Beneath the hood of each AI software are algorithms that churn via knowledge in their very own language, one based mostly on a vocabulary of tokens.
Tokens are tiny items of information that come from breaking down greater chunks of knowledge. AI fashions course of tokens to be taught the relationships between them and unlock capabilities together with prediction, technology and reasoning. The quicker tokens could be processed, the quicker fashions can be taught and reply.
AI factories — a brand new class of information facilities designed to speed up AI workloads — effectively crunch via tokens, changing them from the language of AI to the foreign money of AI, which is intelligence.
With AI factories, enterprises can reap the benefits of the newest full-stack computing options to course of extra tokens at decrease computational value, creating extra worth for purchasers. In a single case, integrating software program optimizations and adopting the newest technology NVIDIA GPUs lowered value per token by 20x in comparison with unoptimized processes on previous-generation GPUs — delivering 25x extra income in simply 4 weeks.
By effectively processing tokens, AI factories are manufacturing intelligence — essentially the most helpful asset within the new industrial revolution powered by AI.
What Is Tokenization?
Whether or not a transformer AI mannequin is processing textual content, photos, audio clips, movies or one other modality, it can translate the info into tokens. This course of is called tokenization.
Environment friendly tokenization helps cut back the quantity of computing energy required for coaching and inference. There are quite a few tokenization strategies — and tokenizers tailor-made for particular knowledge sorts and use instances can require a smaller vocabulary, which means there are fewer tokens to course of.
For giant language fashions (LLMs), quick phrases could also be represented with a single token, whereas longer phrases could also be break up into two or extra tokens.
The phrase darkness, for instance, could be break up into two tokens, “darkish” and “ness,” with every token bearing a numerical illustration, corresponding to 217 and 655. The other phrase, brightness, would equally be break up into “vivid” and “ness,” with corresponding numerical representations of 491 and 655.
On this instance, the shared numerical worth related to “ness” can assist the AI mannequin perceive that the phrases might have one thing in frequent. In different conditions, a tokenizer might assign completely different numerical representations for a similar phrase relying on its which means in context.
For instance, the phrase “lie” may seek advice from a resting place or to saying one thing untruthful. Throughout coaching, the mannequin would be taught the excellence between these two meanings and assign them completely different token numbers.
For visible AI fashions that course of photos, video or sensor knowledge, a tokenizer can assist map visible inputs like pixels or voxels right into a collection of discrete tokens.
Fashions that course of audio might flip quick clips into spectrograms — visible depictions of sound waves over time that may then be processed as photos. Different audio purposes might as a substitute give attention to capturing the which means of a sound clip containing speech, and use one other form of tokenizer that captures semantic tokens, which symbolize language or context knowledge as a substitute of merely acoustic data.
How Are Tokens Used Throughout AI Coaching?
Coaching an AI mannequin begins with the tokenization of the coaching dataset.
Based mostly on the scale of the coaching knowledge, the variety of tokens can quantity within the billions or trillions — and, per the pretraining scaling regulation, the extra tokens used for coaching, the higher the standard of the AI mannequin.
As an AI mannequin is pretrained, it’s examined by being proven a pattern set of tokens and requested to foretell the following token. Based mostly on whether or not or not its prediction is right, the mannequin updates itself to enhance its subsequent guess. This course of is repeated till the mannequin learns from its errors and reaches a goal degree of accuracy, often known as mannequin convergence.
After pretraining, fashions are additional improved by post-training, the place they proceed to be taught on a subset of tokens related to the use case the place they’ll be deployed. These may very well be tokens with domain-specific data for an software in regulation, medication or enterprise — or tokens that assist tailor the mannequin to a particular job, like reasoning, chat or translation. The purpose is a mannequin that generates the best tokens to ship an accurate response based mostly on a consumer’s question — a ability higher often known as inference.
How Are Tokens Used Throughout AI Inference and Reasoning?
