context
43
ctx:discord/blah/random/43Source document
full textrandom-43
text/plain3 KB
doc:agent/random-43/fe0962ed-1d95-4c6d-b094-0ea56c66a694[2026-04-03 13:57] traves_theberge: <@806444151422976035> [2026-04-03 16:39] xenonfun: ollama really has made it easy to launch codex/claude/etc with local inferance. gemma:26b think q4 quant working well. the 31B its near my memory stablity limit but runs fairly fast (MLX native bundling will be better but having issues getting it working, think my python cache needs purged). Better Cascade 2 and Gemma4 very solid looking for 100% local at good clip. (files: Screenshot_2026-04-03_at_12.35.53_PM.png) [2026-04-03 21:44] ajaxdavis: yeah its fucking sick [2026-04-03 22:52] lisamegawatts: https://x.com/i/status/2040067957532049776 [2026-04-11 18:39] xenonfun: (files: AP1GczP-467Qa8rOFMmBYfEisohWl0kQLtskhCoFZ9Gsh08L6VdVY2DSMPSxHQw1620-h2160-s-no-gm.png) [2026-04-11 20:05] lisamegawatts: (files: doraemon-moe-comic.png) [2026-04-14 16:31] xenonfun: `⚠ Large CLAUDE.md will impact performance (41.3k chars > 40.0k) · /memory to edit` - I'm not the one that make the giant claude, why is it treating this like an append only log? [2026-04-15 07:47] xenonfun: (files: Screenshot_2026-04-15_at_3.47.05_AM.png) [2026-04-15 07:48] ajaxdavis: was looking at it earlier, looks like a nice simple way to burn more tokens unknowingly [2026-04-15 07:48] ajaxdavis: seems like a poor way to encroach on the claw/cron types, less control, less visiblity [2026-04-15 07:49] ajaxdavis: if they actually had an observability dashboard i might use it if convenient, "summarise my daily slack" type shit [2026-04-15 07:50] ajaxdavis: but i would need more than their infamous one unmeasurable horizontal bar chart that is more misleading than a windows install progress bar [2026-04-15 07:52] ajaxdavis: yet another token burning thing i didn't ask to be enabled (files: Screenshot_from_2026-04-15_17-52-35.png) [2026-04-15 07:53] ajaxdavis: they should force 50% of their devs to use 20x plan only [2026-04-15 16:54] lisamegawatts: https://youtu.be/2IsR06zf8nQ?si=NUo5AGL6NsQ69jT7 [2026-04-15 16:55] ajaxdavis: did you sleep [2026-04-15 16:55] lisamegawatts: no [2026-04-15 16:55] ajaxdavis: lol ive been sitting in silence doing avanis family tree for 4 hours aha [2026-04-15 16:56] lisamegawatts: lol i got back on call like an hour ago was empty [2026-04-15 16:58] lisamegawatts: rejoined [2026-04-16 15:52] xenonfun: Qwen3.6 8-bit is doing best of anything I have tried so far. (files: Screenshot_2026-04-16_at_11.51.56_AM.png) [2026-04-16 19:37] xenonfun: (files: Screenshot_2026-04-16_at_3.37.17_PM.png) [2026-04-16 19:54] xenonfun: https://lilting.ch/en/articles/ollama-mlx-apple-silicon-preview appearently Ollama is optimized on MLX now, I had been avoid it because last time I checked it was not. [2026-04-17 02:45] ajaxdavis: feels like thats posttrained lol, no way [2026-04-17 04:09] xenonfun: no I just posted the same input that some viral post did, local got real this week. [2026-04-17 04:09] xenonfun: I don't think I even had system prompt in it
Facts in this context
Grouped by subject. Each subject links to its full article.
Xenonfun25 factsex:xenonfun
| claimedCapability | Local Inference Launching |
| deniedResponsibilityFor | Claude Md |
| encounteredIssueWith | Mlx Native Bundling |
| evaluated | Ollama |
| expectsImprovementFrom | Mlx Native Bundling |
| expressedConfusionAbout | Claude Md |
| expressedOptimismAbout | Gemma4 |
| expressedOptimismAbout | Better Cascade2 |
| hardwareConstraint | Memory Stability Limit |
| opinionOn | Ollama |
| parameterConfiguration | Q4 Quantization |
| performanceMetric |