Hugging Face's logo

Datasets:
malaiwah
/
qwen38-27b-fidelity-suite-v3

Tasks:
Text Generation
Languages:
English
German
French
Size:
n<1K
Tags:
quantization
evaluation
kl-divergence
distribution-fidelity
exl3
nvfp4
License:
Dataset card Files
xet
 Community
Dataset Viewer
Duplicate
The dataset viewer is not available for this split.
Cannot load the dataset split (in streaming mode) to extract the first rows.
Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
schema: string
a: struct<report: string, label: string, mean: double>
  child 0, report: string
  child 1, label: string
  child 2, mean: double
b: struct<report: string, label: string, mean: double>
  child 0, report: string
  child 1, label: string
  child 2, mean: double
contexts: int64
difference_a_minus_b: double
median_difference: double
bootstrap_difference: struct<mean: double, ci95_low: double, ci95_high: double, clusters: int64, samples: int64>
  child 0, mean: double
  child 1, ci95_low: double
  child 2, ci95_high: double
  child 3, clusters: int64
  child 4, samples: int64
a_wins: int64
b_wins: int64
largest_a_advantage: list<item: list<item: double>>
  child 0, item: list<item: double>
      child 0, item: double
largest_b_advantage: list<item: list<item: double>>
  child 0, item: list<item: double>
      child 0, item: double
model: string
elapsed_sec: double
quantization: string
captures: list<item: struct<index: int64, sha256: string, shape: list<item: int64>>>
  child 0, item: struct<index: int64, sha256: string, shape: list<item: int64>>
      child 0, index: int64
      child 1, sha256: string
      child 2, shape: list<item: int64>
          child 0, item: int64
kv_cache_dtype: string
suite_token_sha256: string
quantization_config: string
to
{'model': Value('string'), 'quantization': Value('string'), 'quantization_config': Value('string'), 'kv_cache_dtype': Value('string'), 'suite_token_sha256': Value('string'), 'contexts': Value('int64'), 'captures': List({'index': Value('int64'), 'sha256': Value('string'), 'shape': List(Value('int64'))}), 'elapsed_sec': Value('float64')}
because column names don't match
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/utils.py", line 147, in get_rows_or_raise
                  return get_rows(
                      dataset=dataset,
                  ...<4 lines>...
                      column_names=column_names,
                  )
                File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
                  return func(*args, **kwargs)
                File "/src/services/worker/src/worker/utils.py", line 127, in get_rows
                  rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
                File "/src/services/worker/src/worker/utils.py", line 483, in safe_iter
                  yield from ds.decode(False) if ds.features else ds
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2840, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2373, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2398, in _iter_arrow
                  for key, pa_table in self.ex_iterable._iter_arrow():
                                       ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 536, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 419, in _iter_arrow
                  for key, pa_table in self.generate_tables_fn(**gen_kwags):
                                       ~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 343, in _generate_tables
                  self._cast_table(pa_table, json_field_paths=json_field_paths),
                  ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
                  pa_table = table_cast(pa_table, self.info.features.arrow_schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2378, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2306, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              schema: string
              a: struct<report: string, label: string, mean: double>
                child 0, report: string
                child 1, label: string
                child 2, mean: double
              b: struct<report: string, label: string, mean: double>
                child 0, report: string
                child 1, label: string
                child 2, mean: double
              contexts: int64
              difference_a_minus_b: double
              median_difference: double
              bootstrap_difference: struct<mean: double, ci95_low: double, ci95_high: double, clusters: int64, samples: int64>
                child 0, mean: double
                child 1, ci95_low: double
                child 2, ci95_high: double
                child 3, clusters: int64
                child 4, samples: int64
              a_wins: int64
              b_wins: int64
              largest_a_advantage: list<item: list<item: double>>
                child 0, item: list<item: double>
                    child 0, item: double
              largest_b_advantage: list<item: list<item: double>>
                child 0, item: list<item: double>
                    child 0, item: double
              model: string
              elapsed_sec: double
              quantization: string
              captures: list<item: struct<index: int64, sha256: string, shape: list<item: int64>>>
                child 0, item: struct<index: int64, sha256: string, shape: list<item: int64>>
                    child 0, index: int64
                    child 1, sha256: string
                    child 2, shape: list<item: int64>
                        child 0, item: int64
              kv_cache_dtype: string
              suite_token_sha256: string
              quantization_config: string
              to
              {'model': Value('string'), 'quantization': Value('string'), 'quantization_config': Value('string'), 'kv_cache_dtype': Value('string'), 'suite_token_sha256': Value('string'), 'contexts': Value('int64'), 'captures': List({'index': Value('int64'), 'sha256': Value('string'), 'shape': List(Value('int64'))}), 'elapsed_sec': Value('float64')}
              because column names don't match

