misc/ollama: update 0.3.4 → 0.3.6

6 files changed, 67 insertions, 17 deletions

diff --git a/misc/ollama/Makefile b/misc/ollama/Makefile
index 1995935ddfae..be457b2db084 100644
--- a/misc/ollama/Makefile
+++ b/misc/ollama/Makefile
@@ -1,7 +1,6 @@
 PORTNAME=	ollama
 DISTVERSIONPREFIX=	v
-DISTVERSION=	0.3.4
-PORTREVISION=	4
+DISTVERSION=	0.3.6
 CATEGORIES=	misc # machine-learning
 
 MAINTAINER=	yuri@FreeBSD.org
@@ -28,11 +27,13 @@ GO_MODULE=	github.com/${PORTNAME}/${PORTNAME}
 GO_TARGET=	.
 
 USE_GITHUB=	nodefault
-GH_TUPLE=	ggerganov:llama.cpp:6eeaeba:llama_cpp/llm/llama.cpp
+GH_TUPLE=	ggerganov:llama.cpp:1e6f6554aa11fa10160a5fda689e736c3c34169f:llama_cpp/llm/llama.cpp \
+		blabber:go-freebsd-sysctl:503969f:go_sysctl/vendor.x/github.com/blabber/go-freebsd-sysctl
 
 MAKE_ENV=	PATH=${PATH}:${WRKSRC}/llm/build/bsd/x86_64_static/bin # workaround to find vulkan-shaders-gen
 
-PLIST_FILES=	bin/${PORTNAME}
+PLIST_FILES=	bin/${PORTNAME} \
+		bin/ollama-limit-layers
 
 post-patch: # workaround for https://github.com/ollama/ollama/issues/6259 (use of extenral libllama.so)
 	@${REINPLACE_CMD} \
@@ -42,6 +43,8 @@ post-patch: # workaround for https://github.com/ollama/ollama/issues/6259 (use o
 			s| ggml | ${WRKSRC}/llm/build/bsd/x86_64_static/ggml/src/libggml.a |; \
 		' \
 		${WRKSRC}/llm/ext_server/CMakeLists.txt
+	# move vendor.x to vendor
+	@(cd ${WRKSRC}/vendor.x && ${TAR} cf - .) | (cd ${WRKSRC}/vendor && ${TAR} xf -)
 
 pre-build:
 	@${CP} ${WRKSRC}/app/store/store_linux.go ${WRKSRC}/app/store/store_bsd.go
@@ -49,4 +52,7 @@ pre-build:
 		${SETENVI} ${WRK_ENV} ${MAKE_ENV} ${GO_ENV} GOMAXPROCS=${MAKE_JOBS_NUMBER} GOPROXY=off ${GO_CMD} generate ${GO_BUILDFLAGS} \
 			./...
 
