Using model access tokens

In this example we will show how to find model prefixes, models and get short lived session tokens.

Anatomy of a model url

Practicus services follow this pattern:
- [ primary service url ] / [ model prefix ] / [ model name ] / < optional version > /
Sample model addresses:
- https://service.practicus.io/models/practicus/diamond-price/
- https://service.practicus.io/models/practicus/diamond-price/v3/
Please note that Practicus AI model urls always end with a "/"

import practicuscore as prt 

region = prt.regions.get_default_region()

# Let's get model prefixes dataframe
# We can also use the list form with: region.model_prefix_list 
model_prefix_df = region.model_prefix_list.to_pandas()

print("Current model prefixes:")
print("Note: we will use the 'prefix' column in the API urls, and not the 'key'.")

display(model_prefix_df)

print("Current models:")
print("Note: we will use the 'name' column in the API urls, and not 'model_id'")

df = region.model_list.to_pandas()
display(df)

# You can use regular pandas filters
# E.g. let's search for models with a particular model prefix, 
# and remove all models that are not deployed (hs no version)

model_prefix = 'models/practicus'
filtered_df = df[(df['prefix'] == model_prefix) & (df['versions'].notna())]
display(filtered_df)

model = "diamond-price"

api_url = f"{region.url}/{model_prefix}/{model}/"

print("Getting Model API session token for:", api_url)
token = prt.models.get_session_token(api_url)

print("Model access token with a short life:")
print(token)

Getting Model API session token using REST API calls

If your end users do not have access to Practicus AI SDK, they can simply make the below REST API calls with any programming language to get a Model API session token.

# "No Practicus SDK" sample to get a session token

import requests

console_api_url = "http://local.practicus.io/console/api/"

# Option 1 - Use password auth every time you need tokens
print("[Not Recommended] Getting console API access token using password.")
email = "admin@admin.com"
password = "admin"

data = {"email": email, "password": password}
console_login_api_url = f"{console_api_url}auth/"
r = requests.post(console_login_api_url, headers=headers, json=data)
if not r.ok:
    raise ConnectionError(r.status_code)
body = r.json()
refresh_token = body["refresh"]  # Keep refresh tokens safe!
console_access_token = body["access"] 

# Option 2 - Get a refresh token once, and only use that until it expires in ~3 months
print("[Recommended] Getting console API access token using refresh token")
console_access_api_url = f"{console_api_url}auth/refresh/"
headers = {"authorization": f"Bearer {refresh_token}"}
data = {"refresh": refresh_token}
r = requests.post(console_access_api_url, headers=headers, json=data)
if not r.ok:
    raise ConnectionError(r.status_code)
body = r.json()
console_access_token = body["access"]
headers = {"authorization": f"Bearer {console_access_token}"}

# Console API access tokens expire in ~30 minutes
print("Console API access token:", console_access_token)

# Locating model id
print("Getting model id.")
print("Note: you can also view model id using Open API documentation (E.g. https://../models/redoc/), or using Practicus AI App.")
r = requests.get(model_api_url + "?get_meta=true", headers=headers, data=data)
if not r.ok:
    raise ConnectionError(r.status_code)
model_id = int(r.headers["x-prt-model-id"])
print("Model id:", model_id)

# Getting model access token, expires in ~4 hours
print("Getting a model API session token using the console API access token") 
console_model_token_api_url = f"{console_api_url}modelhost/model-auth/"
data = {"model_id": model_id}
r = requests.get(console_model_token_api_url, headers=headers, data=data)
if not r.ok:
    raise ConnectionError(r.status_code)
body = r.json()
model_api_token = body["token"]
print("Model API session token:", model_api_token)

Previous: Create Virtual Envs | Next: Use Polars