ollama build script with patch for llama3 llama.cpp BPE pretokenization issue

build_me.sh

#!/bin/bash

cd $(dirname "$0") || exit 123

# builds a trimmed version of ollama (i.e. check the CPU flags and CUDA architectures below)
# in my case: AMD Ryzen | RTX4090
#
# to be put run in directory cloned from ollama and HEAD pointing to some tag
# git clone [email protected]:ollama/ollama.git
# in that cloned repo-dir: 
# EITHER
#	git checkout v0.1.39
# OR
#	git checkout v0.1.38
#

# https://github.com/ollama/ollama/blob/main/docs/development.md

currenttag=$(git tag --points-at HEAD)

if [ "$currenttag" == "v0.1.38" ] ; then
        echo TAG $currenttag FOUND inserting patch

cat << 'EOF' > llm/patches/06-llama.cpp.diff
cat diff --git a/llama.cpp b/llama.cpp
index 72c10ffc..c5b1c174 100644
--- a/llama.cpp
+++ b/llama.cpp
@@ -4348,7 +4348,9 @@ static void llm_load_vocab(
                 LLAMA_LOG_WARN("%s: CONSIDER REGENERATING THE MODEL             \n", __func__);
                 LLAMA_LOG_WARN("%s: ************************************        \n", __func__);
                 LLAMA_LOG_WARN("%s:                                             \n", __func__);
-                vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_DEFAULT;
+                LLAMA_LOG_WARN("%s:          BLABLA OVERRIDE LLAMA3 BLALBLA     \n", __func__);
+                // vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_DEFAULT;
+                vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_LLAMA3;
             } else if (
                     tokenizer_pre == "default") {
                 vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_DEFAULT;
EOF

elif [ "$currenttag" == "v0.1.39" ] ; then
        echo TAG $currenttag FOUND inserting patch

cat << 'EOF' > llm/patches/06-llama.cpp.diff
diff --git a/llama.cpp b/llama.cpp
index af1aede3..1d50f343 100644
--- a/llama.cpp
+++ b/llama.cpp
@@ -4503,7 +4503,10 @@ static void llm_load_vocab(
                 LLAMA_LOG_WARN("%s: CONSIDER REGENERATING THE MODEL             \n", __func__);
                 LLAMA_LOG_WARN("%s: ************************************        \n", __func__);
                 LLAMA_LOG_WARN("%s:                                             \n", __func__);
-                vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_DEFAULT;
+                // vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_DEFAULT;
+               LLAMA_LOG_WARN("%s:          BLABLA OVERRIDE LLAMA3 BLALBLA     \n", __func__);
+               // vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_DEFAULT;
+                vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_LLAMA3;
             } else if (
                     tokenizer_pre == "llama3"   ||
                     tokenizer_pre == "llama-v3" ||
EOF
else
        echo TAG UNKNOWN NOT CREATING PATCHES...
fi


export BUILD_ARCH="amd64"
# https://github.com/ollama/ollama/blob/main/docs/development.md#advanced-cpu-settings
# OLLAMA_CUSTOM_CPU_DEFS="-DLLAMA_AVX=on -DLLAMA_AVX2=on -DLLAMA_F16C=on -DLLAMA_FMA=on" CMAKE_CUDA_ARCHITECTURES=89 go generate ./...

#https://github.com/ollama/ollama/blob/1b0e6c9c0e5d53aa6110530da0befab7c95d1755/docs/gpu.md
CMAKE_CUDA_ARCHITECTURES=89 go generate ./...

# https://github.com/ollama/ollama/blob/main/docs/development.md#advanced-cpu-settings
export OLLAMA_CUSTOM_CPU_DEFS="-DLLAMA_AVX=on -DLLAMA_AVX2=on -DLLAMA_F16C=on -DLLAMA_FMA=on -DCMAKE_CUDA_ARCHITECTURES=89"
export VERSION=$(git describe --tags --first-parent --abbrev=7 --long --dirty --always | sed -e "s/^v//g")
export GOFLAGS="'-ldflags=-w -s \"-X=github.com/ollama/ollama/version.Version=$VERSION\" \"-X=github.com/ollama/ollama/server.mode=release\"'"

go build .