This example is a 100% local meeting summarization tool that runs on your machine using LiquidAI/LFM2-2.6B-Transcript, a small language model specialized in summarizing meeting transcripts, powered by llama.cpp for fast inference.
What it does
This tool provides a simple CLI to summarize meeting transcripts locally:
- Processes meeting transcripts without sending data to any cloud service
- Uses a specialized 2.6B parameter model optimized for meeting summaries
- Streams tokens in real-time so you can see the summary being generated
- Can be piped with an audio transcription model for a complete audio-to-summary pipeline
Prerequisites
Install uv
If you don’t have uv installed, follow the installation instructions.curl -LsSf https://astral.sh/uv/install.sh | sh
powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"
Quick start
Run with default transcript
Run the tool without cloning the repository using a uv run one-liner:uv run https://raw.githubusercontent.com/Liquid4All/cookbook/refs/heads/main/examples/meeting-summarization/summarize.py
Use a custom transcript
Pass a different transcript file using the --transcript-file argument (supports local files or HTTP/HTTPS URLs):uv run https://raw.githubusercontent.com/Liquid4All/cookbook/refs/heads/main/examples/meeting-summarization/summarize.py \
--transcript-file https://raw.githubusercontent.com/Liquid4All/cookbook/refs/heads/main/examples/meeting-summarization/transcripts/example_2.txt
Clone and customize (optional)
For deeper experimentation and code modification:git clone https://github.com/Liquid4All/cookbook.git
cd cookbook/examples/meeting-summarization
uv run summarize.py \
--model LiquidAI/LFM2-2.6B-Transcript-GGUF \
--hf-model-file LFM2-2.6B-Transcript-1-GGUF.gguf \
--transcript-file transcripts/example_1.txt
How it works
The CLI uses llama.cpp Python bindings to automatically download and build the llama.cpp binary optimized for your platform:
model = Llama(
model_path="LiquidAI/LFM2-2.6B-Transcript-GGUF",
n_ctx=8192,
n_threads=4,
verbose=False,
)
stream = model.create_chat_completion(
messages=[
{"role": "system", "content": system_prompt},
{"role": "user", "content": transcript}
],
max_tokens=2048,
temperature=0.0,
top_p=0.9,
stream=True, # Enable streaming
)
for chunk in stream:
delta = chunk['choices'][0]['delta']
if 'content' in delta:
token = delta['content']
summary_text += token
console.print(token, end='', highlight=False)
The llama.cpp binary is automatically built and optimized for your platform during the first run, so no manual setup is required.
Next steps
Build a complete 2-step pipeline:
- Use an audio transcription model to convert meeting audio to text
- Pipe the transcript into this summarization tool
This entire workflow can run on your machine without any cloud services or API keys.
Source code
View the complete source code on GitHub. 