Datasets:

Thox-ai
/

thox-whitepapers

Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
$schema: string
version: string
generated: timestamp[s]
description: string
changelog: struct<1.1.0: string, 1.0.0: string>
  child 0, 1.1.0: string
  child 1, 1.0.0: string
policy: struct<rule: string, precedence: string, on_unknown: string>
  child 0, rule: string
  child 1, precedence: string
  child 2, on_unknown: string
verified: struct<entity: struct<legal_name: string, styled_name: string, type: string, jurisdiction: string, m (... 1287 chars omitted)
  child 0, entity: struct<legal_name: string, styled_name: string, type: string, jurisdiction: string, management: stri (... 71 chars omitted)
      child 0, legal_name: string
      child 1, styled_name: string
      child 2, type: string
      child 3, jurisdiction: string
      child 4, management: string
      child 5, public_tagline_on_location: string
      child 6, source: string
      child 7, caveat: string
  child 1, mission: struct<statement: string, source: string>
      child 0, statement: string
      child 1, source: string
  child 2, web: struct<website: string, team_page: string, github_org: string, huggingface_orgs: list<item: string>, (... 39 chars omitted)
      child 0, website: string
      child 1, team_page: string
      child 2, github_org: string
      child 3, huggingface_orgs: list<item: string>
          child 0, item: string
      child 4, press_contact: string
      child 5, source: string
  child 3, what_it_is: struct<summary: string, compliance_positioning: string, verticals: list<item: st
...
thox_ai: list<item: string>
          child 0, item: string
      child 3, source: string
  child 10, funding_status: struct<raised: string, in_preparation: string, critical_distinction: string, source: string>
      child 0, raised: string
      child 1, in_preparation: string
      child 2, critical_distinction: string
      child 3, source: string
  child 11, pricing_note: struct<rule: string, reason: string>
      child 0, rule: string
      child 1, reason: string
unknown: list<item: struct<topic: string, status: string, may_say: string, must_not_say: string, reason: stri (... 4 chars omitted)
  child 0, item: struct<topic: string, status: string, may_say: string, must_not_say: string, reason: string>
      child 0, topic: string
      child 1, status: string
      child 2, may_say: string
      child 3, must_not_say: string
      child 4, reason: string
known_false: list<item: struct<claim: string, verdict: string, note: string>>
  child 0, item: struct<claim: string, verdict: string, note: string>
      child 0, claim: string
      child 1, verdict: string
      child 2, note: string
accidentally_correct: list<item: struct<claim: string, verdict: string, note: string>>
  child 0, item: struct<claim: string, verdict: string, note: string>
      child 0, claim: string
      child 1, verdict: string
      child 2, note: string
repo: string
sha256: string
card_version: string
ref: string
commit: string
_comment: string
synced_at: timestamp[s]
path: string
_ref_note: string
to
{'_comment': Value('string'), '_ref_note': Value('string'), 'repo': Value('string'), 'ref': Value('string'), 'commit': Value('string'), 'path': Value('string'), 'sha256': Value('string'), 'synced_at': Value('timestamp[s]'), 'card_version': Value('string')}
because column names don't match
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/utils.py", line 149, 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 129, 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 544, 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 3052, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2559, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2584, 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 558, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 440, 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 361, 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 2381, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2309, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              $schema: string
              version: string
              generated: timestamp[s]
              description: string
              changelog: struct<1.1.0: string, 1.0.0: string>
                child 0, 1.1.0: string
                child 1, 1.0.0: string
              policy: struct<rule: string, precedence: string, on_unknown: string>
                child 0, rule: string
                child 1, precedence: string
                child 2, on_unknown: string
              verified: struct<entity: struct<legal_name: string, styled_name: string, type: string, jurisdiction: string, m (... 1287 chars omitted)
                child 0, entity: struct<legal_name: string, styled_name: string, type: string, jurisdiction: string, management: stri (... 71 chars omitted)
                    child 0, legal_name: string
                    child 1, styled_name: string
                    child 2, type: string
                    child 3, jurisdiction: string
                    child 4, management: string
                    child 5, public_tagline_on_location: string
                    child 6, source: string
                    child 7, caveat: string
                child 1, mission: struct<statement: string, source: string>
                    child 0, statement: string
                    child 1, source: string
                child 2, web: struct<website: string, team_page: string, github_org: string, huggingface_orgs: list<item: string>, (... 39 chars omitted)
                    child 0, website: string
                    child 1, team_page: string
                    child 2, github_org: string
                    child 3, huggingface_orgs: list<item: string>
                        child 0, item: string
                    child 4, press_contact: string
                    child 5, source: string
                child 3, what_it_is: struct<summary: string, compliance_positioning: string, verticals: list<item: st
              ...
              thox_ai: list<item: string>
                        child 0, item: string
                    child 3, source: string
                child 10, funding_status: struct<raised: string, in_preparation: string, critical_distinction: string, source: string>
                    child 0, raised: string
                    child 1, in_preparation: string
                    child 2, critical_distinction: string
                    child 3, source: string
                child 11, pricing_note: struct<rule: string, reason: string>
                    child 0, rule: string
                    child 1, reason: string
              unknown: list<item: struct<topic: string, status: string, may_say: string, must_not_say: string, reason: stri (... 4 chars omitted)
                child 0, item: struct<topic: string, status: string, may_say: string, must_not_say: string, reason: string>
                    child 0, topic: string
                    child 1, status: string
                    child 2, may_say: string
                    child 3, must_not_say: string
                    child 4, reason: string
              known_false: list<item: struct<claim: string, verdict: string, note: string>>
                child 0, item: struct<claim: string, verdict: string, note: string>
                    child 0, claim: string
                    child 1, verdict: string
                    child 2, note: string
              accidentally_correct: list<item: struct<claim: string, verdict: string, note: string>>
                child 0, item: struct<claim: string, verdict: string, note: string>
                    child 0, claim: string
                    child 1, verdict: string
                    child 2, note: string
              repo: string
              sha256: string
              card_version: string
              ref: string
              commit: string
              _comment: string
              synced_at: timestamp[s]
              path: string
              _ref_note: string
              to
              {'_comment': Value('string'), '_ref_note': Value('string'), 'repo': Value('string'), 'ref': Value('string'), 'commit': Value('string'), 'path': Value('string'), 'sha256': Value('string'), 'synced_at': Value('timestamp[s]'), 'card_version': Value('string')}
              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.

