Skip to content

Install

Welcome to this easy 2-step install. Estimated time: 2 minutes.

If you want to host Lamini in your VPC or on prem, check out enterprise installer instructions 🔗.

1. Get your Lamini API key 🔑

Your API key is at https://app.lamini.ai/account. If it's your first time, create a free account by logging in.

Add your key to your environment variables. In your terminal, run: 

export LAMINI_API_KEY="<YOUR-LAMINI-API-KEY>"

Put this line in your ~/.bash_profile or equivalent file, so you don't have to rerun it in a new session. Remember to source ~/.bash_profile after you make the change.

echo "export LAMINI_API_KEY='$LAMINI_API_KEY'" >> ~/.bash_profile
source ~/.bash_profile
echo $LAMINI_API_KEY

2. Run an LLM 🦙

Run an LLM with our REST API or Python SDK.

As a test, run the following command. This makes a batch call to Llama 2 and returns structured JSON:

curl --location "https://api.lamini.ai/v1/completions" \
--header "Authorization: Bearer $LAMINI_API_KEY" \
--header "Content-Type: application/json" \
--data '{
    "model_name": "meta-llama/Llama-2-7b-chat-hf",
    "prompt": ["What is the hottest day of the year?", "What is for lunch?"],
    "out_type": {
        "answer": "str"
    }
}'

Great, you've run your first Lamini API call!

Here is a sample response, with structured JSON schema output: 

[
    {
        "answer": "The hottest day of the year is usually around July 21st or 22nd in the Northern Hemisphere, and January 20th or 21st in the Southern Hemisphere"
    },{
        "answer": "Sandwiches"
    }
]

Now you're ready to start building your own LLMs, which includes heavier batch calls and training LLMs to learn more complex domains and tasks from your data.

Install the latest version of lamini.

pip install --upgrade lamini

This is a python wrapper around our REST API. It also includes many high-level classes and functions to make it easier to work with LLMs.

import lamini

lamini.api_key = "<YOUR-LAMINI-API-KEY>"

As a test, run the LLM and call Llama 2: 

from lamini import LlamaV2Runner

llm = LlamaV2Runner()
response = llm("Tell me a story about llamas.")

print(response)

(Optional) Advanced Python setups

Advanced Python setup: notebook

You have several other options to authenticate if the above methods do not work for you.

If you're in an iPython notebook, you can pass in your Lamini API key to any Python model class, e.g. LLMEngine or LlamaV2Runner, as shown below:

from lamini import LlamaV2Runner

config = { "production.key": "<YOUR-LAMINI-API-KEY>"}
llm = LlamaV2Runner(config=config)
response = llm("Tell me a story about llamas.")

print(response)

You can also create a file at ~/.lamini/configure.yaml with your Lamini API key in it:

production:
    key: "<YOUR-LAMINI-API-KEY>"

This will be implicitly read for any Python model class, e.g. LLMEngine or LlamaV2Runner, without needing to pass in the config variable. As a test:

from lamini import LlamaV2Runner

llm = LlamaV2Runner()
response = llm("Tell me a story about llamas.")

print(response)

Advanced Python setup: VPC or on premise

If you are running Lamini in your VPC or on prem, you can change the URL from Lamini's hosted service to your own server URL:

config = {
    "production.key": "<YOUR-LAMINI-API-KEY>",
    "production.url" : "<YOUR-SERVER-URL-HERE>"
}

Test that it works: 

llm = LlamaV2Runner(config=config)
response = llm("Tell me a story about llamas.")

print(response)

Add the extra url field:

production:
    key: "<YOUR-LAMINI-API-KEY>"
    url: "<YOUR-SERVER-URL-HERE>"

Advanced Python setup: Google Colab

Here's a code snippet to paste in Google Colab that automatically authenticates for you via Google by placing your Lamini API key into the yaml file, as above:

# @title Setup: Authenticate with Google & install the open-source [Lamini library](https://pypi.org/project/lamini) to use LLMs easily
%%capture

from google.colab import auth
import requests
import os
import yaml

def authenticate_lamini():
auth.authenticate_user()
gcloud_token = !gcloud auth print-access-token
lamini_token_response = requests.get('https://api.powerml.co/data_studio/auth/verify_gcloud_token?token=' + gcloud_token[0])
return lamini_token_response.json()['token']

production_token = authenticate_lamini()
!pip install --upgrade lamini

keys_dir_path = '/root/.lamini'
os.makedirs(keys_dir_path, exist_ok=True)

keys_file_path = keys_dir_path + '/configure.yaml'
with open(keys_file_path, 'w') as f:
yaml.dump(config, f, default_flow_style=False)

As a test, run this LLM call in a subsequent cell: 

from lamini import LlamaV2Runner

llm = LlamaV2Runner()
response = llm("Tell me a story about llamas.")

print(response)