fromthesky/pldr-llm-training-dynamics-data

PLDR-LLM Training Dynamics Data

Reported numerical evidence for Training and Inference Dynamics of PLDR-LLMs: Row-Map Collapse, Renormalization, and Predictive Reduction, by Burc Gokden.

• Monograph: Hugging Face Paper Page.
• Scientific code and readers: GitHub repository.
• Numerical evidence: Hugging Face dataset.
• Book: Power Law Graph Attention and PLDR-LLMs: Mathematical Foundations, Training Dynamics, and Predictive Inference, by Burc Gokden.
• Book Companion: Code and edition guidance.
• Citation: CITATION.cff.

This compact dataset contains numerical records, complete reported outcome grids, controls, uncertainty summaries, statistical roles and portable provenance. It includes no manuscript source, PDF, executable helper, corpus text, token array, model checkpoint or large raw observation array.

Access

Use an empty destination and an immutable revision. This anonymous HTTPS recipe needs Git, Git LFS, and sha256sum (GNU coreutils). Install Git LFS with your operating system's package manager, then check git lfs version. The tested route uses Git LFS's HTTPS compatibility endpoint; a separate git-xet client is not required for this recipe. See the Hub download documentation for other transport options.

PLDR_DATA_REV=5b1f9a53ca8a8e4f3c208b3b7b2625b5dd0c2cfe
PLDR_DATA_DIR=pldr-llm-training-dynamics-data
GIT_LFS_SKIP_SMUDGE=1 git -c credential.helper= clone --no-checkout \
  https://huggingface.co/datasets/fromthesky/pldr-llm-training-dynamics-data "$PLDR_DATA_DIR"
git -C "$PLDR_DATA_DIR" lfs install --local
GIT_LFS_SKIP_SMUDGE=1 git -C "$PLDR_DATA_DIR" checkout --detach "$PLDR_DATA_REV"
git -C "$PLDR_DATA_DIR" -c credential.helper= lfs pull origin
git -C "$PLDR_DATA_DIR" lfs fsck
(cd "$PLDR_DATA_DIR" && sha256sum -c SHA256SUMS)

This example selects the original public evidence release, including its historical README and metadata. To fetch the dataset matched to a newer code companion, use that companion's scripts/download_release_inputs.py: its reviewed release-gate.json supplies the exact dataset commit and checksums. A moving branch name is not a release identity. In the current export, the 2,063 unique compressed objects occupy 53,672,829 bytes (about 53.7 MB); allow additional space for Git/LFS storage, metadata and any extraction (about 334.5 MB summed over all 2,081 logical records). Git LFS must materialize the .gz objects: small text files beginning with version https://git-lfs.github.com/spec/v1 are pointers, not evidence bytes. The manifest SHA-256 for the historical example above is d639bc535224e03ae34e71e057f5309916e10e4a7d17b0275e5b146de8d4170a.

Keep the exact dataset inventory. A generic Hub cache/snapshot directory can contain symlinks or .cache metadata and is not a verified substitute for this checkout. Write extracted records, logs and validation outputs outside the dataset directory. Download tools are separate from the reader: after download, verification and reading require only Python 3.11+ and its standard library.

Run these commands from the Training Dynamics code repository, replacing DATA-REPO with the downloaded checkout's path:

python3 scripts/verify_evidence.py --data-repo DATA-REPO
python3 scripts/read_evidence.py --data-repo DATA-REPO --list
python3 scripts/verify_evidence.py --data-repo DATA-REPO --extract /tmp/pldr-evidence

The verifier authenticates all indexed compressed objects, their decompressed bytes and the display/claim links. No private research repository is needed.

The Book Companion reuses these scientific records and supplies its own book correspondence. Dataset coverage, figure/table numbers and the preferred citation remain those of the source monograph; they are not book pagination.

From the Book Companion root, verify the same downloaded bytes with:

python3 scripts/check.py evidence --data-repo DATA-REPO

Contents and interpretation

• index.json maps logical scientific identities to deterministic gzip objects in objects/, with compressed/uncompressed SHA-256 hashes and byte counts.
• coverage.json indexes all table and figure environments by monograph label, final number and chapter subject, along with mathematical statements and conditional empirical claim scopes.
• displays/ logical records transcribe printed table cells at their displayed precision and figure captions/axis annotations. Related numerical-record links aid navigation. They are subject associations, not a claim that the complete raw plotting dependency graph or every underlying point is deposited.
• checks/cache-reconstruction/ logical records give the bounded raw-array reconstruction result, analysis and independent verification. They do not include the external raw arrays or claim native model replay.
• data-dictionary.json records observables, units, centering, denominators, independent statistical units, nested measurements and pairing.
• raw-inputs.json identifies separate raw-array and native replay inputs by portable acquisition role. Large input hashes also occur in protocol records.
• normalization.json binds exported records to acquisition-record content hashes and specifies the metadata policy. Local execution locations and diagnostic environment strings are omitted or replaced by logical roles. Device identifiers and acquisition dates are omitted. Elapsed durations, training-time coordinates, seeds and source positions remain available.
• manifest.json, SHA256SUMS and validation.json bind the prepared dataset and record the numerical-equivalence and coverage checks.

