This release is a pre-release and may not be stable for production use.
nshconfig v3
Typed Python configuration with editable drafts, live interpolation, and independent read-only snapshots. Config files, presets, and execution are ordinary Python modules and functions.
pip install 'nshconfig==3.0.0a0'
import nshconfig as C
class Pairformer(C.Config):
c_z: int = C.interp(lambda c: c.root(Run).c_z)
class Run(C.Config):
project: str
run_name: str
c_z: int = 128
pairformer: Pairformer = Pairformer()
config = Run.draft()
config.project = "af3"
config.run_name = "wide"
config.c_z = 256
assert config.pairformer.c_z == 256
ready = config.finalize()
config.c_z = 384
assert ready.pairformer.c_z == 256
assert config.pairformer.c_z == 384
For short configs, Run(project="af3", run_name="wide", c_z=256) also creates an editable draft. Constructors have checked keyword signatures; .draft() permits required fields to be populated later. Both use .finalize() before execution.
Composition is Python
def wide(config: Run) -> None:
config.c_z = 256
def experiment() -> Run:
config = Run(project="af3", run_name="wide")
wide(config)
return config
# In your ordinary Python entry point:
# train(experiment().finalize())
The runnable AF3 example demonstrates shared model dimensions, presets, nested configs, and training-set/weight validation in a reduced configuration.
Validation and ownership
Writes defer validation. Reads validate requested values and their dependencies; declaration order does not matter. Use Pydantic Annotated metadata for field-local normalization and constraints, and @C.check for final cross-field checks:
from typing import Annotated
class Data(C.Config):
batch_size: Annotated[int, C.Field(gt=0)] = 1
train_sets: list[str] = ["pdb"]
weights: list[float] = [1.0]
@C.check
def aligned(self) -> None:
if len(self.train_sets) != len(self.weights):
raise ValueError("train_sets and weights must align")
Explicit child assignment preserves identity and requires a single owner. Use .copy() for reuse. Declared defaults are copied automatically per parent. Plain incoming lists/dicts become managed containers; aliases read from a config stay live. Final snapshots and computed subtrees reject nested mutation at runtime.
The same schema type describes drafts and finals. ty, Pyright, and mypy check names and value types; completeness and read-only state are runtime guarantees. Managed list/dict values implement sequence/mapping interfaces and are not built-in list/dict instances. Arbitrary mutable opaque leaves are unsupported.
Python 3.10-3.14 and Pydantic 2.13 through the latest 2.x are supported. Install nshconfig[transport] for trusted cloudpickle transport of notebook-defined schemas. v3 is an intentional breaking redesign of v2.
See DESIGN.md for the exact contract, the documentation for guides, and SKILL.md for agent usage.
1 maintainer
Metadata
1 maintainer
Release files for nshconfig 3.0.0a0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | |
|---|---|---|---|
| nshconfig-3.0.0a0.tar.gz | 12.9 kB | Details |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| nshconfig-3.0.0a0-py3-none-any.whl | Python 3 | none | any | Details |
Total release size: 27.0 kB
Release files / nshconfig-3.0.0a0.tar.gz
| Download URL | nshconfig-3.0.0a0.tar.gz |
|---|---|
| Size | 12.9 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
95d9c8a299b43bea5576033d8ac6378fa698b8b40a1e718af30eeafc66085707
|
|
BLAKE2b-256 checksum How to use checksums |
821a9d9f3fc760a7ccf9e6952450e971fbb1e16a5db9c40d5b568c0fbb270f3b
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
uv/0.11.28 {"installer":{"name":"uv","version":"0.11.28","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"24.04","id":"noble","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}
|
Release files / nshconfig-3.0.0a0-py3-none-any.whl
| Download URL | nshconfig-3.0.0a0-py3-none-any.whl |
|---|---|
| Size | 14.1 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
796c84047a1aa6930fd90052d0a52bc9323d2e5bfedaf0dba133da89739b2a13
|
|
BLAKE2b-256 checksum How to use checksums |
67d3b302595d9001ca87c1d60826fcf24c30fef926c0671e189fd00dcba12ac0
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
uv/0.11.28 {"installer":{"name":"uv","version":"0.11.28","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"24.04","id":"noble","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}
|
