GLM MoE DSA — fp8-ue8m0: a tiny toolchain fixture

A fp8-ue8m0 storage/reader fixture, built from a tiny random reference checkpoint. Use it to inspect packing, reconstruction, scope accounting, and own-head fidelity comparisons—not to choose a production quantizer.

These weights are untrained, not an assistant, and not a language-quality benchmark. Generated text has no useful semantic quality. No upstream trained weights or training data are implied by an architecture name.

Weight lineage

This is a quantization of a shared native checkpoint, not an independent random initialization and not a fine-tune. HF base_model / base_model_relation: quantized metadata names the real source model.

Exact construction source: malaiwah/glm-moe-dsa-tiny-random-bf16@a1a973330e4e912678adbd8ea41e8b86b1c9dccb.

All 8 derivatives in this family were measured against the same complete source checkpoint identity:

800be70604bef0c3adfb429774f44756d60bf23d141e82d24a032ce25ce0451e

That identity binds config and weight bytes; equal architecture names or random seeds are not used as a substitute. Quantized storage cases use round-to-nearest packing and do not claim GPTQ/AWQ/ModelOpt optimization or calibration.

Browse the Matched-Weight Quantization Families collection.

At a glance

Size is serialized artifact size, not runtime RAM or a capacity/performance guarantee. Packed-array element counts are not model parameter counts.

What is actually included

Four text decoder layers (one dense, three MoE), MLA attention, alternating full/shared DSA indexers, 8 routed experts/top-2 plus one shared expert, and a complete untied vocabulary head. Native FP32 router buffers are retained; no MTP.

Storage intervention: fp8-ue8m0; CPU RTN format fixture.

• GLM MoE DSA, dense and routed expert matrices; decoded BF16 execution, not full original quantized release
• Output-head policy: native. The comparison replays each side through its own sealed head, including any head error; the reference head is not substituted.
• Reconstructed weights only; no activation quantization, QAT, calibration or GPU kernel validation

Compressed-tensors / ModelOpt labels identify concrete packed component layouts. The stored scales, inverse scales and mixed constituents must all be admitted and consumed by the reader; there is no fallback that treats integer/FP8 storage as native BF16. Mixed formats are described per tensor in the scope rather than by a global bit-width claim.

The repository file inventory includes config.json, generation_config.json, tokenizer.json, tokenizer_config.json, scope.json, build-manifest.json. Exact build/runtime/license inventories are linked below. A file may describe historical provenance without being an executable entry point.

Good community uses—and boundaries

• Learn how to download pinned artifacts, seal a small synthetic token panel, capture hidden states and replay a full vocabulary head.
• Debug model-family adapters, strict tensor loading, storage decoders, or reproducibility tooling without downloading a production-sized checkpoint.
• Reproduce a narrowly scoped result, report an adapter/reader regression, and retain the source, panel and runtime identities needed to explain it.

Not established: trained-model accuracy; useful instruction following; quantizer optimization quality; original production-weight compatibility; GPU/NPU/serving-kernel parity; cross-hardware determinism; long-context behavior outside the recorded panel; throughput or paid-compute admission.

RTN is round-to-nearest packing, not optimization. These cases do not run GPTQ, AWQ or AutoRound optimization, calibrated ModelOpt/CT/QAT training, or activation-aware tuning. CPU reconstruction does not execute the original packed GEMM or serving activation arithmetic.

Model family, root dataset and evidence

Exact construction source: malaiwah/glm-moe-dsa-tiny-random-bf16@a1a973330e4e912678adbd8ea41e8b86b1c9dccb. This is the actual source pin; later documentation commits do not rewrite that history. Family fidelity-root dataset repository and storage evidence bundle. The root model is separate; this child must not be registered or represented as the base model. Canonical roots and their evidence are published; registry links distinguish native bases from child/control artifacts.

Recorded own-head result

Full comparison receipt includes per-context/tokenwise evidence, comparison gates and disclosures. Receipt self-digest: 4e03857ab1c31b045782a4f8258b26199041f212cf13d2f0eece2d6062fe6929. No threshold or usable determinism floor is invented here.

Material disclosures from the observed receipt:

• HEAD-1d: each side replayed through its own sealed head (reference 2364995d4009, candidate 2364995d4009); head error is inside the measurement, as under HEAD-2, and nothing is substituted. The heads are content-identical.
• candidate was captured from a WEIGHTS-ONLY RECONSTRUCTION (microfloat-weight-reconstruction, output bfloat16). The stored weights were decoded before the model forward; native quantized GEMM and serving activation arithmetic are not measured. Decoder provenance is in the sealed runtime receipt. The comparison is advisory.
• candidate was captured from a bf16 materialisation of its weights (microfloat-weight-reconstruction); the checkpoint declares activation quantization (dynamic, reader-declared activation quantization; see sealed decode evidence) that a weights-only capture does not apply, so a served deployment also quantizes activations at runtime. That term is not in this number, which is expected to understate the served divergence; it is not a mathematical bound. The comparison is advisory.

