tutorials/developer-docs/python-api.md

LitGPT High-level Python API

This is a work-in-progress draft for a high-level LitGPT Python API.

 

Model loading & saving

The LLM.load command loads an llm object, which contains both the model object (a PyTorch module) and a preprocessor.

from litgpt import LLM

llm = LLM.load(
    model="url | local_path",
    # high-level user only needs to care about those:
    memory_reduction="none | medium | strong"
    # advanced options for technical users:
    source="hf | local | other"
    quantize="bnb.nf4",
    precision="bf16-true",
    device=""auto | cuda | cpu",
)

Here,

• llm.model contains the PyTorch Module
• and llm.preprocessor.tokenizer contains the tokenizer

The llm.save command saves the model weights, tokenizer, and configuration information.

llm.save(checkpoint_dir, format="lightning | ollama | hf")

 

Inference / Chat

response = llm.generate(
    prompt="What do Llamas eat?",
    temperature=0.1,
    top_p=0.8,
    ...
)

 

Dataset

The llm.prepare_dataset command prepares a dataset for training.

llm.download_dataset(
    URL,
    ...
)

dataset = llm.prepare_dataset(
    path,
    task="pretrain | instruction_finetune",
    test_portion=0.1,
    ...
)

 

Training

llm.instruction_finetune(
    config=None,
    dataset=dataset,
    max_iter=10,
    method="full | lora | adapter | adapter_v2"
)

llm.pretrain(config=None, dataset=dataset, max_iter=10, ...)

 

Serving

llm.serve(port=8000)

Then in another Python session:

import requests, json

response = requests.post(
    "http://127.0.0.1:8000/predict",
    json={"prompt": "Fix typos in the following sentence: Example input"}
)

print(response.json()["output"])