Fine-Tunning Ai Models [V.1]

Fine-tuning lets you get more out of the models available through the API by providing:

• Higher quality results than prompting
• Ability to train on more examples than can fit in a prompt
• Token savings due to shorter prompts
• Lower latency requests

Once a model has been fine-tuned, you won't need to provide as many examples in the prompt.

• This saves costs and enables lower-latency requests.

Fine-tuning Steps

1. Prepare and upload training data
2. Train a new fine-tuned model
3. Evaluate results and go back to step 1 if needed
4. Use your fine-tuned model

Preparing your dataset

Once you have determined that fine-tuning is the right solution (i.e. you’ve optimized your prompt as far as it can take you and identified problems that the model still has), you’ll need to prepare data for training the model. You should create a diverse set of demonstration conversations that are similar to the conversations you will ask the model to respond to at inference time in production.

Each example in the dataset should be a conversation in the same format as our Chat Completions API, specifically a list of messages where each message has a role, content, and optional name. At least some of the training examples should directly target cases where the prompted model is not behaving as desired, and the provided assistant messages in the data should be the ideal responses you want the model to provide.

Example format

In this example, our goal is to create a chatbot that occasionally gives sarcastic responses, these are three training examples (conversations) we could create for a dataset:

Examples of Popular Tools That Use AI Models

You can use different AI models to create tools for a range of complex tasks. As shown below, for example.

ChatGPT: GPT-3.5 Model

ChatGPT is OpenAI’s advanced chatbot that uses the latest GPT LLM to generate relevant, human-like responses to prompts.

For example, here’s how it responded to the prompt “Explain how you work in a few lines:”

GPT stands for Generative Pre-trained Transformer:

• Generative: Means it generates content
• Pre-trained: Means the OpenAI team inputted data (known as pre-training) to help the system understand and respond to specific tasks
• Transformer: Means it uses deep learning capabilities to consider the context of words and predict what comes next

ChatGPT uses the GPT-3.5 model for free users and the latest GPT-4 version for paid plans.

Ask ChatGPT a question, and it’ll answer you conversationally.

But that’s not all it does. The tool can also:

Create marketing content (e.g., social media posts, email newsletters, or landing page copy)

Write cold email templates

Break down complicated concepts in simple terms

Translate text into multiple languages

Create spreadsheet formulas and solve math problems

Summarize and categorize huge documents and meeting notes

*NB: ChatGPT can generate inaccurate and sometimes biased information. So always double-check any content you use!

Semrush Tools: ChatGPT API

Several Semrush AI writing tools use ChatGPT API to help marketers streamline and optimize their processes. Including SEO Writing Assistant, AI Writing Assistant, and ContentShake.

Google Bard: PaLM - Version 2

Bard is Google’s free experimental chatbot that uses the second version of an LLM called Pathways Language Model (PaLM).

Its original AI model was the Language Model for Dialogue Applications (or LaMDA for short). However, PaLM 2 is better at:

• Reasoning
• Translating
• Coding

Google designed Bard to be a complementary experience to Search.

• It works by searching the web in real time for answers.
• Then, it uses its findings to converse with users.

For example, here’s how it responded to the prompt “What’s the weather like in Monticello, Utah?”:

Bard can help you:

• Come up with marketing ideas
• Discover relevant tips and tricks
• Adjust the writing’s tone
• Translate English into multiple languages
• Summarize text and data
• Generate content (e.g., ecommerce product page copy)

When it quotes or includes images, Bard links to sources and citations. This sourcing is a helpful feature other popular chatbots are missing.

DALL-E 2: GLIDE

DALL-E 2 is OpenAI’s text-to-image generator that uses a multimodal model called GLIDE. It stands for Guided Language to Image Diffusion for Generation and Editing.

OpenAI used the GLIDE model to:

1. Improve the original DALL-E.
2. Allow DALL-E 2 to have higher image resolutions + higher-quality photorealism.

DALL-E 2 produces AI images from text prompts. The visuals look like:

• Human-created sketches
• Illustrations
• Paintings
• Photos

For example, here’s what it came up with for the prompt “a photo of a spiky hedgehog laying in the grass”:

*The tool will always generate four variations of AI images that it thinks best match your prompt.

You can use DALL-E 2 images in all types of marketing content. For example:

• Blog articles
• Social media posts
• Landing pages
• Email newsletters
• Community forums

Stable Diffusion XL Playground: Stable Diffusion

Stable Diffusion XL is an AI image generator that uses Stable Diffusion’s API. It’s an open-source model—Its code is available to the public. So any creator can use its capabilities to:

• Set up models
• Build tools

Many users believe Midjourney (another popular AI image generator) uses the Stable Diffusion model. But the team hasn’t confirmed that.

• You can create free images using Stable Diffusion XL in its online Playground. Enter your prompt, choose your style, and generate a result.

For example, here’s what it came up with for “a horse running through a candy cane forest” in cinematic style:

• Stable Diffusion’s official AI application is: DreamStudio
• Through the official AI application: It's allowed to get images without watermarks.

Credit: Semrush OpenAI