Skip to main content

Python client and utilities for tierkreis.

Project description


Tierkreis is a higher-order dataflow graph program representation and runtime designed for compositional, quantum-classical hybrid algorithms.

For a detailed introduction read the paper: Tierkreis: a Dataflow Framework for Hybrid Quantum-Classical Computing.

This repository contains the source for the tierkreis python package, and the protocol buffer definitions for Tierkreis data types and gRPC services (in the protos directory) which define the protocols for the system.

The python package provides a complete development and testing evironment for writing and running Tierkreis program, and allows you to write extensions ("workers") in python. By implementing the gRPC services you can also implement runtimes and workers in other languages.

Getting Started

To install the python package run:

pip install tierkreis

This package is pure python and is compatible with Python 3.9 and above. Tierkreis has a strong, static type system, and full automated type inference is available as an extension on supported platforms via the typecheck optional feature. To install that run:

pip install tierkreis[typecheck]

You can now build a graph (Tierkreis program), optionally type check it and execute it. The recommended environment for this is a Jupyter notebook (especially given some operations are async).

First we need the runtime we are going to execute on and a handle to the primitive functions available on that runtime. The python package comes with the PyRuntime which runs locally in your python environment.

from tierkreis.builder import graph, Namespace, Output, Input
from tierkreis.pyruntime import PyRuntime

cl = PyRuntime([]) # empty list for no extra workers
sig = await cl.get_signature()
ns = Namespace(sig) # get a handle to all functions

The output shows the type signatures of the two functions we will use to add two integers together and unpack a pair. Note Tierkreis functions have named inputs and outputs.

iadd(a: Int, b: Int) -> (value: Int)
unpack_pair(pair: Pair[VarType(a), VarType(b)]) -> (first: VarType(a), second: VarType(b))


The @graph decorator allows you to build graphs using python functions.

def sum_pair(pair: Input) -> Output:
    first, second = ns.unpack_pair(pair) # tierkreis functions can have multiple outputs
    return Output(ns.iadd(first, second))

Calling the decorated function with no arguments (sum_pair()) returns a TierkreisGraph.


In an Jupyter notebook this is immediately visualised, as long as you have Graphviz installed. In a script you can write the image to file with render_graph:

from tierkreis import render_graph

render_graph(sum_pair(), "filename", "pdf")

sum_pair graph

Type check

If you have the typecheck extension installed, you can replace @graph with @graph(type_check_sig=sig), providing the signature retrived from the client as above, and the graph will be type checked when you call the building function. A graph is well-typed if type annotations can be inferred for every edge of the graph. If type check fails, the error message will try to indicate the location of your error.

The type checked version of the graph above looks like:

sum_pair graph

Run graph

We can now run the graph using the client set up earlier:

await cl.run_graph(sum_pair(), pair=(1, 2))

The inputs to the graph are provided via keyword argument, and most of the time you can just provide python values that are auto converted. Here the tuple of integers is automatically converted to the Tierkreis type Pair[Int, Int]. The output is given in Tierkreis form:

{'value': IntValue(value=3)}


For a more involved example see variational.ipynb

Custom workers

Workers are standalone servers which implement a set of functions which can connect to a Tierkreis runtime to add extra primitives. They do this by implementing the worker gRPC service. The tierkreis python package makes it easy to do this by taking care of all the server logic, and the conversion to and from Tierkreis data types. Note that workers are intended to be deployed as part of remote Tierkreis runtimes, but we can use the PyRuntime to test and develop them without any networking code.

For example, we could define a custom function to sum a list:

from tierkreis.worker.namespace import Namespace as WorkerNS
from tierkreis.builder import Const
root = WorkerNS()
custom_ns = root["custom"]

async def sum_list(lst: list[int]) -> int:
    return sum(lst)

cl = PyRuntime([root])
ns = Namespace(await cl.get_signature())["custom"]

def runsum() -> Output:
    return Output(ns.sum_list(Const([1, 2, 3])))
await cl.run_graph(runsum())

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

tierkreis-0.2.0.tar.gz (79.3 kB view hashes)

Uploaded source

Built Distribution

tierkreis-0.2.0-py3-none-any.whl (74.9 kB view hashes)

Uploaded py3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page