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

Qwen3.8-27B distribution-fidelity suite v3 (held-out, 181 x 2048)

The frozen evaluation suite and the captured hidden states that let anyone recompute or contest the KL-divergence numbers published for malaiwah/Qwen3.8-27B-K4 — without a GPU, without downloading any model, and without trusting the publisher.

Scope, stated up front. The captures in this snapshot are the iteration-1 set: malaiwah/Qwen3.8-27B-K4, unsloth/Qwen3.8-27B-NVFP4 and Qwen/Qwen3.8-27B-FP8 against the BF16 reference. The current published checkpoint, malaiwah/Qwen3.8-27B-EXL3-K5K6 (mean KLD 0.008157), was measured on this same suite and this same LM head, but its captures are not in this dataset yet — see What is missing.

Protocol adopted from Kimi-K3 distribution fidelity: capture the BF16 hidden state after the final RMSNorm and before the LM head, then reconstruct complete full-vocabulary distributions offline through one shared BF16 LM head. Storage is 21 MB per context instead of 2.03 GB of fp32 logits, which is what makes many-context evaluation affordable.

What is measured

KL(BF16 reference || candidate) over the entire 248,320-token vocabulary, no top-k, two passes (log-sum-exp normalisers and argmax first, then divergence), float32 within a vocabulary chunk accumulated in float64 across chunks, source-cluster bootstrap. Also Jensen-Shannon divergence, top-1 agreement, tail quantiles, and per-stratum means.

Because both operands go through the same shared head, no candidate's own head quantization is counted. That is what makes the three candidates comparable, and it is why these are body-only numbers.

Headline results reproducible from these files

Analysis partition, 136 contexts, 278,392 scored positions (iteration 1):

candidate weights mean KLD bootstrap 95 % CI median top-1 agreement
Qwen/Qwen3.8-27B-FP8 30.9 GB 0.013126 [0.00981, 0.01709] 0.002343 96.22 %
malaiwah/Qwen3.8-27B-K4 19.2 GB 0.030736 [0.02238, 0.04073] 0.004218 94.50 %
unsloth/Qwen3.8-27B-NVFP4 23.4 GB 0.094978 [0.06858, 0.12688] 0.012911 90.53 %

Paired over the same contexts: K4 beats NVFP4 by 0.064242 (95 % CI [-0.08621, -0.04611], 136/136 contexts); FP8 beats K4 by 0.017611 (95 % CI [0.01256, 0.02368], 136/136 contexts).

Layout

Total 17.75 GiB. Every count and size below is the actual content of this revision.

suite-manifest.json          181 contexts: stratum, source cluster, partition, sentinel
                             flag, token SHA-256, contamination-scan result;
                             partitions 136 analysis / 45 qualification / 32 sentinels
tokens/                      181 x context-NNNN.json, the exact token IDs (2048 each)
reference-hidden/            181 x hidden_NNNN.safetensors, BF16 [2047, 5120], from
                             Qwen/Qwen3.8-27B, plus capture-manifest.json (per-file
                             SHA-256 + the runtime config used)         3.53 GiB
candidate-hidden/            three candidates, same shape and manifest  3.53 GiB each
  k4-online-k6/              malaiwah/Qwen3.8-27B-K4: EXL3 K4 MLP + online-K6 attention
  qwen-fp8/                  Qwen/Qwen3.8-27B-FP8
  unsloth-nvfp4/             unsloth/Qwen3.8-27B-NVFP4
sentinel-hidden/             two extra captures of the K4 runtime over the 32 sentinel
  repeat-02/, repeat-03/     contexts, for the noise floor              0.62 GiB each
lm-head/weight.safetensors   BF16 [248320, 5120], extracted from
                             Qwen/Qwen3.8-27B@1d4bf0f2, SHA-256 25a30fd5…dee4cfff  2.37 GiB
reports/                     8 JSON receipts:
                             report-{k4,fp8,nvfp4}-analysis.json  the three candidates
                             paired-k4-vs-{fp8,nvfp4}.json        the paired comparisons
                             noise-r1-vs-r2.json, noise-r2-vs-r3.json  the noise floor
                             qualification-bf16.json              live-vs-replay error
checksums.txt                SHA-256 of all 985 data files (everything but itself and
                             this README)

tokens/ is authoritative: retokenising source text does not reproduce the evaluation input. Verify integrity with sha256sum --check checksums.txt before use.

