Watsonx.ai Prompt Lab Overview
Key Points
- Watsonx.ai is an enterprise studio that unifies generative AI and traditional machine‑learning tools, letting users build, train, tune, and deploy models tailored to specific business problems.
- In the Prompt Lab, users can craft prompts from scratch or use sample prompts for tasks like summarization, sentiment analysis, or question‑answering, choosing from a curated catalog of foundation models—including IBM’s Granite series and third‑party models such as Llama 2—and adjusting parameters and guardrails to control output quality and safety.
- Completed prompts can be instantly shared via a one‑click API command, saved to a Jupyter Notebook for further testing, or refined iteratively with different models, decoding techniques, and parameter tweaks.
- The platform also includes a Tuning Studio for fine‑tuning foundation models on proprietary data and a no‑code AutoAI environment for building traditional predictive models, enabling seamless collaboration between application developers and data scientists within a single workspace.
Sections
- Untitled Section
- No‑Code Model Building & Synthetic Data - The passage explains how AutoAI enables rapid, no‑code creation, training, and deployment of predictive models, and how Watsonx.ai’s synthetic data generator can fill data gaps by automatically producing large, realistic datasets for model training.
Full Transcript
# Watsonx.ai Prompt Lab Overview **Source:** [https://www.youtube.com/watch?v=swPBNKKPK0E](https://www.youtube.com/watch?v=swPBNKKPK0E) **Duration:** 00:05:28 ## Summary - Watsonx.ai is an enterprise studio that unifies generative AI and traditional machine‑learning tools, letting users build, train, tune, and deploy models tailored to specific business problems. - In the Prompt Lab, users can craft prompts from scratch or use sample prompts for tasks like summarization, sentiment analysis, or question‑answering, choosing from a curated catalog of foundation models—including IBM’s Granite series and third‑party models such as Llama 2—and adjusting parameters and guardrails to control output quality and safety. - Completed prompts can be instantly shared via a one‑click API command, saved to a Jupyter Notebook for further testing, or refined iteratively with different models, decoding techniques, and parameter tweaks. - The platform also includes a Tuning Studio for fine‑tuning foundation models on proprietary data and a no‑code AutoAI environment for building traditional predictive models, enabling seamless collaboration between application developers and data scientists within a single workspace. ## Sections - [00:00:00](https://www.youtube.com/watch?v=swPBNKKPK0E&t=0s) **Untitled Section** - - [00:03:05](https://www.youtube.com/watch?v=swPBNKKPK0E&t=185s) **No‑Code Model Building & Synthetic Data** - The passage explains how AutoAI enables rapid, no‑code creation, training, and deployment of predictive models, and how Watsonx.ai’s synthetic data generator can fill data gaps by automatically producing large, realistic datasets for model training. ## Full Transcript
Watsonx.ai is an enterprise studio that puts generative
AI and traditional machine learning at your fingertips, empowering you
to build, train, tune and deploy
AI models targeted for your specific business needs.
Let's begin our tour of watsonx.ai in the Prompt Lab.
Here you can build a prompt from scratch or select one of the sample prompts
corresponding to common generative AI tasks, including applying summarization
to condense long form content, like an earnings call, into short descriptions.
Or deploying insight extraction and classification
to determine critical insights, like the sentiment of a customer review.
Each sample prompt defaults to the pre-trained foundation
model most appropriate for the task at hand.
Our curated list features foundation models of different sizes and architectures,
including third-party open models like Llama 2
and IBM-developed models like the Granite series.
The models in the Granite series help businesses
build and scale generative AI and we offer certain contractual intellectual property
indemnity protections for these IBM-developed AI models.
You can experiment with different models from the list
and adjust their parameters.
Let's say you want to create a question-answer application
for your company's financial statements.
By clicking the appropriate sample prompt, you can quickly start
prompt engineering in either a freeform or structured
format.
In either format, you can provide natural language instructions
and examples, which helps your model uncover patterns.
You can also add AI guardrails,
which can help prevent your model from receiving or generating
harmful, abusive and profane language.
Okay, let's test your prompt.
Prompt engineering is an iterative process.
You can adjust your prompts, test different models
and tinker with decoding techniques and parameters.
Once you've built the most effective prompt for your use case,
it's easy to share across your organization.
With one click, you can retrieve the API command and send it to a developer.
You can also save the prompt to a Jupyter Notebook.
From there, a data scientist can do additional testing.
Watsonx.ai also includes a Tuning Studio,
enabling your enterprise to harness its proprietary data.
Simply select a foundation model, tuning method and task.
Then upload a dataset.
Once your model is tuned, you can start using it in the Prompt Lab.
Watsonx.ai also supports building traditional machine learning workflows
and models, facilitating collaboration between application developers
and data scientists in a single workspace.
Let's imagine you're looking to create a machine learning model
designed to process and forecast upcoming customer insurance claims.
A data scientist can then quickly create a traditional predictive model that forecasts
the claims amount by connecting to a repository
of historical claims data in a project.
With AutoAI, you can create a machine learning model
from scratch in a no-code environment.
Now you can pull in the training data file,
set the model type, adjust
its parameters and specify the range of options.
You can run experiments in AutoAI.
The most promising pipelines and models are identified on a leaderboard.
You can save the winner as Python code
or immediately deploy it for inferencing.
A developer can use the model endpoint
to integrate the new model with their code and UI.
And just like that, you've built an app.
Now, let's say you didn't have enough data to effectively train your ML model.
In this case, you can leverage watsonx.ai’s synthetic data generator
tool to address your data gaps, whether stored in a database or as a data file,
you can upload your existing data, anonymize columns,
and select how many rows of synthetic data
you want to generate - up to 2.1 billion.
You can also select the type of data file you want output.
Watsonx.ai takes it from there, automatically learning your data’s
distributions and relationships.
In short order, it creates a robust synthetic dataset
that reflects your original training data.
Watsonx.ai is part of the watsonx AI and data platform,
comprised of three core components and complemented by several AI assistants designed
to help you scale and accelerate the impact of
AI with your trusted data across your business.
The core components include: a studio for new foundation models, generative AI
and machine learning; a fit-for-purpose data store
built on an open data lakehouse architecture; and an AI governance toolkit
that accelerates responsible, transparent and explainable AI.
To learn more, start a free trial or reach out
to an IBM representative.