Throughout inference, an AI receives a immediate — which, relying on the mannequin, could also be textual content, picture, audio clip, video, sensor knowledge and even gene sequence — that it interprets right into a collection of tokens. The mannequin processes these enter tokens, generates its response as tokens after which interprets it to the consumer’s anticipated format.
Enter and output languages could be completely different, corresponding to in a mannequin that interprets English to Japanese, or one which converts textual content prompts into photos.
To know a whole immediate, AI fashions should have the ability to course of a number of tokens directly. Many fashions have a specified restrict, known as a context window — and completely different use instances require completely different context window sizes.
A mannequin that may course of a couple of thousand tokens directly would possibly have the ability to course of a single high-resolution picture or a couple of pages of textual content. With a context size of tens of 1000’s of tokens, one other mannequin would possibly have the ability to summarize a complete novel or an hourlong podcast episode. Some fashions even present context lengths of 1,000,000 or extra tokens, permitting customers to enter large knowledge sources for the AI to investigate.
Reasoning AI fashions, the newest development in LLMs, can sort out extra complicated queries by treating tokens in another way than earlier than. Right here, along with enter and output tokens, the mannequin generates a number of reasoning tokens over minutes or hours because it thinks about the best way to resolve a given downside.
These reasoning tokens permit for higher responses to complicated questions, identical to how an individual can formulate a greater reply given time to work via an issue. The corresponding improve in tokens per immediate can require over 100x extra compute in contrast with a single inference move on a conventional LLM — an instance of test-time scaling, aka lengthy considering.
How Do Tokens Drive AI Economics?
Throughout pretraining and post-training, tokens equate to funding into intelligence, and through inference, they drive value and income. In order AI purposes proliferate, new rules of AI economics are rising.
AI factories are constructed to maintain high-volume inference, manufacturing intelligence for customers by turning tokens into monetizable insights. That’s why a rising variety of AI companies are measuring the worth of their merchandise based mostly on the variety of tokens consumed and generated, providing pricing plans based mostly on a mannequin’s charges of token enter and output.
Some token pricing plans provide customers a set variety of tokens shared between enter and output. Based mostly on these token limits, a buyer may use a brief textual content immediate that makes use of only a few tokens for the enter to generate a prolonged, AI-generated response that took 1000’s of tokens because the output. Or a consumer may spend the vast majority of their tokens on enter, offering an AI mannequin with a set of paperwork to summarize into a couple of bullet factors.
To serve a excessive quantity of concurrent customers, some AI companies additionally set token limits, the utmost variety of tokens per minute generated for a person consumer.
Tokens additionally outline the consumer expertise for AI companies. Time to first token, the latency between a consumer submitting a immediate and the AI mannequin beginning to reply, and inter-token or token-to-token latency, the speed at which subsequent output tokens are generated, decide how an finish consumer experiences the output of an AI software.
There are tradeoffs concerned for every metric, and the best steadiness is dictated by use case.
For LLM-based chatbots, shortening the time to first token can assist enhance consumer engagement by sustaining a conversational tempo with out unnatural pauses. Optimizing inter-token latency can allow textual content technology fashions to match the studying pace of a median particular person, or video technology fashions to attain a desired body price. For AI fashions partaking in lengthy considering and analysis, extra emphasis is positioned on producing high-quality tokens, even when it provides latency.
Builders must strike a steadiness between these metrics to ship high-quality consumer experiences with optimum throughput, the variety of tokens an AI manufacturing unit can generate.
To deal with these challenges, the NVIDIA AI platform gives an enormous assortment of software program, microservices and blueprints alongside highly effective accelerated computing infrastructure — a versatile, full-stack answer that permits enterprises to evolve, optimize and scale AI factories to generate the following wave of intelligence throughout industries.
Understanding the best way to optimize token utilization throughout completely different duties can assist builders, enterprises and even finish customers reap essentially the most worth from their AI purposes.
Be taught extra in this e-book and get began at construct.nvidia.com.