Families row, rg and model identify row dynamics, chronological renormalization and model-wide dynamics. They do not imply independent datasets or replications. Numerical values, signs, uncertainties, seeds, counts, source positions and outcomes are preserved. Normalized metadata changes record bytes; new object hashes are provided. A normalized protocol documents an acquisition but does not replace the immutable protocol required for raw reanalysis.

The 30-cell cache confirmation retains eighteen recalibrated aggregate passes, six zero-control transfer passes and six positive-control initial-cache transfer failures, including context exceptions. Other failed precision gates, rejected reductions, null controls and finite-scope qualifications remain reported.

Single-pass consumption of distinct registered RefinedWeb target blocks is the primary training law. Repeated-corpus controls remain separate. Heads, layers, contexts and times are nested or paired; they are not extra independent model replicas. Source-selection records include identifiers, hashes and positions, not corpus contents. Consult each protocol and the monograph for conditioning, calibration/assessment roles and uncertainty conventions.

This is a reported-evidence deposit. Raw-array reduction and native replay need the separately identified assets and the code companion's admission checks. Integrity checking and display transcription do not constitute new scientific replication. Rights and attribution remain with their respective materials; this deposit makes no blanket third-party data-license grant.

Published PDF destinations

The mathematical entries in coverage.json retain the statement's stable id, printed number, original-build anchor, and shared counter. Statement aliases and nested equation/clause labels remain separate. For navigation in the published monograph, use published_anchor and published_pdf_page. The latter is a one-based PDF page index, including front matter.

The top-level published_pdf record identifies arXiv 2609.34130v1, its versioned PDF URL, SHA-256 and page count. All 263 numbered statements have a published destination. The original-build anchors are retained for provenance; 212 of them differ from the published destinations. This mapping concerns mathematical statements; table/figure coverage and numerical records retain their existing meanings.

These entries agree with the code companion's provenance/statement-manifest.json. Its check_formal_manifest.py --data-repo check verifies code/data correspondence using the standard library. To check the actual PDF destinations and pages, run from the code repository:

python3 -m pip install -r requirements-publication.txt
curl -fL https://arxiv.org/pdf/2609.34130v1 -o /tmp/pldr-monograph-2609.34130v1.pdf
python3 scripts/check_published_pdf.py --pdf /tmp/pldr-monograph-2609.34130v1.pdf --data-repo DATA-REPO --output validation/published-pdf.json

The PDF checker authenticates the supplied PDF before resolving its destinations. It uses the separately pinned pypdf package; the evidence reader continues to need only the standard library. Use corresponding code and dataset revisions that include the published-PDF mapping.

Citation

CITATION.cff provides the preferred monograph citation with the arXiv v1 identifier. Also record the dataset Git revision and the manifest.json payload SHA-256 used. These identify the evidence release separately from the manuscript version. Coverage numbers refer to the monograph.

Release validation

RELEASE.md describes atomic metadata refresh and validation through both code companions. VALIDATION.md identifies the retained historical execution record; it is not a claim about a later branch checkout.

Public reference identities

Use the descriptive identities in index.json for lookup and extraction. The public-references.json catalogue distinguishes pldr-data: identities that resolve to indexed public records, pldr-code: identities that resolve to public companion source files, and urn:pldr:unavailable: identities for raw acquisition inputs or historical artifacts that are not distributed. An unavailable identity is not a local path or a download promise. Its locator digest preserves distinct historical roles without exposing internal filenames; it is not a substitute for an artifact-content hash.

normalization.json retains each original acquisition-record hash and identifies the exported bytes. export-provenance.json identifies the historical repository. Recorded seals continue to refer to their original acquisition bytes; normalized records do not claim a new seal, authorization, or preregistration. Numerical values, types, observations, scientific controls and outcomes remain unchanged. Current readers use the current index. Historical aliases and raw-input replay require the corresponding historical software and exact inputs.