SimCT source distribution

Public source distribution for SimCT / KDFlow experiments.

This repository stores a Git bundle, not a direct Git branch mirror. The HF Git server rejects ordinary historical image blobs. The bundle retains the original code commit and complete Git history.

  • Branch: vdt/ops/b200-portable
  • Commit: 4b9d70a58df4aa77d2bc35228207b959fdf6bee5
  • Source repository: https://github.com/sontungkieu/SimCT
  • Runtime, checkpoints, datasets and untracked remote_artifacts are excluded.

Download simct-b200-portable.bundle and source-manifest.json from the same pinned HF revision. Verify SHA-256 against the manifest before Git operations.

First checkout in a new directory:

git clone --branch vdt/ops/b200-portable /absolute/path/simct-b200-portable.bundle simct-code
git -C simct-code rev-parse HEAD

For a subsequent complete bundle at the same absolute path, with simct-code checked out on the same branch and no conflicting changes:

git -C simct-code pull --ff-only /absolute/path/simct-b200-portable.bundle vdt/ops/b200-portable

A Git bundle is a read-only transport; edits from the server require a separately authorized publishing workflow. Server proxy/download compatibility has not been verified by the local upload.

Server launchers and port fixes are integrated under experiments/runai. See experiments/runai/README.md inside the checkout. Combined GPU canary remains pending.

MP-OPD launcher uses microbatch 4. Evaluation runner has explicit author-code and paper-spec profiles. Independent eval source archive: simct-eval-94114c1.zip. Extract into a new eval directory; do not update the active training checkout. See scripts/evaluation/CONTRACT_README.md. CPU tests passed; paper math requires math-verify, company preflight and dataset verification remain pending. No GPU evaluation was run.

Pinned LCB-v6 downloader: download_lcb_v6-323216b.py. Data downloading runs on company storage, not on the personal machine.

MP-OPD A/B/C: real-data adapter oracle and controls; 38 CPU tests pass, 1 skip. No real-GPU diagnostic yet. Download MP-OPD-ABC-HUONG-DAN.md and pull-mp-opd-abc.sh. Code via HF bundle, selected SFT data reused locally or pulled at pinned dataset revision.

Reference conflict fix: explicitly exclude entire ambiguous prompt groups in mechanics prep; default standalone preparation remains strict. Guarded MB4 import profile keeps original run names and distinguishes corrected parity from legacy bypass.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support