git: ee1259996679 - main - misc/ollama: Improve pkg-message
- Go to: [ bottom of page ] [ top of archives ] [ this month ]
Date: Mon, 19 Aug 2024 01:12:11 UTC
The branch main has been updated by yuri: URL: https://cgit.FreeBSD.org/ports/commit/?id=ee1259996679f14ac8b59a10d10614d840d89460 commit ee1259996679f14ac8b59a10d10614d840d89460 Author: Yuri Victorovich <yuri@FreeBSD.org> AuthorDate: 2024-08-19 01:11:36 +0000 Commit: Yuri Victorovich <yuri@FreeBSD.org> CommitDate: 2024-08-19 01:12:09 +0000 misc/ollama: Improve pkg-message --- misc/ollama/Makefile | 5 +++-- misc/ollama/files/{ollama-limit-layers => ollama-limit-gpu-layers} | 2 ++ misc/ollama/pkg-message | 6 +++++- 3 files changed, 10 insertions(+), 3 deletions(-) diff --git a/misc/ollama/Makefile b/misc/ollama/Makefile index be457b2db084..724bc72feceb 100644 --- a/misc/ollama/Makefile +++ b/misc/ollama/Makefile @@ -1,6 +1,7 @@ PORTNAME= ollama DISTVERSIONPREFIX= v DISTVERSION= 0.3.6 +PORTREVISION= 1 CATEGORIES= misc # machine-learning MAINTAINER= yuri@FreeBSD.org @@ -33,7 +34,7 @@ GH_TUPLE= ggerganov:llama.cpp:1e6f6554aa11fa10160a5fda689e736c3c34169f:llama_cpp MAKE_ENV= PATH=${PATH}:${WRKSRC}/llm/build/bsd/x86_64_static/bin # workaround to find vulkan-shaders-gen PLIST_FILES= bin/${PORTNAME} \ - bin/ollama-limit-layers + bin/ollama-limit-gpu-layers post-patch: # workaround for https://github.com/ollama/ollama/issues/6259 (use of extenral libllama.so) @${REINPLACE_CMD} \ @@ -53,6 +54,6 @@ pre-build: ./... post-install: # pending https://github.com/ollama/ollama/issues/6407 - ${INSTALL_SCRIPT} ${FILESDIR}/ollama-limit-layers ${STAGEDIR}${PREFIX}/bin + ${INSTALL_SCRIPT} ${FILESDIR}/ollama-limit-gpu-layers ${STAGEDIR}${PREFIX}/bin .include <bsd.port.mk> diff --git a/misc/ollama/files/ollama-limit-layers b/misc/ollama/files/ollama-limit-gpu-layers similarity index 93% rename from misc/ollama/files/ollama-limit-layers rename to misc/ollama/files/ollama-limit-gpu-layers index 6713a9f9b02b..f17cb5622d0b 100755 --- a/misc/ollama/files/ollama-limit-layers +++ b/misc/ollama/files/ollama-limit-gpu-layers @@ -5,9 +5,11 @@ NUM_GPU=$2 if [ -z "$MODEL" ] || [ -z "$NUM_GPU" ]; then echo "Usage: $0 <model> <num_gpu>" + exit 1 fi ollama show --modelfile $MODEL > Modelfile && echo "PARAMETER num_gpu $NUM_GPU" >> Modelfile && ollama create "$MODEL-num_gpu$NUM_GPU" -f Modelfile && +rm Modelfile && echo "model variant $MODEL-num_gpu$NUM_GPU was created" diff --git a/misc/ollama/pkg-message b/misc/ollama/pkg-message index 95b0d4e58448..936a3494000a 100644 --- a/misc/ollama/pkg-message +++ b/misc/ollama/pkg-message @@ -12,6 +12,10 @@ To run ollama, plese open 2 terminals. This will download and run the AI model "mistral". You will be able to interact with it in plain English. +If the above fails, you can try to run 'ollama start' +with theenvironment variable OLLAMA_NUM_PARALLEL=1. +This is known to help with smaller models. + Please see https://ollama.com/library for the list of all supported models. @@ -19,7 +23,7 @@ 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 +the provided ollama-limit-gpu-layers script to create model flavors with different num_gpu parameters. ollama uses many gigbytes of disk space in your home directory,