AIBuildAI v2.5.10

AIBuildAI uses AI agents to design, train, and evaluate machine-learning models from your data and task description.

This release fixes two problems during data preparation on the v2.5 line.

Fixes

• Data preparation now requests a correction when validation answers are a directory instead of a file. This case no longer ends the run with an unexpected error.
• Image answers can now have several rows for one image. Data preparation no longer rejects these answers because their row count is greater than the number of images.

Install

Download the archive and SHA256SUMS attached to this release. Check and install the package:

:::bash
sha256sum -c SHA256SUMS
tar -xzf aibuildai-linux-x86_64-v2.5.10.tar.gz
cd aibuildai-linux-x86_64-v2.5.10
./install.sh
export PATH="$HOME/.local/bin:$PATH"
aibuildai --version

The last command must report aibuildai v2.5.10.

Run a task

Use the README inside the package to choose a model and set up its sign-in. Model access is separate from your AIBuildAI plan.

Sign in to AIBuildAI and create a task configuration:

:::bash
aibuildai login
aibuildai config > task.yaml

Edit the task name, data path, task instruction, output path, and model in task.yaml. Set the time budgets and parallel candidate count for your machine, then start the task:

:::bash
aibuildai run task.yaml

Requirements

• Linux on an x86_64 processor, with glibc 2.29 or newer.
• A Max plan or above, and access to the model service you select.
• Network access and a writable location for task files and outputs.
• 0.6 GB for the installed program and at least 1.4 GB of free space for a first installation. Task environments and outputs need additional space.
• Memory and computing capacity for the selected task.