> For the complete documentation index, see [llms.txt](https://docs.quantaria.xyz/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.quantaria.xyz/quantaria-overview.md).

# Quantaria Overview

Quantaria builds generative models in a different number format. Instead of dense weight matrices that are shrunk after training, every layer in a Quantaria model is a **tensor network** (a matrix product state or operator, MPS/MPO) from the first training step. The same format is used in physics to describe quantum states, which is what gives the models a direct path to quantum hardware.

{% hint style="info" %}
Quantaria is a research company. Hilbert-1 Nano and Base are live on the [Quantaria Console](/quantaria-console.md). All research figures in these docs come from published papers on the methods Quantaria combines, not from Quantaria's own models. Quantaria models train and run on GPUs today.
{% endhint %}

## The approach

| Idea                                  | What it means                                                                                                  | Basis                                                                                                               |
| ------------------------------------- | -------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------- |
| Tensor-network weights from step zero | Every layer is an MPS/MPO, trained natively, never compressed from a dense model afterwards                    | [Tensor-Network Weights](https://github.com/bhavy2908/dataai_docs/tree/main/architecture/tensor-network-weights.md) |
| Unified multimodal objective          | Next-token loss on text and diffusion loss on image patches over one set of shared tensor cores                | [arXiv:2408.11039](https://arxiv.org/abs/2408.11039)                                                                |
| Tensor Product Attention              | Queries, keys and values factorized from contextual activations; RoPE-compatible, with a much smaller KV cache | [arXiv:2501.06425](https://arxiv.org/abs/2501.06425)                                                                |
| MPO mixture-of-experts                | Experts share one central tensor and keep small expert-specific auxiliary tensors                              | [arXiv:2203.01104](https://arxiv.org/abs/2203.01104)                                                                |
| Rank-adaptive training                | Bond dimensions (χ) shrink or grow during training                                                             | [CoMERA, arXiv:2405.14377](https://arxiv.org/abs/2405.14377)                                                        |
| Schmidt-rank truncation               | Each bond keeps its dominant Schmidt coefficients, the operation that bounds entanglement in a quantum state   | [Tensor-Network Weights](https://github.com/bhavy2908/dataai_docs/tree/main/architecture/tensor-network-weights.md) |
| Canonical-form gauging                | Cores kept as left/right isometries for numerically stable contraction                                         | [Tensor-Network Weights](https://github.com/bhavy2908/dataai_docs/tree/main/architecture/tensor-network-weights.md) |

## Why tensor networks

A dense layer stores every entry of a large matrix. A tensor network stores a chain of small cores whose size is set by the bond dimension χ. Published work on tensor-train and related factorizations reports large parameter and memory reductions on the layers where they are applied. See [Research Foundations](https://github.com/bhavy2908/dataai_docs/tree/main/architecture/research-foundations.md) for each result and its paper.

The same structure carries over to quantum computing. A bond of dimension χ carries at most log₂χ bits of entanglement entropy across its cut, and an MPS with bond dimension 2^k can be prepared by a circuit whose gates act on k+1 qubits. That makes tensor-network layers natural candidates for quantum hardware. See [Quantum Path](https://github.com/bhavy2908/dataai_docs/tree/main/architecture/quantum-path.md).

## Principles

* **Matched baselines.** Every model is compared with a dense transformer of the same size on the same data.
* **Cited numbers.** Every research figure links to its paper.
* **Hardware-honest.** When a result came from a simulator, Quantaria says so. Quantum layers ship in production models only where real hardware supports them.
* **Results published either way.** Evaluation results are published whether or not they favour the tensor-network model.

## Where to go next

* [Models](/models.md) for the full model line-up and status
* [Getting Started](https://github.com/bhavy2908/dataai_docs/tree/main/guides/getting-started.md) to try Hilbert-1
* [Architecture](https://github.com/bhavy2908/dataai_docs/tree/main/architecture/tensor-network-weights.md) for the technical detail


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://docs.quantaria.xyz/quantaria-overview.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.