Reproduce a number 

git clone https://github.com/malaiwah/qwen38-27b-exl3
hf download malaiwah/qwen38-27b-fidelity-suite-v3 --repo-type dataset --local-dir suite-v3

# recompute our published K4 result from the shipped captures (no model needed)
python qwen38-27b-exl3/tools/fidelity.py replay \
  --reference suite-v3/reference-hidden \
  --candidate suite-v3/candidate-hidden/k4-online-k6 \
  --head suite-v3/lm-head/weight.safetensors \
  --suite suite-v3 --filter analysis --out my-k4.json

# then compare receipts
python qwen38-27b-exl3/tools/fidelity.py paired \
  --a my-k4.json --b suite-v3/reports/report-nvfp4-analysis.json \
  --a-label mine --b-label nvfp4 --out my-paired.json

To score your own checkpoint, capture its hidden states over the same tokens (fidelity.py capture --suite suite-v3 ...) and replay against reference-hidden/.

Held-out corpus, and the contamination scan

stratum contexts source
literary 52 Project Gutenberg, public domain
code 52 CPython Lib/ at tag v3.12.8, PSF licence
scientific 52 arXiv abstracts via the arXiv API
encyclopedic 17 Wikipedia extracts, English, CC BY-SA 4.0
multilingual 8 Wikipedia extracts in de, fr, es, ja, zh, ru, it, pt, CC BY-SA 4.0

Why held out matters: the previous version of this suite was built from exllamav3's bundled calibration corpora — the same text the EXL3 candidate was calibrated on, while the NVFP4 and FP8 candidates were calibrated elsewhere. That biased the comparison. Re-measuring on held-out text moved our own number from 0.026231 to 0.030736 (+17 %) and moved FP8's from 0.019309 to 0.013126 (-32 %). The published v3 numbers are the honest ones.

A 160-character shingle scan of every context against every exllamav3 calibration corpus reports 0 contaminated contexts, 0 hits (67,818 calibration shingles). Recorded in suite-manifest.json -> contamination_scan.

Token IDs are a lossless encoding of the source text, so the Wikipedia-derived strata carry CC BY-SA 4.0 attribution requirements; the Gutenberg strata are public domain; CPython is PSF-licensed; arXiv abstracts remain under their authors' terms. lm-head/weight.safetensors is one tensor extracted from Qwen/Qwen3.8-27B (Apache-2.0) and is redistributed under that licence.

Controls shipped with the data

control receipt result why it matters
runtime-repeat noise floor reports/noise-r1-vs-r2.json, reports/noise-r2-vs-r3.json 0.000000 mean KLD between captures 1-2 and 2-3 of the same runtime over 32 sentinel contexts, top-1 1.0 every reported difference is far above measurement noise; this runtime is bit-deterministic across process restarts, unlike the TP16 reference runtime whose floor was 0.0032
harness self-check reproducible with fidelity.py replay reference scored against itself returns exactly 0.000000 no bias in densification or window handling
replay qualification reports/qualification-bf16.json mean KL(live served logits || replayed logits) = 6.54e-04, top-1 98.999 % over 6 contexts the replay path is not exact; see caveat 1

The withdrawn CUDA-graph control

Earlier revisions of this card listed a fourth control, "CUDA-graph parity 0.000000 between graph and eager captures of the same checkpoint". It is withdrawn and is not shipped here. fidelity.py capture takes a single prefill forward, and the graph mode under test (cudagraph_mode=FULL_DECODE_ONLY) captures no prefill graph, so that measurement compared two runs of the same eager prefill code and could not have measured the decode path at all. The only claim it supports is "enabling graph decode does not change prefill numerics".

