context
130
ctx:discord/blah/general/130Source document
full textgeneral-130
text/plain3 KB
doc:agent/general-130/30495226-3aae-4719-ab1b-a289b0d8db17[2026-04-11 22:09] girvo: >British Empire Displacement well that makes a lot of sense. bloody british [2026-04-11 22:48] lisamegawatts: since the small world model shows that any group of 6 people can be chained to reach one another and the ancestry data of 6 people are being analyzed, solve the graph equation to find a genuine connection [2026-04-11 22:58] girvo: unified memory is so neat when you can process it in a massively parallel way (files: image.png) [2026-04-11 22:59] ajaxdavis: we still on meet, come share what you did [2026-04-11 23:01] girvo: making coffee for the missus, but i'll be around later! i'm just playing with CUDA on the GB10 to see what it's like :3 [2026-04-11 23:51] ajaxdavis: The Sisterhood of the Traveling Pants [2026-04-12 02:54] girvo: https://huggingface.co/MiniMaxAI/MiniMax-M2.7 oh here we go! M2.7 is honestly very impressive in my testing, and now it's open weight 😮 [2026-04-12 06:15] ajaxdavis: can you get your model working by tomorrow or so, used so much of my weekly limit % doing genealogy [2026-04-12 07:52] girvo: man I wish I could fit M2.7 locally. maybe one day, if I cluster more of these things [2026-04-12 07:56] girvo: Though, at 229B params its actually _smaller_ than I expected for such a high quality model [2026-04-13 09:39] ajaxdavis: https://github.com/anthropics/claude-code/issues/45756#issuecomment-4231739206 [2026-04-13 09:39] ajaxdavis: might drop my max tokens down anyway, barelly had a problem with 200k [2026-04-13 11:16] girvo: JFC, I knew something was up [2026-04-13 19:05] lisamegawatts: i find using 1m context works well and there is no use in compacting that, it loses everything and is cheaper to start fresh. took a cue from <@187062569415081984> and have the agent write a handoff doc at end of session [2026-04-13 19:35] xenonfun: I did watching it get progressively dumber/less dillegent yesterday, and its been pretty obvious they been playing with cache experation (I thought more then just lowering default TTL I assumed their own backend issues might be regularly losing parts of cache, plus more peopel using 1M and thus higher evictions.) I can kinda live within a 1hr cache window, but 5m no. [2026-04-13 19:36] xenonfun: codex has a handoff button in the desktop UI [2026-04-13 21:47] lisamegawatts: what happens if you fork? i just forked a full context one that ran out before i got the handoff [2026-04-13 22:59] xenonfun: I haven't been doing that, pretty sure its a brand new ~800K context write. had been using /btw a fair bit, which I assume uses the cached context but throws away response [2026-04-14 00:04] xenonfun: JANG version working pretty good both Gemma4 and MiniMax (files: Screenshot_2026-04-13_at_8.01.58_PM.png, Screenshot_2026-04-13_at_8.03.45_PM.png) [2026-04-14 22:11] girvo: Talk me out of buying a second GX10 so I can run MiniMax M2.7 at home lol
Facts in this context
Grouped by subject. Each subject links to its full article.
Girvo18 factsex:girvo
| associatedFile | files: image.png |
| commentedOnSize | actually _smaller_ than I expected |
| expressedDesire | fit M2.7 locally |
| expressedOpinion | M2.7 is honestly very impressive |
| expressedOpinionAbout | British Empire |
| expressedRealization | I knew something was up |
| mentioned | British Empire Displacement |
| mentioned | the missus |
| mentioned | CUDA |
| mentionedFutureCondition | cluster more of these things |
| mentionedHardware | GB10 |
| performedAction | making coffee |
| praised | unified memory |
| rdfs:label | girvo |
| rdf:type |