FizzyFort Documentation

FizzyFort simplifies AI deployment and fine-tuning, allowing companies to focus on product development rather than infrastructure. It offers efficient data cleaning from unstructured sources like text, PDFs, and markups, and supports the creation of high-quality training datasets. With comprehensive fine-tuning tools, scalable model deployments, and flexible cloud integrations, FizzyFort streamlines AI implementation.

FizzyFort provides powerful data cleaning and transformation tools to prepare raw data for AI model training. The platform's robust algorithms generate numerous fine-tune jobs that ensure accuracy, scalability, and performance, optimizing AI models for real-world applications with minimal manual intervention.

FizzyFort adheres to GDPR compliance, ensuring data is AES encrypted. This ensures that only the individuals who uploaded the documents within the company have access, maintaining privacy and data integrity throughout the processing lifecycle.

Fizzy Data facilitates the handling of diverse, unstructured datasets, including text, PDFs, ZIP files, markups, Excel, PowerPoint, Word documents, and more. Our novel algorithm can generate 1000 questions out of 1MB worth of data in 30 seconds, making it the quickest system to generate training data in the market.

Users can easily drag and drop files from their computer into the system, triggering automatic processing.

After reviewing the data, users can edit the generated questions, with a future feature for AI-assisted mass editing. The data can then be sent to the files collection in FizzyFort to initiate fine-tuning, deleted, or downloaded as a JSONL file for on-premise fine-tuning.

Fine-tuning jobs can be easily initiated with an intuitive setup process designed for AI model configurations.

A real-time dashboard allows users to monitor the progress of fine-tuning tasks. The platform ensures optimal metrics by automatically adjusting parameters to prevent overfitting or underfitting.

To fine-tune a model, simply select the model, choose "Fine-Tune Model," pick an uploaded file, and confirm.

This section stores all your JSONL files. Users can delete unnecessary files or upload new JSONL files for model fine-tuning on FizzyFort.

To achieve optimal results from fine-tuning, we recommend a minimum of 100 examples. Below is an example of a correctly formatted JSONL file for upload:

{"messages":[{"role":"system","content":"You are FizzyBot..."},{"role":"user","content":"Name two companies that Daimler has invested in as part of its acquisitions and partnerships."},{"role":"assistant","content":"Daimler has invested in Storedot and Careem as part of its acquisitions and partnerships."}]}
{"messages":[{"role":"system","content":"You are FizzyBot..."},{"role":"user","content":"What was the original establishment year of DaimlerChrysler Aviation?"},{"role":"assistant","content":"DaimlerChrysler Aviation was established in 1998."}]}
{"messages":[{"role":"system","content":"You are FizzyBot..."},{"role":"user","content":"What was the total amount Daimler directly invested in Chrysler after the sale?"},{"role":"assistant","content":"Daimler directly invested US$2 billion in Chrysler after the sale."}]}
{"messages":[{"role":"system","content":"You are FizzyBot..."},{"role":"user","content":"What was the name of the company that acquired the 25 percent stake of MV Agusta back from Daimler?"},{"role":"assistant","content":"MV Holding acquired the 25 percent stake of MV Agusta back from Daimler in December 2017."}]}
{"messages":[{"role":"system","content":"You are FizzyBot..."},{"role":"user","content":"What was a significant indication of the automakers' interest in automated cars in 2015?"},{"role":"assistant","content":"The acquisition of Nokia's Here digital maps division by BMW, Daimler AG, and Volkswagen Group for €2.8 billion was a significant indication of their interest in automated cars."}]}
{"messages":[{"role":"system","content":"You are FizzyBot..."},{"role":"user","content":"What was the purpose of the joint venture established by BYD and Daimler?"},{"role":"assistant","content":"The purpose of the joint venture was to produce luxury electric vehicles."}]}

FizzyFort offers flexible deployment options across various cloud providers with minimal configuration required. Once deployed, the model can be accessed via an API endpoint for use.

FizzyFort offers an intuitive chat feature that lets you test your deployed or base model. To access it, go to the Quick Deploy section, click "View" for any deployed model, then select "Chat." From there, you can adjust general parameters and fine-tune prompts and settings to match your use case.

Each company receives a unique API key for secure access to the API.

To call the completion endpoint, use the following example:

curl -X POST 'https://app.fizzyfort.com/api/callDeployedModel?deploymentName=my-deployment&company=ExampleCorp' \
    -H 'Content-Type: application/json' \
    -H 'x-api-key: Your api key here' \
    -d '{
      "user": "How cool is FizzyFort",
      "system": "You are FizzyBot, a helpful assistant. You also think FizzyFort rocks!",
      "history": [
        {"role": "user", "content": "Hello!"},
        {"role": "assistant", "content": "Hi! How can I help you today?"}
      ],
      "temperature": 0.7,
      "max_tokens": 100,
      "top_p": 1.0,
      "frequency_penalty": 0.5,
      "presence_penalty": 0.5
    }'

Ensure to replace "my-deployment", "ExampleCorp", and "Your api key here" with your specific deployment name, company name, and API key.

Currently, the token limit is set to 50,000 tokens, corresponding to 300 requests per minute. This will increase as FizzyFort grows so keep an eye on any new developments