ollama: local LLM serving with ROCm on Krackan Point (gfx1151)
Enable ollama-rocm bound to 127.0.0.1:11434 with gfx1151 (RDNA 3.5) overrides, flash attention, q8_0 KV cache, and preload qwen3:14b.
This commit is contained in:
parent
a72fb75811
commit
f31ecbf6e6
2 changed files with 18 additions and 0 deletions
|
|
@ -17,6 +17,7 @@
|
|||
./modules/security.nix
|
||||
./modules/power.nix
|
||||
./modules/xray.nix
|
||||
./modules/ollama.nix
|
||||
./modules/gaming.nix
|
||||
];
|
||||
|
||||
|
|
|
|||
17
modules/ollama.nix
Normal file
17
modules/ollama.nix
Normal file
|
|
@ -0,0 +1,17 @@
|
|||
{ config, pkgs, lib, ... }:
|
||||
|
||||
{
|
||||
services.ollama = {
|
||||
enable = true;
|
||||
package = pkgs.ollama-rocm;
|
||||
rocmOverrideGfx = "11.5.1"; # gfx1151, Krackan Point RDNA 3.5
|
||||
host = "127.0.0.1";
|
||||
port = 11434;
|
||||
environmentVariables = {
|
||||
HSA_OVERRIDE_GFX_VERSION = "11.5.1";
|
||||
OLLAMA_FLASH_ATTENTION = "1";
|
||||
OLLAMA_KV_CACHE_TYPE = "q8_0";
|
||||
};
|
||||
loadModels = [ "qwen3:14b" ];
|
||||
};
|
||||
}
|
||||
Loading…
Add table
Add a link
Reference in a new issue