I ordered 2 DGX Sparks for a POC at my company. My reseller notified me this morning that they are officially EOL and that they don't have a direct replacement to sell me at this time. Anyone else seen this?
This must be related to the announcements yesterday of the N1x laptops and desktop.
I constantly get messages like this after a session has gone on for a little while. I see that it sometimes compresses by itself okay. Sometimes I have to go call it myself manually. However, most of the time that doesn't work either, ending with e.g.
· ⠋ compressing 103 messages (~108,084 tok)…
· error: timeout: session.compress
At this point I have to start a new session, which I don't want to / shouldn't have to do.
In ~/.hermes/config.yam, Under `model:`, I have set:
context: 65536
max_tokens: 8192
I've read through docs and chatted back and forth with AI about this, but haven't found good settings to solve this.
EDIT:
Some additional data:
* I have a second model dedicated to context compression (
* Both models are locally hosted on my network (llama-cpp)
This, plus the recent SpaceX announcement, makes the VCX fund very interesting.