+post-install: # pending https://github.com/ollama/ollama/issues/6407
+	${INSTALL_SCRIPT} ${FILESDIR}/ollama-limit-layers ${STAGEDIR}${PREFIX}/bin
+
 .include <bsd.port.mk>
diff --git a/misc/ollama/distinfo b/misc/ollama/distinfo
index c436195aa415..86cae5d113d3 100644
--- a/misc/ollama/distinfo
+++ b/misc/ollama/distinfo
@@ -1,7 +1,9 @@
-TIMESTAMP = 1722929347
-SHA256 (go/misc_ollama/ollama-v0.3.4/v0.3.4.mod) = ec7b6ce8fd820503671d269fd43ea5ae7275b1ff095c77d6422ac8540ccb410b
-SIZE (go/misc_ollama/ollama-v0.3.4/v0.3.4.mod) = 2992
-SHA256 (go/misc_ollama/ollama-v0.3.4/v0.3.4.zip) = d9bc3d3901272f8fbc9b57ce1af1caa2209127978bb5f72257b6a3fb84161969
-SIZE (go/misc_ollama/ollama-v0.3.4/v0.3.4.zip) = 1832031
-SHA256 (go/misc_ollama/ollama-v0.3.4/ggerganov-llama.cpp-6eeaeba_GH0.tar.gz) = f749f494dc5e672926179524cf28c834151edba5acae18e132c954df76ffdf33
-SIZE (go/misc_ollama/ollama-v0.3.4/ggerganov-llama.cpp-6eeaeba_GH0.tar.gz) = 19003544
+TIMESTAMP = 1724010094
+SHA256 (go/misc_ollama/ollama-v0.3.6/v0.3.6.mod) = 16c078d8f0b29f84598fb04e3979acf86da41eb41bf4ff8363548e490f38b54e
+SIZE (go/misc_ollama/ollama-v0.3.6/v0.3.6.mod) = 2992
+SHA256 (go/misc_ollama/ollama-v0.3.6/v0.3.6.zip) = 94d2376c9555dd03a76cb093f3213e8155811874db7eab6aff2941d0e75dce07
+SIZE (go/misc_ollama/ollama-v0.3.6/v0.3.6.zip) = 1842735
+SHA256 (go/misc_ollama/ollama-v0.3.6/ggerganov-llama.cpp-1e6f6554aa11fa10160a5fda689e736c3c34169f_GH0.tar.gz) = b2b3137f734fc9a202fac710986f1de837e0ae69a0b532d4cbadb90748f4cb73
+SIZE (go/misc_ollama/ollama-v0.3.6/ggerganov-llama.cpp-1e6f6554aa11fa10160a5fda689e736c3c34169f_GH0.tar.gz) = 19016896
+SHA256 (go/misc_ollama/ollama-v0.3.6/blabber-go-freebsd-sysctl-503969f_GH0.tar.gz) = 1f497780d88f6f056b8d8f59b6aa129566c8041d16d1cda89f212accf88f3165
+SIZE (go/misc_ollama/ollama-v0.3.6/blabber-go-freebsd-sysctl-503969f_GH0.tar.gz) = 2117
diff --git a/misc/ollama/files/ollama-limit-layers b/misc/ollama/files/ollama-limit-layers
new file mode 100755
index 000000000000..6713a9f9b02b
--- /dev/null
+++ b/misc/ollama/files/ollama-limit-layers
@@ -0,0 +1,13 @@
+#!/bin/sh
+
+MODEL=$1
+NUM_GPU=$2
+
+if [ -z "$MODEL" ] || [ -z "$NUM_GPU" ]; then
+	echo "Usage: $0 <model> <num_gpu>"
+fi
+
+ollama show --modelfile $MODEL > Modelfile &&
+echo "PARAMETER num_gpu $NUM_GPU" >> Modelfile &&
+ollama create "$MODEL-num_gpu$NUM_GPU" -f Modelfile &&
+echo "model variant $MODEL-num_gpu$NUM_GPU was created"
diff --git a/misc/ollama/files/patch-FreeBSD-compatibility b/misc/ollama/files/patch-FreeBSD-compatibility
index ca206363f042..046ca9b3417c 100644
--- a/misc/ollama/files/patch-FreeBSD-compatibility
+++ b/misc/ollama/files/patch-FreeBSD-compatibility
@@ -1,14 +1,19 @@
 -- patch based on https://github.com/ollama/ollama/issues/1102#issuecomment-2270042340
 
