chore: clean up setup

7 files changed, 26 insertions, 71 deletions

diff --git a/.gitignore b/.gitignore
index 560d82b..845ecc6 100644
--- a/.gitignore
+++ b/.gitignore
@@ -1,2 +1 @@
-/inference/models
-/embedding/ollama
+/inference/ollama
diff --git a/docker-compose.yml b/docker-compose.yml
index 1b58f68..bb5046b 100644
--- a/docker-compose.yml
+++ b/docker-compose.yml
@@ -1,65 +1,29 @@
 services:
-  inference:
-    image: ghcr.io/ggml-org/llama.cpp:server-rocm
-    restart: always
-    command: >-
-      -m /models/gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf
-      --host 0.0.0.0
-      --port 8080
-      --api-key sk-7bbd5413f34151830df2ff33b4956ec10bb8029d73a6d53d3da30a59640b3232
-      --ctx-size 262144
-      -ngl 12
-      --jinja
-      --alias "Gemma 4"
-    volumes:
-      - type: bind
-        source: ./inference/models
-        target: /models
-        read_only: false
-    networks:
-      - local-ai
-    ports:
-      - 8080:8080
-    privileged: true
-    shm_size: 16g
-    security_opt:
-      - seccomp:unconfined
-    cap_add:
-      - SYS_PTRACE
-    group_add:
-      - video
-    devices:
-      - /dev/kfd:/dev/kfd
-      - /dev/dri:/dev/dri
 
-  embedding:
-    image: ollama/ollama:latest
+  inference:
+    image: ollama/ollama:rocm
     restart: always
     environment:
-      OLLAMA_MAX_QUEUE: 8192
+      OLLAMA_MAX_QUEUE: 512
+      OLLAMA_MAX_LOADED_MODELS: 1
+      OLLAMA_NUM_PARALLEL: 1
+      OLLAMA_KEEP_ALIVE: 15m
+      OLLAMA_FLASH_ATTENTION: 1
+      OLLAMA_GPU_OVERHEAD: 2147483648
+      OLLAMA_NO_CLOUD: 1
+      LLAMA_ARG_IMAGE_MAX_TOKENS: 1120
     volumes:
       - type: bind
-        source: ./embedding/ollama
+        source: ./inference/ollama
         target: /root/.ollama
         read_only: false
-    networks:
-      - local-ai
+      - type: bind
+        source: ./inference/modelfiles
+        target: /root/modelfiles
+        read_only: false
     ports:
       - 11434:11434
-    privileged: true
-    security_opt:
-      - seccomp:unconfined
-    cap_add:
-      - SYS_PTRACE
-    group_add:
-      - video
     devices:
       - /dev/kfd:/dev/kfd
       - /dev/dri:/dev/dri
-
-networks:
-  local-ai:
-    driver: bridge
-
-volumes:
-  search-cache:
+    shm_size: 4gb
diff --git a/embedding/.gitkeep b/embedding/.gitkeep
deleted file mode 100644
index e69de29..0000000
--- a/embedding/.gitkeep
+++ /dev/null
diff --git a/inference/modelfiles/qwen2.5-coder-128k.Modelfile b/inference/modelfiles/qwen2.5-coder-128k.Modelfile
new file mode 100644
index 0000000..ac6d19c
--- /dev/null
+++ b/inference/modelfiles/qwen2.5-coder-128k.Modelfile
@@ -0,0 +1,3 @@
+FROM qwen2.5-coder:14b
+
+PARAMETER num_ctx 131072
diff --git a/inference/modelfiles/qwen2.5-coder-64k.Modelfile b/inference/modelfiles/qwen2.5-coder-64k.Modelfile
new file mode 100644
index 0000000..09395dd
--- /dev/null
+++ b/inference/modelfiles/qwen2.5-coder-64k.Modelfile
@@ -0,0 +1,6 @@
+FROM qwen2.5-coder:14b
+
+PARAMETER num_ctx 65536
+
+SYSTEM """You are a helpful assistant with access to tools. When you decide to call a tool, you MUST output ONLY the raw JSON object matching the tool call schema. Do NOT wrap the JSON block in markdown code blocks like ```json or prefix it with any text. Output the JSON object directly."""
+
diff --git a/searxng/config/limiter.toml b/searxng/config/limiter.toml
deleted file mode 100644
index 95e2dc1..0000000
--- a/searxng/config/limiter.toml
+++ /dev/null
@@ -1,8 +0,0 @@
-[botdetection.ip_limit]
-
-# activate link_token method in the ip_limit method
-link_token = false
-
-[botdetection.ip_lists]
-block_ip = []
-pass_ip = []
diff --git a/searxng/config/settings.yml b/searxng/config/settings.yml
deleted file mode 100644
index dc4946d..0000000
--- a/searxng/config/settings.yml
+++ /dev/null
@@ -1,9 +0,0 @@
-use_default_settings: true
-
-search:
-    formats:
-        - json
-server:
-    secret_key: "ae9bcaa7ce98ba26d5f352a23e97722d93921f42b97525d32e4a9906c9b7a287"
-    limiter: false
-    image_proxy: true