A token is the industry term for the volume our models operate. From the input you are giving them to work with to the output they are generating for you.

Think of them like paying for liters of water you consume from your city system and drain back into it.

Tokens are a form of measurement which can be expressed in different units like 1 liter of water is the same as 1000 milliliter of water. Similarly, 1 token consists of 4 characters.

BTW did you know that the average word consists of 4.5 characters? So 1 token is almost a word!

To put this all into perspective again. The text up to here has around 670 characters, 124 words and makes roughly 169 tokens.

If Alta would have created this block it would have costed:

[# of total tokens] x [per token price of Alta] = cost

169 x 0.00002 = 0.0038 USD

We work most of the time with our own proprietary NeoCortex models.

Other models which we can host, optimize and operate for you include:
- GPT Neo 2.7 bn.
- GPT-J 6 bn.
- GPT NeoX 20 bn.
- OPT: Open Pre-trained Transformer Language Models (125M to 66 bn)
- FairSeq (13 bn.)
- CodeGen (16 bn.)
- Bloom 560m
- Bloom 1.1 bn.
- Bloom 1.7 bn.
- Bloom 3 bn.
- Bloom 7.1 bn.
- Bloom 176 bn. (available soon)
- t5 Small
- t5 Base
- t5 Large
- t5 3b
- t5 11b
- Stable Diffusion hosted API

In case you have more questions reach out to our integration team.

We calculate your charges going by the amount of input and output tokens you are working with.

A generation task with our "Alta" model for LongForm is most of the times very output heavy.

Think about the following input prompt:
“The NeoCortex text generation API helps you in your workflow” (60 character = 15 tokens)

Here is the actual output:
“What is the NeoCortex?
NeoCortex is a deep learning text generation API, which uses only natural language to generate texts. The generated text is designed to be human-readable and understandable, helping you in your workflow. The API is easy to learn, with a simple interface. This makes it perfect for use by developers who want to create their own applications.” (360 characters = 90 Tokens)

Makes a total token count of 15 + 90 = 105 Tokens.

0.02$ divided by 1000 times 105 = 0.0021 $ would be charged for your request.

We calculate your charges going by the amount of input and output tokens you are working with.

A generation task with our "Alta" model for LongForm is most of the times very output heavy.

Think about the following input prompt:
“The NeoCortex text generation API helps you in your workflow” (60 character = 15 tokens)

Here is the actual output:
“What is the NeoCortex?
NeoCortex is a deep learning text generation API, which uses only natural language to generate texts. The generated text is designed to be human-readable and understandable, helping you in your workflow. The API is easy to learn, with a simple interface. This makes it perfect for use by developers who want to create their own applications.” (360 characters = 90 Tokens)

Makes a total token count of 15 + 90 = 105 Tokens.

0.02$ divided by 1000 times 105 = 0.0021 $ would be charged for your request.

Yes.

When you signup to our platform and generate an API key you automatically have 10$ to try out our APIs for 30 days.

For help around your integration reach out to our integration team.

When we fine tune your models we are taking your data and teaching a model how to behave on it and what is expected.

Like a nice suit we are tailoring you a text generation, classification, extraction model on your very needs.

All we need from you is a dataset with high quality data from your workflow.

We see first improvements from datasets as small as 500 observations.

In case you have more questions reach out to our integration team.

Yes.

Base functionalities require less than 3-lines of code. For this sake we also have multiple OSS packages for easy integration (PyPi, NPM-JS).

Furthermore, we offer you a dedicated integration support for your onboarding time.

While newcomers require around 4 to 6 weeks to see first results. Somebody switching to our infrastructure finds their way around in less than 2 weeks time.

While we try to manage our API endpoints as self explanatory as possible. We are always happy to help you find the best model.

In case you look for direct guidance reach out to our integration team.

Running large language models (LLM) requires powerful computing machinery in form of GPUs.

Building the infrastructure, maintaining and optimizing for it comes at high cost. We are experts in it having scaled down our models for our user facing products where response times in "ms" matter and cost optimization our margin driver is.

With TextCortex you have the benefit of benefitting from our NeoCortex system in two cost saving ways:

1) Pay-per-token
Ideal for smaller scale operations. For example, less 100k request a month or smaller models with less than 10 bn. parameters.

You run in a shared environment with other customers.

2) Pay-per-GPU-hour
Ideal for large scale process operation. For example, millions of request or models with more than 20 bn. parameters.
‍
You pay us for managing and hosting models depending on the flat active GPU hourly rate your models run.

This is a priority option in which you run dedicated resources just for your case.
‍‍
In case you look for more guidance reach out to our integration team.