Author: Jingwei Huang (Jingwei.Huang@utsouthwestern.edu)
EnsemLLMs can run a group of large language models on OpenAI, AzureOpenAI, Ollama, and local deployments.
To run models on Ollama, you need to install Ollama first. See how to install Ollama at:
https://github.com/ollama/ollama
-
Create a project folder as "demo" shows.
-
Create prompt template file for the problem of interest in the project home folder. Create a system message file (optional) if prefer.
The variables of interest and their possible values, variable list, the expected JSON output template, can be embedded in a prompt template with the following place holders:
-
{{<Variable-Value Dictionary>}}for variable-value table -
{{<Variable List>}}for variable list -
{{<json_output_template>}}for expected JSON output template
-
Create a variable-value dictionary in yaml, as demo show.
-
Specify expected flattern JSON output template in yaml, as demo show.
-
Specify configuration in "ensemble_config.yaml"
-
Modify setup.py for setting up the paths to local packages in EnsemLLMs.
-
"./demo/demo.ipynb" provides an example to run EnsemLLMs for ECG report labeling, with a single case absert from MIMIC-IV ECG dataset.
-
"./example_project_small/demo_lung.ipynb" demos with a very small dataset about how to use EnsemLLMs to predict lung cancer TNM staging.
Following UT Southwestern Office for Technology Development, the project is using the license from The University of Texas Southwestern Medical Center.