THOX Whitepapers

Technical whitepapers from THOX.ai's in-house model programs: how the device models behind ThoxAir, ThoxMini, ThoxClip, and ThoxKey are selected, trained, evaluated, and published. Every result in these papers is stated as measured; held or unpromoted models are reported with the same prominence as promoted ones.

Your AI. Your Data. Your Rules.™

Paper	Current edition	Program	Models covered
Flagships for Every Device: The Wave C Training Program (v0.4.2)	v0.4.2, October 30, 2026 — latest alias	Wave C flagship program + C.1/C.2 correction cycles	ThoxAir-0.5B (promoted, Wave C.2 build), ThoxEdge-0.8B (promoted), ThoxMini-Phi4Mini (documented baseline), ThoxClip-Qwen3-0.6B (documented baseline)
First Device Models, Trained In-House (second edition)	Second (hardened) edition, October 30, 2026 — latest alias	Waves A–B: training rails, voice-command pilot, edge fast-track wave	ThoxFlux-350M, ThoxEdge-0.8B, ThoxEdge-0.5B, ThoxNano-360M, ThoxNano-135M, LFM3 voice-command adapter
MeshStack v2: Distributed Inference for the Private Device Mesh (v1.0.0)	v1.0.0, October 30, 2026 — latest alias	MeshStack v2 distributed inference (design paper)	Reference implementation: pipeline sharder, KV graph, secrets grant DAG, KV-affinity scheduler, mesh session manager

Version integrity

Every versioned filename in this dataset serves its own edition permanently. Earlier editions are never overwritten: when a paper is revised, the new edition is published under its own versioned filename and only the _latest alias moves to the current edition. (An earlier link-stability practice that replaced older versioned files was reversed on October 30, 2026; each restored edition was verified by anonymous download against its local build.)