Re-measured on real decode steps (32 prompts x 32 greedy tokens, temperature 0, fixed seed, via the serving endpoint), graph and eager agree on 24/32 exact sequences with mean |delta logprob| 0.0118 on the chosen token, and each mode is internally deterministic (32/32 self-repeat). Unquantised BF16 on the same build drifts by the same amount (24/32, 0.0128), so the drift belongs to this build's CUDA-graph decode path and is not caused by the quantisation. Full result: docs/27-graph-decode-drift-control.md. These are decode-sequence receipts, not hidden-state captures, so they live in the research repo (receipts/decode-parity-*.json) rather than in this dataset.

What this snapshot now contains, and what it still does not

Added (this revision): the whole K5/K6 family's hidden-state captures, so every headline in the family's model cards is independently recomputable without a GPU or a checkpoint download:

directory build mean KLD (analysis partition)
candidate-hidden/k5k6-hydrated-offline-k6 attention serialized offline at K6 0.007406
candidate-hidden/k5k6-online-k6 attention BF16 on disk, encoded K6 at load 0.008157
candidate-hidden/k5k6-online-k5 same download, overlay K5 0.012135
candidate-hidden/k5k6-online-k4 same download, overlay K4 0.027530
candidate-hidden/k4-online-k6 iteration-1 build 0.030736
candidate-hidden/qwen-fp8 official FP8 0.013126
candidate-hidden/unsloth-nvfp4 Unsloth NVFP4 0.094978

reports-k5k6/ carries the matching replay reports and every paired receipt, including the two as-served (asymmetric-head) reports that attribute the K6 head: +0.000127 on the online build, +0.000125 on the hydrated one.

Still missing, stated plainly:

  • A post-selection result. Every number here is on the 136-context analysis partition, which guided recipe selection, and this suite's qualification partition is not source-disjoint from it: all 27 qualification clusters also appear in analysis, because the builder split contexts rather than source clusters. A v4 suite is being built from new documents with a group split; until it lands, treat these as development-set numbers.
  • Language coverage is narrower than the tags suggest. The multilingual stratum is 6 German and 1 Russian context; the builder tolerated under-filled strata. It now fails instead, and v4 fetches more languages.
  • No downstream task, multimodal-quality or long-context accuracy data. This dataset measures distribution divergence only.
  • No near-duplicate or semantic contamination scan. The exact 160-character shingle scan found zero overlap with exllamav3's calibration corpus, and an independent reviewer reproduced that at 80 characters too, but near-duplicates are untested.

Caveats, stated plainly

  1. Replay is not exact. Our live-vs-replayed qualification is 6.54e-04, about 500x worse than the 1.23e-06 the reference protocol reports on Kimi-K3. This was tested: storing the hidden states in fp32 instead of BF16 moved it only to 6.25e-04 (-4.5 %), so operand rounding is ~5 % of the floor and the rest is the implementation difference between the serving runtime's logit path and our replay. Paired comparisons are unaffected because both arms use the identical replay path, but absolute differences below ~1e-3 are not resolvable with these artifacts.
  2. 2048-token contexts only. Nothing here measures long context, and the model supports 262,144.
  3. No dialogue/instruction, mathematics/reasoning, or structured/tool-call strata — exactly the distributions a thinking model is used for. 41 source clusters, versus 827 in the reference artifact.
  4. KLD ranks candidates within this artifact's frozen identities. It does not substitute for coding, reasoning, long-context, multimodal, tool-use or free-running generation evaluation, and thresholds from other models, corpora or tokenizers do not transfer.

Provenance

item identity
reference model Qwen/Qwen3.8-27B @ 1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0
candidates malaiwah/Qwen3.8-27B-K4; unsloth/Qwen3.8-27B-NVFP4 @ 9c73e2da; Qwen/Qwen3.8-27B-FP8 @ 017b9c7a
runtime voipmonitor/vllm@sha256:820181fbbc975cd5291c411cda9771d58fecee1636d916f508f47230df20592b (Gilded Gnosis r34), vLLM 0.11.2.dev280+gilded.gnosis.v20...r34
hardware 1x RTX PRO 6000 Blackwell Server Edition, SM120, driver 595.58.03, TP1
capture forward hook on language_model.model.norm via collective_rpc; enforce_eager, max_num_seqs=1, one prefill chunk per context
suite token SHA-256 3f9d17f1b55f64872ad3ac19c8711654e09ba70b7ca14b0851525088fe735691
harness + docs https://github.com/malaiwah/qwen38-27b-exl3
Downloads last month
617
Total file size:
51 GB