mlenv
Snapshot the machine learning environment to a single file, and diff two snapshots to see exactly what changed.
"It trained fine last week" usually comes down to something in the stack
moving: a different CUDA build behind PyTorch, a minor Python bump, a driver
upgrade, a GPU swapped on a shared box. mlenv captures all of that into one
JSON file you can commit, share, and compare, with the changes most likely to
affect results ranked first.
$ mlenv snapshot -o good.json # when results were correct
$ mlenv snapshot -o now.json # after something changed
$ mlenv diff good.json now.json
risk section key change
high cuda torch_cuda 12.1 -> 11.8
high python version 3.11.9 -> 3.12.0
low packages rich 13.7.0 -> 13.8.0
Install
$ pip install mlenv-cli # from PyPI, once released
$ pip install git+https://github.com/jmweb-org/mlenv # latest, available now # core: Python, platform, packages, CUDA, env vars
$ pip install "mlenv-cli[gpu]" # adds GPU model and driver capture via NVML
mlenv has no heavy dependencies and runs anywhere. GPU details are read
through NVML only when the gpu extra is installed and a driver is present;
without it, every other section is still captured.
What it captures
| Section | Contents |
|---|---|
python |
Version, implementation, interpreter path |
platform |
OS, release, architecture, libc |
packages |
Every installed distribution and its version |
cuda |
PyTorch build, its CUDA and cuDNN versions, availability |
gpus |
GPU count, model per index, driver version |
env |
Training-relevant variables (CUDA_*, NCCL_*, OMP_NUM_THREADS, ...) |
Commands
$ mlenv snapshot -o env.json # write a snapshot (omit -o to print to stdout)
$ mlenv show # pretty-print the current environment
$ mlenv diff a.json b.json # show changes, highest risk first
$ mlenv diff a.json b.json --json # machine-readable changes
$ mlenv diff a.json b.json --check # exit non-zero on any high-risk change
In CI
--check turns the diff into a gate. Commit a baseline snapshot and fail the
build when the runner drifts from it in a way that could change results:
- run: mlenv snapshot -o current.json
- run: mlenv diff baseline.json current.json --check
Risk levels
A change is high risk when it routinely moves numerical results: anything
under cuda, a GPU model or count or driver change, or a Python minor-version
bump. A major-version bump of a sensitive package (torch, numpy,
tensorflow, jax, transformers) is high; a smaller bump is medium.
Everything else is low. Only high-risk changes trip --check.
Exit codes
| Code | Meaning |
|---|---|
| 0 | Ran; no high-risk change (or --check not set) |
| 1 | --check found a high-risk change |
| 2 | A snapshot file was missing or invalid |
License
MIT. See LICENSE.
Metadata
Release files for mlenv-cli 0.2.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| mlenv_cli-0.2.0.tar.gz | 12.6 kB | Without provenance | Details for mlenv_cli-0.2.0.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| mlenv_cli-0.2.0-py3-none-any.whl | Python 3 | none | any | Details for mlenv_cli-0.2.0-py3-none-any.whl |
Total release size: 24.8 kB
Release files / mlenv_cli-0.2.0.tar.gz
| Download URL | mlenv_cli-0.2.0.tar.gz |
|---|---|
| Size | 12.6 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
0e8baf7db1da8bfbb6aac29f68760553c10d6b579539464c1f8770b25743936d
|
|
BLAKE2b-256 checksum How to use checksums |
469c391a900afa994bf675bf54ca24395dc35aa6925b88efd0daa74d688bf5f3
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
uv/0.11.20 {"installer":{"name":"uv","version":"0.11.20","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":null}
|
Release files / mlenv_cli-0.2.0-py3-none-any.whl
| Download URL | mlenv_cli-0.2.0-py3-none-any.whl |
|---|---|
| Size | 12.1 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
9a4af45c9e022c245821d58099e4174d4e4f67c3fbaf38c8f99a47c5a1b8ba0e
|
|
BLAKE2b-256 checksum How to use checksums |
d2ca815077fef529107c947f740f9463cd956ad16e8456c712d136e038c31e7b
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
uv/0.11.20 {"installer":{"name":"uv","version":"0.11.20","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":null}
|
Security for mlenv-cli 0.2.0
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.