Digital Lab Contribute
Contribute

Built in the open.
Better with more minds.

SYMBEX-1 is an inspectable engineering project. Read the implementation, reproduce a result, report a discrepancy, or help extend the toolchain.

01 / CORRECTNESS

Check every bit.

Extend bit-exact tests, isolate exporter mismatches, or improve chip-vs-reference verification. A small reproducible failure is a useful contribution.

Inspect the test suite ↗
02 / REPRODUCIBILITY

Repeat the experiment.

Re-run a benchmark with recorded seeds, validation split, model shape and hardware conditions. Keep correctness and task accuracy beside timing.

Start a reproduction ↗
03 / EXPLANATIONS

Make the next step clearer.

Improve setup instructions, add a focused example, review an encoding explanation, or help with English and Spanish documentation.

Find or open an issue ↗
WHAT’S NEXT / FROM THE REPOSITORY

Good questions
still to answer.

These are documented directions, not capabilities the website claims are complete.

Clean evaluation.

Complete validation-split fixes and re-measure affected results. Separate variant selection from the final test evaluation.

New workloads.

Develop 1D convolutional sensor workloads and human activity recognition comparisons, including the prior-work binary baseline.

More targets.

Extend work on Cortex-M33 and RISC-V / RP2350. Measure the device rather than assuming performance transfers between chips.

Better training.

Explore distillation from the official floating-point MLPerf ResNet-8 and improve the accuracy / resource tradeoff.

Energy measurements.

Latency and memory are measured today. Energy is an open measurement task; no energy reduction is claimed here.

THE PEOPLE / FROM THE PROJECT README

Engineering is collaborative.

Attribution follows the inspected README. See the repository history for additional contributions.

RESEARCH & LICENSING

A work in progress.
A source you can read.

The project README says a preprint is in preparation. A citation will be added when it is available; no publication is represented as existing here.

Project license

The main project carries the GNU Affero General Public License, version 3. Read the canonical license before redistributing or integrating the code.

Read LICENSE ↗
Report an issue or propose a change

Use the GitHub issue tracker with your revision, board, configuration, expected result and actual output. For code changes, include the relevant checks and a clear reproduction.

Open the issue tracker ↗
Previous results and version history

The README retires earlier gated and bitslice results after correctness issues were found. The current fully binary engine replaces an earlier convolutional path with an INT8 first layer. This website presents the current documented path and labels provisional results.

Read the update history ↗
PICK A SMALL, VERIFIABLE NEXT STEP

Make the work better.

Explore the repository ↗