FAQ

Gretel Playground FAQ

Frequently Asked Questions (FAQ)

1. What types of data can I work with using Playground? Gretel Playground is designed to support tabular data containing any combination of numeric, categorical, and text modalities. This flexibility allows you to work seamlessly across various types of datasets, catering to a broad range of data generation and augmentation tasks.

2. How can I get started? Log in or create a free Gretel account, and access Playground by selecting it in the left navigation or using the SDK through your API key.

3. What can I do with Playground? You can generate tabular data from natural language or SQL prompts, edit existing datasets, augment data, fill in missing values, experiment interactively in the console, and generate and edit data at scale using our batch API and SDK.

4. Why is my feedback important? Your feedback helps us prioritize our development roadmap. By sharing your experience and suggestions, you directly contribute to shaping the future features and improvements of Playground.

5. What about larger datasets and advanced features? We're committed to rapidly increasing the scale of datasets that Playground can handle and are continuously working on enhancing the AI's capabilities. Expect regular updates and improvements based on user feedback.

6. Can I use Playground to work with my existing datasets? Absolutely! Playground is designed to assist in editing and augmenting existing datasets. You can fill in missing values, make corrections, or extend your datasets using natural language prompts.

7. How can I learn to use Playground effectively? Start with the getting started guide, prompts guide, and examples and tutorials. You can also reach out to us if you have more questions.

8. Is Playground a model or an application? It's actually both. Playground is a compound AI system that leverages multiple transformer-based models, including Gretel’s own fine-tuned LLM.

9. How does Gretel Playground overcome the limitations of traditional LLMs in data generation tasks? Traditional LLMs are limited by their context windows and struggle with tasks that exceed these limits or require precise mathematical operations. Gretel Playground overcomes these by using an agent-based approach that plans tasks, delegates operations beyond the scope of LLMs, and ensures high-quality output without the complexities for the user.

10. Can I run Gretel Playground in my own cloud or VPC? Currently, Playground runs inside Gretel’s managed cloud. We are working to make it available in any public cloud.

11. What else is coming for Playground? Data quality and diversity, as well as advanced agent capabilities and some LLM model updates, are still under development.

12. Are there safety checks for prompts submitted to Playground?

    At Gretel, we are committed to promoting fair and equitable use of our AI systems. We firmly stand against any hateful, discriminatory, or otherwise harmful content. All prompts submitted to Playground undergo safety and alignment checks to ensure they adhere to our guidelines, utilizing the safety checks built into the LLMs. Content flagged as potentially harmful will be reviewed by our security team, and violators may have their access revoked. We take these measures seriously to maintain a safe and respectful environment for all users. For more information on what constitutes acceptable use, please visit our guidelines at ​https://gretel.ai/terms.

13. What data sources is Playground trained on? Playground is trained on high quality, structured and semi-structured tabular datasets with permissible licenses, that have been curated and organized across over 20 industry verticals including Healthcare, Biotechnology, Finance, Telecommunications, Government, Pharma, Retail, and others. Goals with model training include familiarizing the model with industry specific dataset formats, teaching data correlations inside analytics and machine learning datasets, and improving task performance for being able to fill in missing values, clean data, or generate data at scale for analytics and machine learning use cases.

14. What large language model (LLM) does Gretel Playground use for generating tabular data? Gretel Playground uses a mixture of expert models including foundation models and Gretel's fine-tuned model specialized in generating tabular data. Data generation requests may utilize a combination of models to compare and optimize performance.

15. Can you share the details of each LLM that Gretel Playground uses? Certainly! There are currently five options available for customers:

    Name	Model	Description
    auto	Auto-selected model	This setting automatically selects the best model from the list below to generate high-quality data at scale. Note: please read each description carefully to understand specific constraints of each model and, if applicable, to make a different model selection when using Playground to best suit your use case.
    Gretel Custom Model (Industry fine-tuned) gretelai/Mistral-7B-Instruct-v0.2/industry	Gretel's proprietary LLM	Gretel's proprietary model is based on Mistral-7b and fine-tuned by Gretel on curated and synthetic industry-specific datasets from 10+ verticals. Data generated from this LLM is owned by the user and can be used for any downstream task without licensing concerns.
    Gretel Llama-3.1-8B-Instruct gretelai/Llama-3.1-8B-Instruct	Gretel's LLM + Llama 3.1 model	Built with Llama 3.1. Gretel's LLM and Llama 3.1 are both used in this option. This option offers high quality and data available for commercial use. For more please see Llama 3.1 official license and policy on Github.
    Gretel Azure GPT-3.5 Turbo gretelai-azure/gpt-3.5-turbo	Gretel's LLM + Azure OpenAI models	Gretel's LLM along with Azure OpenAI models are both leveraged. This option offers excellent free text capabilities and speed, but data generated from this model may have certain restrictions. Please see Azure's documentation for possible restrictions.
    Gretel Google Gemini Pro gretelai-google/gemini-pro	Gretel's LLM + Google Gemini Pro models	Both Gretel's LLM along with Google Gemini Pro models are leveraged in this option. This option offers excellent free text capabilities and speed, but data generated from this model may have certain restrictions. Please read Google's documentation to understand possible restrictions.