Runtime requirements and safe local reproduction

Replay the published evidence without loading a model

This uses QFS's existing NumPy FP64 comparator in a Torch-free environment. It reads the stored hidden states and each side's own head. No model forward, remote model code, GPU, upload or registry mutation is involved. Runtime receipts name the actual backend; last-bit differences from another FP64 implementation are not a new quality claim. Synthetic exact controls remain zero.

Set QFS to a reviewed Quant Fidelity Suite checkout and use Bash:

: "${QFS:?Set QFS to your reviewed QFS checkout}"
WORK=$(mktemp -d)
export OMP_NUM_THREADS=2 MKL_NUM_THREADS=2 OPENBLAS_NUM_THREADS=2
python3.12 -m venv "$WORK/replay-env"
"$WORK/replay-env/bin/pip" install 'numpy==2.5.3' 'huggingface-hub==1.30.0'
"$WORK/replay-env/bin/hf" download malaiwah/qfs-microfloat-tiny-cpu-format-v1 --repo-type dataset \
  --revision b1690c1b72beafc812f2286ec6a8fe2ced338edc --include 'raw/candidates/fp8-ue8m0/evidence/**' --local-dir "$WORK/evidence"
"$WORK/replay-env/bin/python" "$QFS/bin/fidelity_dataset.py" verify \
  "$WORK/evidence/raw/candidates/fp8-ue8m0/evidence/root" --verify-tensors
"$WORK/replay-env/bin/python" "$QFS/bin/fidelity_dataset.py" verify \
  "$WORK/evidence/raw/candidates/fp8-ue8m0/evidence/candidate" --verify-tensors
"$WORK/replay-env/bin/python" "$QFS/bin/fidelity_dataset.py" compare \
  --reference "$WORK/evidence/raw/candidates/fp8-ue8m0/evidence/root" \
  --candidate "$WORK/evidence/raw/candidates/fp8-ue8m0/evidence/candidate" \
  --out "$WORK/replayed" --device cpu --replay-device numpy --replay-dtype float32 \
  --vocab-chunk 8192 --verify-tensors --own-heads --force-compute

Reconstructed format comparisons are intentionally advisory and normally return exit code 2 while writing a valid receipt. Inspect that receipt; do not silence refusals or interpret an advisory result as a production-quality ranking.

Capture the actual checkpoint

Capture uses a separate pinned Torch CPU environment. The input below is the original token-panel format, not a sealed capture's internal panel/ folder. The native source, tokenizer and model revisions remain explicit. Set AUTHOR to your own HF handle; the dataset repository argument is attribution only and nothing is uploaded by these commands.

: "${AUTHOR:?Set AUTHOR to your Hugging Face handle}"
"$WORK/replay-env/bin/hf" download malaiwah/glm-moe-dsa-tiny-random-fp8-ue8m0-rtn-format --revision b35a3365ea79b5b5e72f778819ee2dc7610975ec --local-dir "$WORK/model"
"$WORK/replay-env/bin/hf" download malaiwah/glm-moe-dsa-tiny-random-bf16 --revision a1a973330e4e912678adbd8ea41e8b86b1c9dccb --local-dir "$WORK/source"
"$WORK/replay-env/bin/hf" download malaiwah/qfs-fixture-root-captures-v1 --repo-type dataset \
  --revision f53b204091c988ce4a2161af81886f3018745559 --include 'requirements-capture.txt' --include 'roots/glm_moe_dsa/input-panel/**' \
  --local-dir "$WORK/inputs"
python3.12 -m venv "$WORK/capture-env"
"$WORK/capture-env/bin/pip" install -r "$WORK/inputs/requirements-capture.txt"
"$WORK/capture-env/bin/python" "$QFS/bin/fidelity_dataset.py" architectures prepare \
  --architecture glm_moe_dsa --model-dir "$WORK/model" \
  --model-repository malaiwah/glm-moe-dsa-tiny-random-fp8-ue8m0-rtn-format --model-revision b35a3365ea79b5b5e72f778819ee2dc7610975ec \
  --panel "$WORK/inputs/roots/glm_moe_dsa/input-panel" --tokenizer-root "$WORK/source" \
  --author "$AUTHOR" --dataset-repository "$AUTHOR/glm-moe-dsa-tiny-random-fp8-ue8m0-rtn-format-capture" \
  --dataset-id "fidelity--$AUTHOR.glm-moe-dsa-tiny-random-fp8-ue8m0-rtn-format" --out "$WORK/workflow" \
  --role quant --scope-file "$WORK/model/scope.json" --codec fp8_e4m3 --bits 8.0 \
  --reference hf://malaiwah/glm-moe-dsa-tiny-fidelity-root-v1@b8ed427805abdfd6b8b20db10ca5107e840d0497

