[Draft] Add generic logic structure and backends functionalities #11

Doomsk · 2024-06-27T13:35:04Z

Design a logic and structure for the backends to work according to some defined rules and have the same standard approach.

According to qadence 2 structure design (notion page):
[x] Add Dialect, Bytecode and Interface generic api classes to accommodate for current and future backends
[x] Adjust pulser and pyqtorch logic and data structure to working with the apis above
[] Use Embedding from pyqtorch v1.3.0

add more types and utils functions

# Conflicts: # qadence2_platforms/backend/api.py

fix some issues

kaosmicadei · 2024-06-28T13:56:01Z

Should we add the QuInstruct extra attributes to this PR, too?

class QuInstruct:
    def __init__(self, name: str, support: Support, *args: Any, **attributes: Any) -> None:
        self.name = name
        self.support = support
        self.args = args
        self.attr = attributes

    def get(self, attribute: str, default_value: Any | None = None) -> Any:
        return self.attr.get(attribute, default_value)

Doomsk · 2024-07-01T08:57:58Z

Should we add the QuInstruct extra attributes to this PR, too?

class QuInstruct:
    def __init__(self, name: str, support: Support, *args: Any, **attributes: Any) -> None:
        self.name = name
        self.support = support
        self.args = args
        self.attr = attributes

    def get(self, attribute: str, default_value: Any | None = None) -> Any:
        return self.attr.get(attribute, default_value)

definitely we can use it for that

awennersteen · 2024-07-02T09:59:10Z

qadence2_platforms/backend/pulser/fresnel_eom/functions.py

+    match direction:
+        case "x":
+            phase = 0.0
+        case "y":
+            phase = np.pi / 2
+        case _:
+            phase = direction


Remember to drop support for Python 3.9 if we do this.

I think that's fine, since 3.8 was the difficult one (shipped with Ubuntu 20), and 3.9 is already outside Numpy support.

good point!

it's almost fully dropped; I'm taking care of the remaining parts of 3.9

awennersteen · 2024-07-02T10:55:16Z

pyproject.toml

@@ -26,7 +26,9 @@ classifiers = [

 # always specify a version for each package
 # to maintain consistency
-dependencies = ["pyqtorch",
+dependencies = [
+  "pulser",


I think pulser-core, pulser-simulation and later with torch, will be sufficient and we can avoid the pulser-pasqal module.

awennersteen · 2024-07-02T10:59:52Z

I'm getting a bit confused here. Can you link me to any written idea for the backend platform?
I will continue this in Slack

awennersteen · 2024-07-02T11:11:10Z

qadence2_platforms/backend/pulser/backend.py

+Fresnel = replace(PulserAnalogDevice.to_virtual(), dmm_objects=(_dmm,))
+FresnelEOM = Fresnel


Shouldn't we call this OrionAlpha

I'm just using the names that I'm aware of; if you have some extra information, let me know and I'll change it!

awennersteen · 2024-07-02T11:15:36Z

qadence2_platforms/backend/pulser/interface.py

+                simulation = QutipEmulator.from_sequence(
+                    sequence=self.sequence.evaluate(self.embedding, values),
+                    with_modulation=True,


so, this is the "old fashioned" API to pulser simulation.
I understand the more "modern" way is to separately create the emulator and then pass it the sequence.
This way is also more compatible with tensor network backends where you want to adjust settings. perhaps that should be made possible in the API?

that's the intention. for now, we are testing functionalities

awennersteen · 2024-07-02T11:15:54Z

qadence2_platforms/backend/pulser/interface.py

+                    with_modulation=True,
+                )
+                result = simulation.run()
+                return result.sample_final_state(N_samples=num_shots)


Run should return the state, no?

awennersteen · 2024-07-02T11:18:28Z

qadence2_platforms/backend/pulser/interface.py

+        np.ndarray,
+    ]
+):
+    def __init__(


For running on the real device (and proposed as default on Pulser coming up) usage of the Batch concept is required.
It might be good now that we're redesigning the API to make it batch aware or compatible.

I can imagine that this would be similar to what I say a few lines below on instantiating the emulator separately.

definitely will take that into account!

awennersteen · 2024-07-02T11:20:26Z

qadence2_platforms/compiler.py

+def compile(
+    model: Model,
+    backend_name: str,
+    device: Optional[str] = None,
+    output_bytecode: bool = False,
+) -> Union[RuntimeInterfaceApi, BytecodeApi]:
+    dialect_module: ModuleType = get_dialect_instance(backend_name)
+    dialect: DialectApi = dialect_module.Dialect(backend_name, model, device)
+    if output_bytecode:
+        return dialect.compile_bytecode()
+    return dialect.compile()


This might be a premature comment, but, I'd expect that the backend should be able to hook in and provide it's own compiler passes.

