Text Generation
Transformers
Safetensors
PyTorch
English
llama
facebook
meta
llama-2
text-generation-inference

Merge method/script

#1
by khoantap - opened

This is such a good model, would you mind sharing the merging method or merging script ? Thank you

We use ties merge(https://github.com/cg123/ties-merge.git). The only problem we encountered was the inability to use the installed protobuf package. According to system prompts, we set PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=python and resolved this issue.
Here is the merge script we are using:

PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=python \
        python ties_merge.py \
                ${BASE_MODEL} \
                ${TARGET_MODEL} \
                --merge ${SUB_MODEL_1} \
                --merge ${SUB_MODEL_2} \
                --cuda
khoantap changed discussion status to closed

Sign up or log in to comment