Repolode
Submit
deepopen-com

deepopen-com/deepopen

A non-autoregressive Python SDK that answers predefined classification and scoring questions over text or JSON, returning labels, scores and probabilities instead of free-form text.

View on GitHubWebsite
Stars
3,075
Forks
343
Open issues
17
Last commit
4 days ago

What is it?

DeepOpen is an intent-classification framework and SDK for structured decisions. You define questions with constrained output types, such as choosing one label from options, scoring an ordinal level, or estimating whether a statement is true, and the model returns answers with probabilities and confidence values. It does not generate free-form text, which keeps outputs within a fixed set of types.

The SDK loads upstream Laya checkpoints through a router that selects English, multilingual or typed-decisions models based on the input. It also ships question templates for routing, guard, moderation, triage and email tasks. Installation is from the GitHub repository with Python 3.10 or newer, and CPU or CUDA PyTorch is supported.

The repository also includes Banking77 and CLINC150 training and evaluation guides, with reported accuracy figures. Predictions can be wrong, and the confidence scores need validation on your own data before use in consequential decisions.

Key features

  • 01Answers predefined choice, score and yes/no-style questions over text or JSON input.
  • 02Returns labels, option probabilities and confidence values rather than generated text.
  • 03Router selects English, multilingual or typed-decisions checkpoints per input.
  • 04Provides question templates for routing, guard, moderation, triage and email tasks.
  • 05Runs on CPU or GPU through PyTorch, with a local checkpoint directory option.

Use cases

  • Classifying customer support requests into departments or intents.
  • Routing incoming emails or tickets by urgency or topic.
  • Flagging content for moderation with probability scores that a team can threshold.
  • Prototyping structured classification over text without training a model from scratch.

Quick start

  1. 1

    Install the SDK from the GitHub repository

    python -m pip install "git+https://github.com/deepopen-com/deepopen.git"
  2. 2

    Verify the installation by printing the version

    python -c "import deepopen; print(deepopen.__version__)"
  3. 3

    Load the English checkpoint and run a classification question in Python (see the README quick start example)

Things to know

  • Outputs are constrained predictions that can be wrong, and probabilities need validation on your own data.
  • Model weights are downloaded from the Hub on first use, and the README notes that base checkpoints are overconfident and near chance on some tasks.
  • Benchmark results depend on task-specific training that installing the SDK does not reproduce.
  • Model and dataset licenses are separate from the Apache-2.0 SDK license and must be reviewed.

In collections

Natural language processing tools & libraries

Topics

#intent-classification#nlp#text-classification#python-sdk#structured-output#llm-alternative

FAQ

What is deepopen?
A non-autoregressive Python SDK that answers predefined classification and scoring questions over text or JSON, returning labels, scores and probabilities instead of free-form text. DeepOpen is an intent-classification framework and SDK for structured decisions. You define questions with constrained output types, such as choosing one label from options, scoring an ordinal level, or estimating whether a statement is true, and the model returns answers with probabilities and confidence values. It does not generate free-form text, which keeps outputs within a fixed set of types.
Is deepopen free to use? What license does it have?
deepopen is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is deepopen still maintained?
Actively maintained. The last commit to deepopen was on Nov 3, 2026.
What can I use deepopen for?
Classifying customer support requests into departments or intents. Routing incoming emails or tickets by urgency or topic. Flagging content for moderation with probability scores that a team can threshold. Prototyping structured classification over text without training a model from scratch.
How do I install deepopen?
Install the SDK from the GitHub repository: python -m pip install "git+https://github.com/deepopen-com/deepopen.git" Verify the installation by printing the version: python -c "import deepopen; print(deepopen.__version__)" Load the English checkpoint and run a classification question in Python (see the README quick start example)
Where can I get deepopen?
The source code and installation instructions are on GitHub: https://github.com/deepopen-com/deepopen

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects