10-12-2026, 02:24 PM
That matches what I was seeing. The pattern around prefill/decode split has been pretty similar for me.
|
70B still worth the pain with llama.cpp
|
|
« Next Oldest | Next Newest »
|
| Messages In This Thread |
|
70B still worth the pain with llama.cpp - by batchscope72 - 10-12-2026, 04:48 AM
Re: 70B still worth the pain with llama.cpp - by stackarch - 10-12-2026, 05:53 AM
Re: 70B still worth the pain with llama.cpp - by phasescope - 10-12-2026, 02:24 PM
|