Turify Templates

Start with a clean slate to build your own custom flow from scratch.

Compare between prompt templates using template chaining. Visualize response quality across models.

A simple evaluation with a prompt template, some inputs, and three models to prompt. Visualizes variability in response length.

Compares response quality across different system prompts. Visualizes how well it sticks to the instructions to only print Racket code.

Test whether different LLMs know the first sentences of famous books.

Chain one prompt into another to extract entities from a text response. Plots number of entities.

Estimate how sycophantic a chat model is: ask it for a well-known fact, then tell it it's wrong, and check whether it apologizes or changes its answer.

Asks an LLM to estimate the gender of a person, given a profession and salary.

Check for whether models refuse to generate stereotypes about people from different countries.

Extracts named entities from a dataset of tweets, and double-checks the output against multiple eval criteria.

Extract information from a dataset and output it in a structured JSON format using OpenAI's structured outputs feature.

Basic example showing whether a given prompt triggered tool usage.

Check whether asking for a different format (YAML, XML, JSON, etc.) changes the content.

Run the HumanEvals Python coding benchmark to evaluate LLMs on Python code completion, entirely in your browser. A classic!

Get a sense of different model's robustness against prompt injection attacks.

Uses a Tabular Data Node to evaluate LLM performance on basic math problems. Compares responses to expected answer and plots performance.

Uses an LLM scorer to test whether LLMs know the difference between lifetimes of male and female mosquitos.

Shows images of a fox, sparrow, and a pig as a computer scientist and a gamer, using Dall-E2.