Este repositorio contiene el código fuente y el enlace a nuestro documento.
ScaleEval es un marco de meta-evaluación asistido por agente-debate que aprovecha las capacidades de los agentes LLM comunicativos múltiples. Este marco admite discusiones múltiples para ayudar a los humanos a discernir a los evaluadores basados en LLM más capaces. Los usuarios pueden suministrar sus presentaciones, criterios y escenarios de LLM con nuestro marco para realizar meta-evaluación.

pip install scaleeval
export OPENAI_API_KEY=XXXX.YYYY.ZZZ
export ANTHROPIC_API_KEY=XXXX.YYYY.ZZZSe requiere Python 3.9+.
git clone [email protected]:GAIR-NLP/scaleeval.git
cd scaleeval
pip install -e .
# install pre-commit hooks
pip install pre-commit
pre-commit install # this is necessary before you commit
git init
git add .
pre-commit run cd scaleeval
python evaluation.pycriteria/metaeval_creativity : puntuación de 1 a 5 para cada envío de LLM en función de los criterios de creatividad y decide qué sumisión es mejor.criteria/metaeval_helpfulness : puntuación de 1 a 5 para cada sumisión de LLM en función de los criterios de ayuda, y decida qué sumisión es mejor.criteria/metaeval_interpretability : puntuación de 1 a 5 para cada envío de LLM en función de la interpretabilidad y decide qué sumisión es mejor.criteria/metaeval_reasoning : puntuación de 1 a 5 para cada presentación de LLM basada en los criterios de razonamiento, y decida qué sumisión es mejor.scaleeval/responses/gpt3.5/brainstorming.jsonl : respuestas generadas a partir de GPT3.5-TURBO basadas en las indicaciones en el escenario de lluvia de ideas.scaleeval/responses/claude-instant/brainstorming.jsonl : Respuestas generadas a partir de Claude-Instant basadas en las indicaciones en el escenario de lluvia de ideas. cd criteria¡Agregue sus propias descripciones de criterios a un nuevo archivo .yaml! Siga el formato a continuación:
criteria :
[ your defined criteria ]:
"1" : ...
"2" : ...
"3" : ...
"4" : ...
"5" : ...Recopile sus envíos de LLM y guárdelos en un archivo .jsonl con el siguiente formato:
{ "prompt" : ..., "category" : ..., "response" : ...}Aquí hay un ejemplo:
{ "prompt" : "Can you tell me how to make chocolate chip cookies?" , "category" : "brainstorming" , "response" : "Certainly! Here's a simple recipe for making chocolate chip cookies: n n Ingredients: n - 1 cup (2 sticks) unsalted butter, softened n - 1 cup granulated sugar n - 1 cup packed brown sugar n - 2 large eggs n - 1 teaspoon vanilla extract n - 3 cups all-purpose flour n - 1 teaspoon baking soda n - 1/2 teaspoon salt n - 2 cups chocolate chips n n Instructions: n 1. Preheat your oven to 350 u00b0 F (175 u00b0 C) and line a baking sheet with parchment paper. n 2. In a large mixing bowl, cream together the softened butter, granulated sugar, and brown sugar until well combined and creamy. n 3. Add the eggs one at a time, beating well after each addition. Then, stir in the vanilla extract. n 4. In a separate bowl, whisk together the flour, baking soda, and salt. Gradually add this dry mixture to the wet ingredients, mixing until just combined. n 5. Fold in the chocolate chips until evenly distributed throughout the dough. n 6. Using a cookie scoop or tablespoon, drop rounded portions of dough onto the prepared baking sheet, spacing them about 2 inches apart. n 7. Bake in the preheated oven for 10-12 minutes, or until the edges are golden brown. The centers may still appear slightly soft, but they will firm up as they cool. n 8. Remove the baking sheet from the oven and let the cookies cool on the sheet for a few minutes before transferring them to a wire rack to cool completely. n 9. Repeat the process with the remaining dough until all the cookies are baked. n n Enjoy your homemade chocolate chip cookies!" }¡Considere citar nuestro documento si encuentra útiles nuestro repositorio y resultados!
@article{chern2024scaleeval,
title={Can Large Language Models be Trusted for Evaluation? Scalable Meta-Evaluation of LLMs as Evaluators via Agent Debate},
author={Chern, Steffi and Chern, Ethan and Neubig, Graham and Liu, Pengfei},
journal={arXiv preprint arXiv:2401.16788},
year={2024}
}