\
  The most prestigious law school admissions discussion board in the world.
BackRefresh Options Favorite

Hermes Agent needs at least 48gb of system RAM to breathe just FYI

If you're thinking about fucking with this, know that it wil...
https://imgur.com/a/o2g8xYK
  04/04/26
misread this as hamas agent and then misread hamas agent as ...
Must Poast Always
  04/04/26
https://i.imgur.com/tcwjqbX.jpeg
https://imgur.com/a/o2g8xYK
  04/04/26
~90gb of VRAM occupied but only one GPU is doing work (trans...
https://imgur.com/a/o2g8xYK
  04/04/26
Predict whether it will finish translating this book or run ...
https://imgur.com/a/o2g8xYK
  04/04/26


Poast new message in this thread



Reply Favorite

Date: April 4th, 2026 11:35 AM
Author: https://imgur.com/a/o2g8xYK


If you're thinking about fucking with this, know that it will idle at 3gb, then spike to whatever. Right now it's translating a Russian book for me, with all of the translation being done by a LLM, but Hermes still thinks it needs a > 40gb. I've even seen it go a hair over 48gb once.

https://i.imgur.com/RiiszXn.jpeg

(http://www.autoadmit.com/thread.php?thread_id=5853424&forum_id=2],#49793224)



Reply Favorite

Date: April 4th, 2026 11:36 AM
Author: Must Poast Always (No Future)

misread this as hamas agent and then misread hamas agent as meaning mossad agent

i think jafar's radio waves just hit my pineal gland

(http://www.autoadmit.com/thread.php?thread_id=5853424&forum_id=2],#49793225)



Reply Favorite

Date: April 4th, 2026 11:39 AM
Author: https://imgur.com/a/o2g8xYK


https://i.imgur.com/tcwjqbX.jpeg

(http://www.autoadmit.com/thread.php?thread_id=5853424&forum_id=2],#49793235)



Reply Favorite

Date: April 4th, 2026 11:46 AM
Author: https://imgur.com/a/o2g8xYK


~90gb of VRAM occupied but only one GPU is doing work (translating Russian), so Hermes is effectively using 42gb of system RAM in addition to all that:

https://i.imgur.com/xkI8eF0.jpeg

Yes I could offload more layers onto that 5060 ti doing nothing, but it won't affect how much the agent uses. The 3060 runs nomic-embed-text in a separate container so I can make vector databases

(http://www.autoadmit.com/thread.php?thread_id=5853424&forum_id=2],#49793246)



Reply Favorite

Date: April 4th, 2026 12:42 PM
Author: https://imgur.com/a/o2g8xYK


Predict whether it will finish translating this book or run out of memory:

https://i.imgur.com/bzbYq7D.png

(http://www.autoadmit.com/thread.php?thread_id=5853424&forum_id=2],#49793401)