Inspect workflow.json. This launcher executes its exact capture/verification commands and selects the Torch-free interpreter only for the final comparison:

export OMP_NUM_THREADS=2 MKL_NUM_THREADS=2 OPENBLAS_NUM_THREADS=2
"$WORK/replay-env/bin/python" - "$WORK/workflow/workflow.json" <<'PY'
import json, subprocess, sys
workflow = json.load(open(sys.argv[1]))
for step in workflow["commands"]:
    argv = list(step["argv"])
    if step["step"] == "compare":
        argv[0] = sys.executable
    result = subprocess.run(argv)
    if result.returncode:
        raise SystemExit(result.returncode)
PY

Use the actual root at malaiwah/glm-moe-dsa-tiny-fidelity-root-v1@b8ed427805abdfd6b8b20db10ca5107e840d0497. The community collection groups models and captures. Custom code, where required above, is explicitly pinned and executed only after your consent; hash verification is provenance, not a sandbox.

Licensing and detailed provenance

The fixture repository retains its mit license in LICENSE. This covers the fixture only according to that exact text; architecture inspiration is not relicensing of upstream weights, cards, code, or configuration. Dependencies retain their own licenses (native Transformers implementation: Apache-2.0). No upstream trained tensors are claimed to be copied.

The exact source license is retained, unchanged, from malaiwah/glm-moe-dsa-tiny-random-bf16. The storage conversion does not grant new upstream rights.

Immutable provenance, historical cards, source inventories and full caveats

• Model/card snapshot used while authoring: malaiwah/glm-moe-dsa-tiny-random-fp8-ue8m0-rtn-format@b35a3365ea79b5b5e72f778819ee2dc7610975ec.
• Full preserved earlier model card, SHA-256 7b7c998d5f5d86f114c805a9bd823b73990541a66052752150d786f47598b956. This is an immutable historical record, including original build pins, measured resource tables, command logs, limitations and failed-attempt provenance. Historical publisher commands are not part of the local-use recipe above.
• build-manifest.json.
• qfs-source.json.
• cpu-reproduction-summary.json.
• scope.json.
• Retained reproduction tools/dependencies and source-inventory.json. Source paths in recorded argv are historical execution locations; restore the dataset’s raw trees and map them to your own workspace. Do not substitute moving upstream branches or an unreviewed publisher workflow.

Selected original provenance fields (full tensor/component evidence remains in the linked inventories):

{
  "architecture_inspiration": "GLM-5.2-SIQ-Fruit-bf16 geometry and dense/MoE plus full/shared DSA schedules; no Fruit or GLM model weights reused.",
  "generation": "Native GlmMoeDsaForCausalLM random initialization in FP32, rounded to the published BF16 artifact; native router correction buffers retained FP32.",
  "generator_sha256": "19c3489f1bc7e8e9323cb26afbead562816a6918d6823af252eb6d04dff8e81e",
  "limitations": [
    "Random initialization; not trained or distilled.",
    "Not a quantization or behavioral approximation of a released GLM model.",
    "MTP intentionally absent (zero next-n-prediction layers); no unmatched MTP tensors.",
    "This build manifest describes generated bytes, not proof of clean loading or capture; see separately measured verification evidence."
  ],
  "parameter_count": 277824,
  "seed": 20260907,
  "versions": {
    "safetensors": "0.8.0",
    "tokenizers": "0.23.2",
    "torch": "2.11.0+cpu",
    "transformers": "5.16.1"
  },
  "source": {
    "repository": "malaiwah/glm-moe-dsa-tiny-random-bf16",
    "revision": "a1a973330e4e912678adbd8ea41e8b86b1c9dccb"
  },
  "format": "fp8-ue8m0; CPU RTN format fixture",
  "classification": "lossy",
  "architecture_restrictions": [
    "GLM MoE DSA, dense and routed expert matrices; decoded BF16 execution, not full original quantized release"
  ],
  "activation_omissions": [
    "Reconstructed weights only; no activation quantization, QAT, calibration or GPU kernel validation"
  ],
  "head_policy": "native"
}