Fixes for saving the gemma tokenizer

This commit is contained in:
Niels Geens
2025-05-22 18:22:15 +02:00
parent 29b35ffccf
commit 19dd9fa2ee
5 changed files with 25 additions and 66 deletions

View File

@@ -3,7 +3,7 @@
#SIA_INSTALL_NO_WEB=1
SIA_INSTALL_NO_NOTEBOOK=1
#SIA_INSTALL_NO_CORE=1
SIA_INSTALL_NO_ITB=1
#SIA_INSTALL_NO_ITB=1
SIA_INSTALL_NO_MISTRAL_INFER=1
#SIA_INSTALL_NO_GEMMA_INFER=1
#SIA_INSTALL_NO_GEMMA_TRAIN=1

View File

@@ -1,6 +1,7 @@
#!/bin/bash
#SIA_INSTALL_NO_WEB=1
#SIA_INSTALL_NO_LLAMA_CPP=1
#SIA_INSTALL_NO_NOTEBOOK=1
#SIA_INSTALL_NO_CORE=1
#SIA_INSTALL_NO_ITB=1
@@ -20,6 +21,12 @@ if [ -z "${SIA_INSTALL_NO_WEB}" ]; then
)
fi
if [ -z "${SIA_INSTALL_NO_LLAMA_CPP}" ]; then
echo "Installing venv for llama.cpp"
python3 -m venv /root/venvs/llama_cpp
/root/venvs/llama_cpp/bin/pip install /root/sia/modules/llama.cpp
fi
if [ -z "${SIA_INSTALL_NO_NOTEBOOK}" ]; then
echo "Installing venv for running notebooks"
python3 -m venv /root/venvs/notebook

View File

@@ -14,4 +14,6 @@ fi
mkdir -p "$OUTPUT_DIR"
gemma_train --model "google/gemma-3-12b-it" --output-dir "$OUTPUT_DIR"
gemma_train --model "google/gemma-3-12b-it" --output-dir "$OUTPUT_DIR"
cp "$OUTPUT_DIR"/tokenizer/* "$OUTPUT_DIR"/merged/
llama-convert-hf-to-gguf --outfile "$OUTPUT_DIR/model.gguf" --outtype q8_0 "$OUTPUT_DIR/merged"