summaryrefslogtreecommitdiff
path: root/docker-compose.yml
diff options
context:
space:
mode:
authorFelix Morgner <felix.morgner@gmail.com>2026-09-18 11:03:57 +0200
committerFelix Morgner <felix.morgner@gmail.com>2026-09-18 11:03:57 +0200
commitc6a1c53abe3f9d059a62b2eef9b81ed40a33515f (patch)
tree95244f541ce9b0a2da5f468877ad83713b35499b /docker-compose.yml
parent9496b3ade6bdaa77b58dfd1ac4b3ed89c5067834 (diff)
downloadlocal-ai-c6a1c53abe3f9d059a62b2eef9b81ed40a33515f.tar.xz
local-ai-c6a1c53abe3f9d059a62b2eef9b81ed40a33515f.zip
llama.cpp: disable mtp
Diffstat (limited to 'docker-compose.yml')
-rw-r--r--docker-compose.yml9
1 files changed, 2 insertions, 7 deletions
diff --git a/docker-compose.yml b/docker-compose.yml
index 8079b2a..1b58f68 100644
--- a/docker-compose.yml
+++ b/docker-compose.yml
@@ -4,18 +4,13 @@ services:
restart: always
command: >-
-m /models/gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf
- -md /models/mtp-gemma-4-26B-A4B-it.gguf
- -mm /models/mmproj-BF16.gguf
--host 0.0.0.0
--port 8080
--api-key sk-7bbd5413f34151830df2ff33b4956ec10bb8029d73a6d53d3da30a59640b3232
- --ctx-size 131072
- -ngl 14
- -ngld -1
+ --ctx-size 262144
+ -ngl 12
--jinja
--alias "Gemma 4"
- --spec-type draft-mtp
- --spec-draft-n-max 4
volumes:
- type: bind
source: ./inference/models