In the context of Generative Artificial Intelligence (GenAI), a token represents a unit of text that the model processes. Tokens can vary in length, ranging from a single character to an entire word, depending on the language and the model's implementation.
Common Words
Words like "cat" or "dog" are typically counted as single tokens
Punctuation
Punctuation marks and spaces are treated as individual tokens
Character Ratio
A token is roughly equivalent to 3.7 English characters
Key Concept
For users of Ask Sage, tokens are allocated based on their subscription plan. These tokens can be used to access various platform features, such as ingesting data into a vector database (Ask Sage Dataset), or making predictions using the models available on the platform.
Token Counting
The total token count encompasses both the user's prompt and the model's response. For instance, if a user submits a prompt that is 10 tokens long and the model generates a response of 20 tokens, the overall token count for that interaction would be 30 tokens.
Important Considerations
The cost of tokens can vary depending on the model used
Token costs in the prompt and response may differ
Users should monitor their token usage to manage costs effectively
Example Interaction
User: "What is the capital of France?"
Token Count: 7 tokens
Model: "The capital of France is Paris."
Token Count: 8 tokens
Total Token Count: 15 tokens
Breakdown
The user prompt consists of 7 tokens
The model's response consists of 8 tokens
The total interaction uses 15 tokens
Be mindful of the token costs, as they can vary by model
The cost of tokens in the prompt and response may differ
What are Ask Sage Tokens?
Model-Agnostic Currency
Ask Sage tokens are unique in that they are model-agnostic and serve as the central currency for interacting with the Ask Sage platform. These tokens enable users to access and utilize any model on the platform without additional effort or extra costs.
Key Benefit
Ask Sage provides users with a seamless experience and flexibility by offering a unified token system that can be used across all models on the platform. This simplifies the process of managing tokens and ensures a consistent user experience across different models.
Inference Tokens
Used for making predictions and generating responses
Chat interactions
Plugin/Agent usage
Model predictions
Training Tokens
Used for data ingestion and processing
Dataset creation
Vector embeddings
RAG preparation
Managing Your Tokens
Users can view their token usage and manage their subscription plans through the Tokens menu/settings tab. Here, they can access detailed information about their tokens.
Subscription: View your current subscription plan
Inference Tokens: Check the number of inference tokens used and remaining
Training Tokens: Check the number of training tokens used and remaining
Auto-Refill
Tokens replenish automatically on the first day of each month. Users can also purchase additional tokens if needed.
Enterprise Users
If you are associated with an Enterprise plan and not an Admin, reach out to your Admin for more tokens.
Ask Sage Inference Tokens
Inference Token Usage
Ask Sage Inference Tokens are used to make predictions or inferences using the models available on the platform. These tokens are consumed when users submit prompts, use the plugins/agents to generate responses, or interact with the models in any way that requires processing text data.
Chat Interactions
Tokens consumed for every message exchange
Plugin Usage
Agent and plugin operations consume tokens
Model Predictions
Complex models may use more tokens
Additional Tokens
Users can purchase additional inference tokens if they exceed their allocated limit or require more tokens for their tasks. This flexibility ensures that users can continue to use the platform without interruptions or restrictions.
Important
Inference tokens are not transferrable between months, so it is essential for users to utilize their tokens effectively within the allocated time frame.
Pricing Note
Inference tokens cost varies based on the model. The information being processed via a prompt has a different cost rate than the response generated by the model.
Ask Sage Training Tokens
Training Token Usage
Ask Sage Training Tokens play a pivotal role for users interested in enhancing their experience with the Retrieval-Augmented Generation (RAG) feature. These tokens are primarily used for creating and populating datasets.
1
Upload Data
Upload files or text to your dataset
→
2
Process Embeddings
Data is converted to vector embeddings
→
3
Store Vectors
Vectors stored in database for RAG
Cost-Effective
Training tokens are priced at an economical rate of $0.80 per 1 million tokens, making it a cost-effective option for your data training endeavors.
Workbooks
Ask Sage Workbooks are similar to datasets, but note that the data sources ingested into a workbook use training tokens for the initial ingestion process. However, once the data is in the workbook, and users perform actions on the data, it will use inference tokens for any subsequent actions.
Important
Training tokens are designed for use within their allocation month and do not roll over to the next month. Strategic planning of training token usage is essential.
Token Refill and Subscription Plans
Refill Options
Enterprise Users
If you are on an Enterprise plan and are not an Admin, please contact your Admin for additional tokens. The following information is primarily intended for individual users.
Ask Sage provides users with the option to refill their tokens or upgrade their subscription plans according to their needs. Users can purchase extra tokens or move to a higher subscription tier for increased access.
Monthly Auto-Refill
Tokens automatically replenish on the first day of each month
Purchase Additional
Buy extra tokens anytime to meet your needs
Upgrade Plan
Move to a higher tier for more monthly tokens
Reminder
Tokens do not carry over from one month to the next, so it is crucial for users to make the most of their tokens within the designated time frame.