The papers

Flagships for Every Device: The Wave C Training Program — v0.4.2

The flagship program record: base-model slate selection under the THOX license policy, the 49-question factuality gate (GATE49) with pre-registered publish bars, two corpus correction cycles, the fp-LoRA vs QLoRA recipe measurement, and the closing per-slot state of every device model. Program closed with two of four slots promoted.

Edition: v0.4.2, 47 pages, 588,002 bytes, SHA-256 cea173de0b3ec1417c08887a20f6b23bb188b4a91e68f5909dc034d1ae2e8918.
Earlier editions remain at their own filenames: v0.3.0, v0.4.0, and v0.4.1.
v0.4.2 is the post-publication review revision. It integrates the Research Loop requirements audit (whose fixes landed in v0.4.1), an external hostile review, and a ChatGPT-lane (Imogen) review packet: publish bars are restated against the measured null policy with per-category null results, the promotion decision function and its deviations are printed, the fix-cycle scores carry their development-set status, and a new section reports GATE49 and probe results for the artifacts exactly as served (via Ollama) before and after a template/capability/facts-block fix, together with bare and grounded base-model baselines. Every review finding is dispositioned (adopted / adopted in part / not adopted, with reasons) in the program's review record.
Supplement: wavec-supplement/ (83 files) — the vendored GATE49 bundle (scorer, facts package, facts card v1.1.0, corrected grounding prompt, probe), the per-run gate and probe reports for every program run, the served-path and baseline reports (pre- and post-fix), the null-policy report, the registry entries the runs were launched from, and a SHA-256 manifest.
Models: Thox-ai/ThoxAir-0.5B-WaveC2-Merged, Thox-ai/ThoxEdge-0.8B-WaveC-Merged, Thox-ai/ThoxMini-Phi4Mini, Thox-ai/ThoxClip-Qwen3-0.6B; Ollama: thox-ai/thoxair-0.5b, thox-ai/thoxedge-0.8b, thox-ai/thoxmini-phi4mini, thox-ai/thoxclip-qwen3-0.6b.

First Device Models, Trained In-House — second edition

The predecessor paper: the training and verification rails, the voice-command pilot, and the first edge wave (Waves A–B) that established the factuality gate and the honest-reporting standard the Wave C program runs on.

Edition: second (hardened) edition, 44 pages, SHA-256 f10fe70802316bd4ff62b8bf2bec9f1023c55a47d0dd719d935415f13460b178, served at both the versioned filename above and the _latest alias; a DOCX of the same edition is included.
The second edition applies the post-publication Research Loop audit and external hostile review in full: the failure taxonomy is complete (seven classes), a measured constant-abstainer floor (25/49 through the real scorer) is reported beside the results, the development-set status of the gate is stated, and the paper's earlier printed Clip commit identifier is corrected to the Hub-verified history with the discrepancy disclosed.
Supplement: waves-ab-supplement/ — the GATE49 bundle, the six gating Modelfiles, the constant-abstainer outputs, a README, and a manifest. The supplement records honestly that Wave A/B per-question gate outputs were not retained on the run hosts and were not reconstructed.

MeshStack v2: Distributed Inference for the Private Device Mesh — v1.0.0

A design paper, in a different register from the two program records above: it specifies how a model too large for any one THOX device is sharded across the private device mesh — pipeline parallelism as the mesh form, a KV graph with grant-gated cache transfer, a secrets grant DAG, and a KV-affinity scheduler — against an audited account of what the fleet's serving paths do today.

Edition: v1.0.0, 26 pages, 853,496 bytes, SHA-256 b290f187a8edde5b9e5c5c111e6c619adeb88c4f649a67d00b27c99b459a727a, served at its versioned filename under meshstack/; the meshstack/ _latest alias carries the current edition, and a DOCX of the same edition is included.
Status, stated plainly: the implementation behind the paper is a reference implementation with simulation semantics. The paper makes no performance claims; benchmark measurements are a future, separately gated wave, and no artifact described has been qualified on physical device hardware.
Paper repository: ttracx/thox-meshstack-v2-whitepaper (release v1.0.0); implementation repository: ttracx/thox-meshstack-inference.

License

Publisher

THOX.AI LLC — thox.ai — press@thox.ai

Downloads last month: 313

Total file size:

9.11 MB