#
✕ nick:eiln
#asahi-dev · 2023-03-28 14:17
eilnlike i said, i've been eyeing the thread, and once reviewed/staged/stable i'll send a patch
#asahi-dev · 2023-03-28 14:17
eilnthis is less about email, but it sure doesn't help
#asahi-dev · 2023-03-28 14:17
eilnmaz: of course some unaffiliated normie would feel out of place in disturbing a massively corpo & proprietary operation (Cloud AI chip btw, WTF?)
#asahi-dev · 2023-03-28 13:58
eilnmaz: i feel the opposite. the qaic patch is at v5(!) now, and i have no idea of its review status. only qualcomm would know if v5 gets scraped again.
#asahi-dev · 2023-03-28 13:42
eilnmaz: ill send a patch if it gets merged.
#asahi-dev · 2023-03-28 13:30
eilnbut i *really* dont want to join a @quicinc.com thread
#asahi-dev · 2023-03-28 13:29
eilnthat they're comparing a ulong to int ret in find_and_map_user_pages()
#asahi-dev · 2023-03-28 13:29
eilnlike i wanted to tell https://lore.kernel.org/all/1679932497-30277-5-git-send-email-quic_jhugo@quicinc.com
#asahi-dev · 2023-03-28 13:29
eilnemail terrifies me
#asahi-dev · 2023-03-28 11:54
eilnthank you
#asahi-dev · 2023-03-28 11:53
eilnill add that to the adt wiki
#asahi-dev · 2023-03-28 11:51
eilnthank you!
#asahi-dev · 2023-03-28 11:49
eilnthey're not in preboot and i dont want to install ventura :(
#asahi-dev · 2023-03-28 11:49
eilnis there an easy way to get M2 device trees?
#asahi-dev · 2023-03-26 05:52
eilnmarcan: mind CONFIG_DRM_ACCEL in the next release? :)
#asahi-dev · 2023-03-25 23:08
eilnso much for getting off that site /s
#asahi-dev · 2023-03-25 23:07
eilnso that's why i was getting emails at midnight https://news.ycombinator.com/item?id=35301447
#asahi-dev · 2023-03-24 15:46
eilnlolz i should also get off that site
#asahi-dev · 2023-03-24 15:46
eiln"And for local model development you don't need your rig to be powerful as training wont be done on your local machine anyway and MacBooks are good enough for that (a lot better than windows for sure). "
#asahi-dev · 2023-03-24 15:45
eilni clicked on their profile out of fascination and their comment just before that is:
#asahi-dev · 2023-03-24 15:45
eilnit's so specific it's believable, like gpt speak
#asahi-dev · 2023-03-24 15:45
eilnactually the whole point of the compiler is to chuck everything in one go
#asahi-dev · 2023-03-24 15:45
eilncoreml would agree with that; there's a size threshold to get compiled on the ane
#asahi-dev · 2023-03-24 15:45
eilnit's the opposite case too, large i/o takes advantage of locality
#asahi-dev · 2023-03-24 15:44
eilnlol ive got a 2048x2048 rgb output in front of me
#asahi-dev · 2023-03-24 15:44
eilnsome hn comment says " Its FP16/Int8 inference only (cause you can only access it via apple frameworks that dosent support training). Also its only used if your data is small enough (4mb cache) it wont be useful for big transformers/ big images processing in a while."
#asahi-dev · 2023-03-24 15:44
eilnalso your mastodon post reminded me
#asahi-dev · 2023-03-24 15:44
eilnmarcan: PR'd in m1n1
#asahi-dev · 2023-03-24 07:07
eilnif any t6000 (esp ultra) can read & send a trace that'd be awesome
#asahi-dev · 2023-03-24 07:07
eilnmarcan: im gonna dump the ane device tree mess over at #asahi-re
#asahi-dev · 2023-03-22 06:44
eilnfuck that was loud
#asahi-dev · 2023-03-22 06:44
eilnit may be related to suspend, happened a few moments after pausing audio
#asahi-dev · 2023-03-22 05:10
eilnit's faintly beeping now, should reboot
#asahi-dev · 2023-03-22 05:09
eilnsecond time today, is this a known issue?
#asahi-dev · 2023-03-22 05:09
eiln[ 4640.615406] i2c-apple 235018000.i2c: Issuing reset due to status 0x08210100 (xfstatus 0x00000000)
#asahi-dev · 2023-03-22 05:09
eiln[ 4640.615280] i2c-apple 235018000.i2c: Issuing reset due to status 0x08210100 (xfstatus 0x00000000)
#asahi-dev · 2023-03-22 05:09
eiln[ 4640.615261] cs42l83 2-0048: ASoC: error at snd_soc_component_update_bits on cs42l83.2-0048 for register: [0x00002001] -6
#asahi-dev · 2023-03-22 05:08
eilnah fuck my earphones just started distorting then bleeping
#asahi-dev · 2023-03-20 03:38
eilnmarcan: that and some ML coreml stuff on macos (non hypervisor). after this release sounds good though.
#asahi-dev · 2023-03-20 01:30
eilnneed some t6000 coreml samples
#asahi-dev · 2023-03-20 01:30
eilnmarcan: nothing urgent, but is there a way to request help? akin to the yak page
#asahi-dev · 2023-03-14 03:06
eilnmarcan: got it
#asahi-dev · 2023-03-14 00:07
eilnmarcan: created a pull in m1n1 with the ane tunables
#asahi-dev · 2023-03-11 09:04
eilnhttps://github.com/eiln/aneex
#asahi-dev · 2023-03-11 09:04
eilncode's copy-pasted from the compute examples i pushed yesterday
#asahi-dev · 2023-03-11 09:04
eilnalso should get faster once i get to hardware queue sched
#asahi-dev · 2023-03-11 09:04
eilncompared to https://github.com/ggerganov/whisper.cpp/discussions/548#discussioncomment-5212716 which are MBP numbers
#asahi-dev · 2023-03-11 09:04
eilnshaves off a solid 0.01s too
#asahi-dev · 2023-03-11 09:04
eilnhell fucking yeah it runs https://bashify.io/images/0Y7fdf
#asahi-dev · 2023-03-11 08:36
eilnthey all convert successfully.. lets see if it runs

plus ›