LLM Fine-tuning Use Case

Structured Information Extraction

Extracting structured information from unstructured text has applications across industries like extracting patient information from medical reports or financial data from quarterly call transcripts. However data extraction can be a manual, time-consuming endeavor. Learn how to fine-tune open-source LLMs to automatically generate structured outputs for downstream use.

The Predibase Solution

Generate structured outputs from unstructured text

Efficiently fine-tune open-source LLMs like Llama2 with built-in best practice optimizations like LoRA and quantization
Instantly deploy and prompt your fine-tuned LLM on serverless endpoint
Automate the generation of structured outputs like JSON

Unestructured Text

Reports

Call Transcripts

Web scrapes

Documents

Emails

Structured Data Extraction with LLMs

Example Use Cases

Patient information from medical reports

Financial data from quarterly call transcripts

Product metadata from product descriptions

Fine-tune and serve your own LLM for Information Extraction

Easily and efficiently fine-tune Llama-2 with built-in optimizations such as quantization and LoRA to generate structured JSON outputs. Instantly serve and prompt your fine-tuned LLM with cost-efficient serverless endpoints built on top of open-source LoRAX. Read the full tutorial.

# Kick off the fine-tune job and track the learning curves for your adapter in the Predibase UI
adapter = pb.finetuning.jobs.create(
    config={
        "base_model": "meta-llama/Llama-2-13b",  # specify a HuggingFace LLM to fine-tune
        "epochs": 3,
        "learning_rate": 0.0002,
    },
    dataset=my_dataset,  # Upload your dataset to Predibase beforehand.
    repo="my_adapter",
    description='Fine-tune "meta-llama/Llama-2-13b" with my dataset for my task.',
)

# Dynamically load fine-tuned adapter for serverless inference
prompt = """
        Your task is a Named Entity Recognition (NER) task. Predict the category of
        each entity, then place the entity into the list associated with the
        category in an output JSON payload. Below is an example:

        Input: EU rejects German call to boycott British lamb .
        Output: {{"person": [], "organization": ["EU"], "location": [], "miscellaneous": ["German", "British"]}}
        Now, complete the task.

        Input: {input}
        Output:
"""

print(
    client.generate(
        prompt,
        adapter_id="my_adapter/3",  # Specify adapter/version for inference
        max_new_tokens=512,
        temperature=0.1,
    ).generated_text
)

Example code in Predibase for illustrative purposes only

Resources to Get Started

Ready to efficiently fine-tune and serve your own LLM?

Try Predibase for Free