Skip to content

v1.10.0 - Inviting Interop

Choose a tag to compare

View all tags
@Strilanc Strilanc released this 25 Oct 08:35
· 287 commits to main since this release
v1.10.0
285ad58
This commit was signed with the committer’s verified signature. The key has expired.
Strilanc Craig Gidney
GPG key ID: 0A3D851F2B7FF00B
Expired
Verified
Learn about vigilant mode.

Flagship changes:

  • Use open source decoders that are 100x to 1000x faster than before! (pymatching v2 and fusion_blossom)
  • Produce 2d and 3d diagrams of circuits and detector error models, with built-in viewers for Jupyter notebooks!
  • Sample directly from detector error models. Record the errors that occurred, edit them, and play them back!

Notable changes:

  • Added stim.Circuit.diagram
    • Added diagram type "detector-slice-text"
    • Added diagram type "detector-slice-svg"
    • Added diagram type "match-graph-svg"
    • Added diagram type "match-graph-3d"
    • Added diagram type "match-graph-3d-html"
    • Added diagram type "timeline-text"
    • Added diagram type "timeline-svg"
    • Added diagram type "timeline-3d"
    • Added diagram type "timeline-3d-html"
  • Added stim.DetectorErrorModel.diagram
    • Added diagram type "match-graph-svg"
    • Added diagram type "match-graph-3d"
    • Added diagram type "match-graph-3d-html"
  • Added methods to stim.PauliString:
    • stim.PauliString.from_numpy
    • stim.PauliString.from_unitary_matrix
    • stim.PauliString.to_numpy
    • stim.PauliString.to_tableau
    • stim.PauliString.to_unitary_matrix
  • Added methods to stim.TableauSimulator:
    • stim.TableauSimulator.__init__ now has an optional seed argument
    • stim.TableauSimulator.set_state_from_stabilizers
    • stim.TableauSimulator.set_state_from_state_vector
    • stim.TableauSimulator.do now accepts stim.CircuitInstruction and stim.CircuitRepeatBlock
  • Added methods to stim.Tableau:
    • stim.Tableau.iter_all
    • stim.Tableau.from_circuit
    • stim.Tableau.from_numpy
    • stim.Tableau.from_stabilizers
    • stim.Tableau.from_state_vector
    • stim.Tableau.from_unitary_matrix
    • stim.Tableau.to_circuit
    • stim.Tableau.to_numpy
    • stim.Tableau.to_pauli_string
    • stim.Tableau.to_state_vector
  • Added methods to stim.DetectorErrorModel:
    • stim.DetectorErrorModel.flattened
    • stim.DetectorErrorModel.rounded
    • stim.DetectorErrorModel.compile_sampler
  • Added stim.CompiledDemSampler class with methods:
    • stim.CompiledDemSampler.sample
    • stim.CompiledDemSampler.sample_write
  • Improvements to sinter
    • Added support for pymatching v2
    • Added support for fusion_blossom decoder
    • Added flag --title to sinter plot
    • Added flag --miny to sinter plot
    • Added flag --plot_args_func to sinter plot
    • Added flag --failure_units_per_shot_func to sinter plot
    • Added flag --failure_unit_name argument to sinter plot
    • Added grid lines to the plots made by sinter plot
    • Added filter_func, failure_units_per_shot_func, and failure_unit_name arguments to sinter.plot_error_rate and sinter.plot_discard_rate
    • Added sinter.predict_observables method
    • Added ability to discard shots based on observables being mispredicted
      • Added --postselected_observables_predicate flag to sinter collect
      • Added sinter.Task.postselected_observables_mask field
      • Added postselected_observable_mask argument to sinter.sample_decode
  • Added stim diagram command for producing diagrams, with flags --in, --type, --out, --tick, --remove_noise, --filter_coords.
  • Added stim sample_dem command for directly sampling from detector error model files, with flags --err_out, --err_out_format, --in, --obs_out, --obs_out_format, --out, --out_format, --replay_err_in, --replay_err_in_format, --seed, --shots, --err_out, --err_out_format, --replay_err_in, --replay_err_in_format.
  • Added separate_observables=False argument to stim.CompiledDetectorSampler.sample
  • Added bit_packed=False argument to stim.CompiledDetectorSampler.sample
  • Added bit_packed=False argument to stim.CompiledMeasurementSampler.sample
  • Deprecated stim.CompiledDetectorSampler.sample_bit_packed
  • Deprecated stim.CompiledMeasurementSampler.sample_bit_packed

Notable bug fixes:

  • Fixed stim.Circuit.flattened not fusing all operations
  • Fixed stim.Circuit.without_noise not fusing all operations
  • Fixed PAULI_CHANNEL_2 not being marked as targeting pairs of qubits
  • Fixed sinter's internal python files not being marked as private, so they don't appear as autocomplete suggestions
  • Fixed stim.Circuit.detector_error_model failing when errors affected more than 15 detectors (unless decompose_errors=True)
  • Fixed stim.Circuit.without_noise not making independent copies of the operations (!)
  • Fixed sinter only sampling using the first decoder given to it, instead of all decoders, when given multiple decoders
  • Fixed not redirecting stdout and stderr to python's when calling stim.main
  • Fixed unnecessary overhead in the conversion to/from numpy arrays (10x faster for small arrays)
  • Fixed a segfault and bad parsing logic in stim.read_shot_data_file (!). Fuzz tested the read/write methods against the python reference methods to avoid this happening again.

Notable dev changes:

  • Fixed files being opened in text mode instead of binary mode, resulting in bad parsing/serialization on Windows due to \n bytes being turned into \r\n and vice versa.
  • Templatized classes like simd_word so AVX and SSE code can coexist in the same binary
  • Added stim_python_bindings cmake target
  • The cmake build now uses file lists which can be regenerated using a script
  • Added dev/ directory for scripts
  • File lists used by cmake are now generated automatically instead of maintained manually
  • The polyfill simd_word that uses 64 bit words now only uses one word instead of two (reducing padding overhead)
  • To allow supporting a larger variety of decoders, sinter no longer includes each decoder in its requirements. Decoders need to be installed separately to use them.
  • Added -fPIC to libstim cmake target
  • Sinter now passes most information via the disk instead of through the multiprocessing API. This reduced the workload on the manager and the startup costs of batches, but made it much more disk limited on high core count machines.
  • The API reference no longer shows deprecated methods (these methods are still listed in the .pyi stubs file)
  • All doc strings now wrap at 80 characters, and this is enforced by continuous integration.
  • The command line reference documentation has been substantially improved (e.g. flag information now grouped with the command instead of separate)
  • Added python 3.11 support and OSX ARM support to the set of prebuilt wheels
Assets 31
Loading