version bump sci-ml/ollama-0.40.4

sci-ml/ollama/files/ollama-0.31.1-gcc17.patch

@@ -0,0 +1,10 @@
+--- a/common/common.h	2026-07-04 12:25:21.565559439 +0200
++++ b/common/common.h	2026-07-04 12:25:43.052343559 +0200
+@@ -8,6 +8,7 @@
+ #include "ggml.h"
+ 
+ #include <set>
++#include <fstream>
+ #include <sstream>
+ #include <string>
+ #include <string_view>

sci-ml/ollama/files/ollama-0.31.1-ggml.patch

@@ -0,0 +1,11 @@
+--- a/llama/server/CMakeLists.txt	2026-07-04 07:08:27.528453937 +0200
++++ b/llama/server/CMakeLists.txt	2026-07-04 07:09:07.889047823 +0200
+@@ -508,7 +508,7 @@
+     # RUNTIME covers executables and Windows DLLs; LIBRARY covers .so on Linux
+     install(TARGETS llama-server llama-quantize
+         RUNTIME DESTINATION ${_base_dest} COMPONENT llama-server OPTIONAL)
+-    set(_llama_server_base_libs ggml-base ggml llama mtmd)
++    set(_llama_server_base_libs llama mtmd)
+     if(TARGET llama-common)
+         list(APPEND _llama_server_base_libs llama-common)
+     endif()

sci-ml/ollama/files/ollama-0.34.2-cmake.patch

@@ -0,0 +1,8 @@
+--- a/mlx/CMakeLists.txt	2026-07-04 13:53:35.494218027 +0200
++++ b/mlx/CMakeLists.txt	2026-07-04 13:54:08.428885766 +0200
+@@ -1,4 +1,4 @@
+-cmake_minimum_required(VERSION 3.5)
++cmake_minimum_required(VERSION 3.10)
+ 
+ project(mlx)
+ 

sci-ml/ollama/files/ollama.confd

@@ -0,0 +1,36 @@
+# Ollama allows cross-origin requests from 127.0.0.1 and 0.0.0.0 by default.
+# Additional origins can be configured with OLLAMA_ORIGINS.
+# export OLLAMA_ORIGINS="<ip>"
+# Allow connections from any origin (or specify exact domains/IPs like "http://192.168.1.50:*")
+# export OLLAMA_ORIGINS="*"
+# Explicitly bind to all network interfaces (needed if running behind a proxy or remote API access)
+# export OLLAMA_HOST="0.0.0.0"
+
+# log to syslog
+# output_logger="logger -t \"$RC_SVCNAME\" -p daemon.info"
+# error_logger="logger -t \"$RC_SVCNAME\" -p daemon.err"
+
+# log to file
+output_log="/var/log/ollama/ollama.log"
+error_log="/var/log/ollama/ollama.log"
+
+# Set the default context size - default is 4096 token
+# export OLLAMA_CONTEXT_LENGTH=8192
+# export OLLAMA_CONTEXT_LENGTH=16384
+# export OLLAMA_CONTEXT_LENGTH=32768
+# export OLLAMA_CONTEXT_LENGTH=131072
+# export OLLAMA_CONTEXT_LENGTH=262144
+
+# Optimizations
+#
+# Enable Flash Attention (drastically reduces VRAM usage for long contexts)
+# export OLLAMA_FLASH_ATTENTION=true
+# Compress the Key-Value cache (saves VRAM at a tiny precision cost) default is f16
+# export OLLAMA_KV_CACHE_TYPE=q4_0
+#
+# Number of parallel model requests.
+# export OLLAMA_NUM_PARALLEL=1
+#
+# export OLLAMA_VULKAN=false
+# export OLLAMA_DEBUG=INFO
+# export OLLAMA_DEBUG=1 

sci-ml/ollama/files/ollama.init

@@ -0,0 +1,9 @@
+#!/sbin/openrc-run
+# Copyright 1999-2025 Gentoo Authors
+# Distributed under the terms of the GNU General Public License, v2
+
+supervisor=supervise-daemon
+description="Ollama Service"
+command="/usr/bin/ollama"
+command_args="serve"
+command_user="ollama:ollama"

sci-ml/ollama/files/ollama.service

@@ -0,0 +1,15 @@
+[Unit]
+Description=Ollama Service
+After=network-online.target
+
+[Service]
+ExecStart=/usr/bin/ollama serve
+User=ollama
+Group=ollama
+# GPU access: render -> /dev/dri/renderD* + /dev/kfd (ROCm), video -> legacy/NVIDIA
+SupplementaryGroups=render video
+Restart=always
+RestartSec=3
+
+[Install]
+WantedBy=default.target