This example shows how LLAMA can be used to summarize gong call transcripts
- Download LLAMA model from https://huggingface.co/TheBloke/Llama-2-13B-chat-GGML/resolve/main/llama-2-13b-chat.ggmlv3.q5_K_S.bin and place it in a ./models folder
- Create a virtual environment:
python -m venv venv - Activate the virtual environment:
source venv/bin/activate - Install the requirements:
pip install -r requirements.txt - On Mac M1, M2 etc make sure to build llama-cpp-python like this:
CMAKE_ARGS="-DLLAMA_METAL=on" FORCE_CMAKE=1 pip install --force llama-cpp-python. This will build the Metal version of LLAMA which is much faster than the CPU version. - Run the summarization script:
python summarize.py <gong_transcript_html_file> <key_speaker_name> <output_file>
Go to any Gong call, click on the transcript tab, and save the HTML file. Then run the following command:
python summarize.py gong-transcript.html "Ward" summary.txt
(Main) Speaker name is used for semi-intelligent chunking of the text.
After a coffee or two this will result in a new text file 'summary.txt' with the summary of the call. At the bottom of the file you will have a total summary of the call, and above that you will have a summary of each chunk of the call.