I'm also a bit confused by the bytecode and dialect, but I don't see anything written anywhere

@awennersteen when I put the docstring, things will be more clear (hopefully)
but long story short, return type RuntimeInterface has all the relevant methods from the specific backend and it will just uses what implementation the backends supports for that particular method.

awennersteen · 2024-07-02T11:22:21Z

tests/test_pyq_compiler.py

+    f_params = {"x": torch.rand(1, requires_grad=True)}
+    wf = compiled_model(pyq.zero_state(2), f_params)
+    dfdx = torch.autograd.grad(wf, f_params["x"], torch.ones_like(wf))[0]
+    assert not torch.all(torch.isnan(dfdx))


Suggested change

assert not torch.all(torch.isnan(dfdx))

assert not torch.any(torch.isnan(dfdx))

awennersteen · 2024-07-02T11:38:20Z

When working with people in the Hamiltonian or Simulation teams they generally always request access to intermediate stage information from the emulator and I understand that this is crucial for their workflow.

I fear that the current approach here will not enable that, but it should be fairly easy to make this available I think.

RolandMacDoland

Thanks @Doomsk. This is a massive PR and I just noticed that I didn't press the submit button after my previous review. Sorry about this but here it is. I haven't gone through all of it in all fairness, will review in steps.

RolandMacDoland · 2024-07-04T15:08:41Z

.pre-commit-config.yaml

@@ -13,14 +13,14 @@ repos:
      - id: black

 -   repo: https://github.com/charliermarsh/ruff-pre-commit
-    rev: "v0.4.2"
+    rev: "v0.4.10"


Could you please check if the template-python-project is up-to-date with pre-commits ?

RolandMacDoland · 2024-07-04T15:09:08Z

pyproject.toml

 docstring-quotes = "double"

 [tool.mypy]
 python_version = "3.10"
-warn_return_any = true
+warn_return_any = false


it doesn't like when certain types of returns happen, namely if you do a function factory

RolandMacDoland · 2024-07-22T17:11:11Z

pyproject.toml

@@ -26,7 +26,10 @@ classifiers = [

 # always specify a version for each package
 # to maintain consistency
-dependencies = ["pyqtorch",
+dependencies = [


Do these dependencies have to be installed by default ? Or could they be added as extra ?

I'll change it once refactoring parts of the Embedding

RolandMacDoland · 2024-07-22T17:20:42Z

qadence2_platforms/backend/bytecode.py

+    def __init__(
+        self,
+        backend: str,
+        sequence: InstructionsObjectType,


It is highly confusing that sequence is of instruction type (sort of) and instructions is of bytecode type.

RolandMacDoland · 2024-07-22T17:21:33Z

qadence2_platforms/backend/bytecode.py

+
+    def __init__(
+        self,
+        backend: str,


Shouldn't that be a backend name ?

RolandMacDoland · 2024-07-22T17:41:19Z

qadence2_platforms/backend/embedding.py

+    """
+
+    model: Model
+    param_buffer: ParameterBufferApi


Suggested change

param_buffer: ParameterBufferApi

param_buffer: ParameterBuffer

RolandMacDoland · 2024-07-22T17:41:32Z

qadence2_platforms/backend/embedding.py

+
+    model: Model
+    param_buffer: ParameterBufferApi
+    mapped_vars: EmbeddingMappingResultType


Suggested change

mapped_vars: EmbeddingMappingResultType

mapped_vars: EmbeddingMapType

RolandMacDoland · 2024-07-22T17:42:11Z

qadence2_platforms/backend/interface.py

+)
+
+
+class RuntimeInterfaceApi(


Suggested change

class RuntimeInterfaceApi(

class RuntimeInterface(

Interface is redundant.

RolandMacDoland · 2024-07-22T17:49:41Z

qadence2_platforms/backend/pulser/backend.py

+
+    if model.register.grid_type == "triangular":
+        layout = TriangularLatticeLayout(n_traps=61, spacing=spacing_unit)
+        transform = np.array([[1.0, 0.0], [0.5, 0.8660254037844386]])


Is this some unit conversion ?

to transform the triangular shape from a regular square one

RolandMacDoland · 2024-07-22T17:51:33Z

qadence2_platforms/backend/pulser/embedding.py

+# TODO: move `from_model` and `name_mapping` to a new `ModelParser` class and
+#  make `Embedding` independent of `Model`; also no need to have `ParameterBuffer`
+
+def np_call(call: Call) -> Callable[[dict, dict], np.ndarray]:


Isn't that covered in Embedding ?

