump.cloud.vinney.xyz/nevent1qqswpmm78pduxlen8py7ydp0ha4nldu3wyvak2nezmgdfq2lewjz4aszyqf5jgadw7neu3wamar6fn5txmslcyns7r0hxhe5q4jesav6tj9avptzkq2 In reply to <a href='/nevent1qqsyu4hkmhlyelnvc3wmqs7949t6554f7fnky3ddud2w9jqcgvua8pgm7dt9f'>nevent1q…dt9f</a><br/>_________________________<br/><br/>If you&#39;re just going to start the thing up, do a couple of things and unload it, what&#39;s the point? <br/><br/>GPT Luna is free. <br/><br/><span itemprop="mentions" itemscope itemtype="https://schema.org/Person"><a itemprop="url" href="/npub17trshr98kk3p76hljadj29lf0e8ug5e3l09vw5pe9t8unqeydcjsyq9dsg" class="bg-lavender dark:prose:text-neutral-50 dark:text-neutral-50 dark:bg-garnet px-1"><span>SirCrocky of Cosmic Hi-Country</span> (<span class="italic">npub17tr…9dsg</span>)</a></span> 2026-09-25T22:24:22Z https://njump.cloud.vinney.xyz/nevent1qqsyu4hkmhlyelnvc3wmqs7949t6554f7fnky3ddud2w9jqcgvua8pgzyqf5jgadw7neu3wamar6fn5txmslcyns7r0hxhe5q4jesav6tj9avpxd993 In reply to <a href='/nevent1qqs9899hyqpcxxtcsxpnz90r6r4nykgsfq3t89s9wutg4d3vqk7c3tsken4mr'>nevent1q…n4mr</a><br/>_________________________<br/><br/>The guy is looking for the constraints.<br/><br/>It&#39;s context window. They only address 262k.<br/><br/>So if you&#39;re familiar with the length of a task that would fill that up, and how much vram it would take to physically accomodate that, then there&#39;s the &#34;minimum&#34; vram requirement for serious tasks - that is only part of the story.<br/><br/>You&#39;re running multiple agents and other programs.<br/><br/>Arm soc lets you adjust the vram alloc.<br/><br/><span itemprop="mentions" itemscope itemtype="https://schema.org/Person"><a itemprop="url" href="/npub17trshr98kk3p76hljadj29lf0e8ug5e3l09vw5pe9t8unqeydcjsyq9dsg" class="bg-lavender dark:prose:text-neutral-50 dark:text-neutral-50 dark:bg-garnet px-1"><span>SirCrocky of Cosmic Hi-Country</span> (<span class="italic">npub17tr…9dsg</span>)</a></span> 2026-09-25T22:22:21Z