 new file mode 100644
---- gpu/gpu_bsd.go.orig	2024-08-06 07:41:26 UTC
+--- gpu/gpu_bsd.go.orig	2024-08-18 20:03:12 UTC
 +++ gpu/gpu_bsd.go
-@@ -0,0 +1,101 @@
+@@ -0,0 +1,122 @@
 +//go:build dragonfly || freebsd || netbsd || openbsd
 +
 +package gpu
 +
 +import "github.com/ollama/ollama/format"
++//import sysctl "github.com/lorenzosaino/go-sysctl" // sysctl: this is Linux-only, see https://github.com/lorenzosaino/go-sysctl/issues/7
++import sysctl "github.com/blabber/go-freebsd-sysctl/sysctl" // sysctl: this is FreeBSD-only basic library
++import (
++	"log/slog"
++)
 +
 +/*
 +#cgo CFLAGS: -I/usr/local/include
@@ -98,8 +103,24 @@ new file mode 100644
 +}
 +
 +func GetCPUMem() (memInfo, error) {
-+	size := C.sysconf(C._SC_PHYS_PAGES) * C.sysconf(C._SC_PAGE_SIZE)
-+	return memInfo{TotalMemory: uint64(size)}, nil
++	// all involved sysctl variables
++	sysctl_vm_page_size, _ := sysctl.GetInt64("vm.stats.vm.v_page_size") // memory page size
++	sysctl_hw_physmem, _ := sysctl.GetInt64("hw.physmem") // physical memory in bytes
++	sysctl_vm_free_count, _ := sysctl.GetInt64("vm.stats.vm.v_free_count") // free page count
++	sysctl_vm_swap_total, _ := sysctl.GetInt64("vm.swap_total") // total swap size in bytes
++
++	// individual values
++	total_memory := uint64(sysctl_hw_physmem)
++	free_memory := uint64(sysctl_vm_free_count) * uint64(sysctl_vm_page_size)
++	free_swap := uint64(sysctl_vm_swap_total) // wrong to use the total swap size here, should be vm.swap_free, see https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=280909
++
++	slog.Debug("gpu_bsd.go::GetCPUMem::GetCPUMem", "total_memory", total_memory, "free_memory", free_memory, "free_swap", free_swap)
++
++	return memInfo{
++		TotalMemory: uint64(total_memory),
++		FreeMemory: uint64(free_memory),
++		FreeSwap: uint64(free_swap),
++	}, nil
 +}
 +
 +func (l GpuInfoList) GetVisibleDevicesEnv() (string, string) {
diff --git a/misc/ollama/files/patch-llm_llama.cpp_ggml_src_vulkan-shaders_CMakeLists.txt b/misc/ollama/files/patch-llm_llama.cpp_ggml_src_vulkan-shaders_CMakeLists.txt
index 3ada7c2debc2..a9ef6bec163e 100644
--- a/misc/ollama/files/patch-llm_llama.cpp_ggml_src_vulkan-shaders_CMakeLists.txt
+++ b/misc/ollama/files/patch-llm_llama.cpp_ggml_src_vulkan-shaders_CMakeLists.txt
@@ -1,9 +1,10 @@
---- llm/llama.cpp/ggml/src/vulkan-shaders/CMakeLists.txt.orig	2024-08-08 21:55:59 UTC
+--- llm/llama.cpp/ggml/src/vulkan-shaders/CMakeLists.txt.orig	2024-08-06 15:33:39 UTC
 +++ llm/llama.cpp/ggml/src/vulkan-shaders/CMakeLists.txt
-@@ -1,5 +1,6 @@ add_executable(${TARGET} vulkan-shaders-gen.cpp)
+@@ -2,6 +2,7 @@ add_executable(${TARGET} vulkan-shaders-gen.cpp)
  
  set(TARGET vulkan-shaders-gen)
  add_executable(${TARGET} vulkan-shaders-gen.cpp)
 +target_link_libraries(${TARGET} PRIVATE pthread)
  install(TARGETS ${TARGET} RUNTIME)
  target_compile_features(${TARGET} PRIVATE cxx_std_11)
+ target_link_libraries(vulkan-shaders-gen PUBLIC Threads::Threads)
diff --git a/misc/ollama/pkg-message b/misc/ollama/pkg-message
index 90096ac82cef..95b0d4e58448 100644
--- a/misc/ollama/pkg-message
+++ b/misc/ollama/pkg-message
@@ -15,6 +15,13 @@ You will be able to interact with it in plain English.
 Please see https://ollama.com/library for the list
 of all supported models.
 
+The command "ollama list" lists all models downloaded
+into your system.
+
+When the model fails to load into your GPU, please use
+the provided ollama-limit-layers script to create
+model flavors with different num_gpu parameters.
+
 ollama uses many gigbytes of disk space in your home directory,
 because advanced AI models are often very large.
 Pease symlink ~/.ollama to a large disk if needed.