RolandMacDoland · 2024-07-29T11:41:58Z

Hey @Doomsk you mentioned you'll be carrying on with it right ?

Doomsk added 7 commits June 25, 2024 04:10

add dialect, bytecode and pulser backend logic

df067bf

fix inconsistencies in instructions, dialect instructions logic

0a3fed0

add more types and utils functions

add interface structure and other minor logic designs

c8c9f0a

add interface structure and other minor logic designs

b26cad0

Merge branch 'refs/heads/main' into em/backend_dialect

3bbb78a

# Conflicts: # qadence2_platforms/backend/api.py

add apis and logic

a4a93a8

add apis, logic and proper structure for pulser and pyqtorch

f9e85f1

Doomsk requested a review from kaosmicadei June 27, 2024 13:35

Doomsk self-assigned this Jun 27, 2024

Doomsk requested review from dominikandreasseitz and removed request for kaosmicadei June 27, 2024 13:40

Doomsk added draft refactor feature labels Jun 27, 2024

Doomsk added 12 commits June 27, 2024 18:06

add sequence api and logic to pulser and pyqtorch

bfc60e4

add bytecode, interface to pulser

2b7a3b0

add dialect, embedding and interface to pulser backend

0c32dd4

add features to complete the logic; some tests

961a67a

fix some issues

fix pulser partial sequence, functions

137053b

add pulser to dependencies

1ab1bc3

fix circular reference

578ed5f

fix circular reference

38b7835

make qubit_positions mandatory

1aa7e1b

fix typos

ffaad21

oops

173dfd7

Merge branch 'refs/heads/main' into em/backend_dialect

f9206fb

awennersteen reviewed Jul 2, 2024

View reviewed changes

Doomsk mentioned this pull request Jul 4, 2024

[Feature] Implement a State result type #5

Open

3 tasks

Doomsk added 3 commits July 16, 2024 17:23

fix things

da4f33c

add external ir

c226300

add external ir

49b675c

RolandMacDoland requested review from awennersteen and RolandMacDoland and removed request for dominikandreasseitz and awennersteen July 22, 2024 16:30

RolandMacDoland reviewed Jul 22, 2024

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[Draft] Add generic logic structure and backends functionalities #11

[Draft] Add generic logic structure and backends functionalities #11

Doomsk commented Jun 27, 2024 •

edited by dominikandreasseitz

Loading

kaosmicadei commented Jun 28, 2024

Doomsk commented Jul 1, 2024

awennersteen Jul 2, 2024

Doomsk Jul 2, 2024

Doomsk Jul 2, 2024

awennersteen Jul 2, 2024

awennersteen commented Jul 2, 2024

awennersteen Jul 2, 2024

Doomsk Jul 2, 2024

awennersteen Jul 2, 2024

Doomsk Jul 2, 2024

awennersteen Jul 2, 2024

awennersteen Jul 2, 2024

Doomsk Jul 2, 2024

awennersteen Jul 2, 2024

Doomsk Jul 2, 2024

awennersteen Jul 2, 2024

awennersteen commented Jul 2, 2024

RolandMacDoland left a comment

RolandMacDoland Jul 4, 2024

RolandMacDoland Jul 4, 2024

Doomsk Jul 29, 2024

RolandMacDoland Jul 22, 2024

Doomsk Jul 29, 2024

RolandMacDoland Jul 22, 2024

RolandMacDoland Jul 22, 2024

RolandMacDoland Jul 22, 2024

RolandMacDoland Jul 22, 2024

RolandMacDoland Jul 22, 2024

RolandMacDoland Jul 22, 2024

Doomsk Jul 29, 2024

RolandMacDoland Jul 22, 2024

RolandMacDoland commented Jul 29, 2024

		Fresnel = replace(PulserAnalogDevice.to_virtual(), dmm_objects=(_dmm,))
		FresnelEOM = Fresnel

	assert not torch.all(torch.isnan(dfdx))
	assert not torch.any(torch.isnan(dfdx))

	param_buffer: ParameterBufferApi
	param_buffer: ParameterBuffer

	mapped_vars: EmbeddingMappingResultType
	mapped_vars: EmbeddingMapType

[Draft] Add generic logic structure and backends functionalities #11

Are you sure you want to change the base?

[Draft] Add generic logic structure and backends functionalities #11

Conversation

Doomsk commented Jun 27, 2024 • edited by dominikandreasseitz Loading

kaosmicadei commented Jun 28, 2024

Doomsk commented Jul 1, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

awennersteen commented Jul 2, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

awennersteen commented Jul 2, 2024

RolandMacDoland left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

RolandMacDoland commented Jul 29, 2024

Doomsk commented Jun 27, 2024 •

edited by dominikandreasseitz

Loading