Skip to main content

Open Apache-2.0 numeric-format SSOT bridge from t27 (GoldenFloat + MXFP4) to tt-lang kernels.

Project description

tt-lang-t27

License: Apache-2.0 arXiv:2606.05017

Open numeric-format SSOT bridge from gHashTag/t27 (GoldenFloat family, Apache-2.0, Tiny Tapeout silicon) into tenstorrent/tt-lang kernel author workflows.

What this is

A pure-Python side-package -- zero coupling with tt-lang upstream. Reads the t27 FORMAT-SPEC-001.json registry, validates kernel functions against bit-precise conformance vectors, and (optionally) hands off to a Coq formal oracle.

What this is NOT

  • Not a fork of tt-lang.
  • Not a patch to tt-lang.
  • Not a build-time dependency of tt-lang.
  • Not a CLA / copyright crossing.

Both repos remain Apache-2.0, separately governed.

Install

pip install tt-lang-t27       # from PyPI (planned)
# or
pip install git+https://github.com/gHashTag/tt-lang-t27

No Coq / no MLIR / no ttnn dependencies.

Quickstart

from tt_lang_t27 import t27_kernel, load_registry, load_vectors

@t27_kernel(
    fmt="GF16",
    registry_path="format-spec-001.json",
    vectors_path="gf16_conformance_v0.json",
)
def my_matmul(a, b):
    return a @ b

Every call logs a deterministic provenance tag:

[t27] kernel=my_matmul fmt=GF16 ssot=d9b76c5 tag=0a5ecb845def
[t27]   vector-check n=21 fmt=GF16

CLI

tt-lang-t27-conform \
  --registry format-spec-001.json \
  --vectors  gf16_conformance_v0.json
# OK conform=true reasons=0 sha256=<hex>

tt-lang-t27-mxfp4-conform \
  --vectors  mxfp4_conformance_v0.json
# OK mxfp4_conform=true reasons=0 sha256=<hex>

MXFP4 cross-validation (new in v0.2)

tt_lang_t27.mxfp4 is a pure-Python reference codec for OCP MXFP4 (S1E2M1, block_size=32, E8M0 shared scale). Constants pinned to tt-metal kMxFp4Params:

  • block_size = 32
  • scale_bias = 0x7F (E8M0 bias 127)
  • elem_exp_bits = 2, elem_man_bits = 1, elem_exp_bias = 1
  • saturation sat_pos_bits = 0x7, sat_neg_bits = 0xF
  • inf / nan: NotRepresentable

The vectors/mxfp4_conformance_v0.json pack carries 12 32-element blocks with expected scale byte + 32 nibbles + packed bytes hex. Any reference implementation (including tt-metal's pack_as_mxfp4_tiles<float>) can verify bit-exact parity by re-encoding each input_f32 and comparing to expected_bytes_hex.

Provenance tag formula

sha256(kernel_name || ssot_commit || fmt || anchor_hash)

anchor_hash = SHA-256("phi^2 + 1/phi^2 = 3").

References

Anchor

phi^2 + 1/phi^2 = 3 = L_2 (Lucas number).

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

tt_lang_t27-0.2.0.tar.gz (16.8 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

tt_lang_t27-0.2.0-py3-none-any.whl (13.6 kB view details)

Uploaded Python 3

File details

Details for the file tt_lang_t27-0.2.0.tar.gz.

File metadata

  • Download URL: tt_lang_t27-0.2.0.tar.gz
  • Upload date:
  • Size: 16.8 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/6.1.0 CPython/3.13.12

File hashes

Hashes for tt_lang_t27-0.2.0.tar.gz
Algorithm Hash digest
SHA256 edf0429439f688cdc7f7c23353bffa167b210f97d6402972aab0a3e9ad0494a6
MD5 20b29073fa549d959c207554c81857de
BLAKE2b-256 d763732a391e98467a258cb093cf80bbb9d7f6c1c0dd8874aec93f65187fa4f2

See more details on using hashes here.

Provenance

The following attestation bundles were made for tt_lang_t27-0.2.0.tar.gz:

Publisher: publish.yml on gHashTag/tt-lang-t27

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

File details

Details for the file tt_lang_t27-0.2.0-py3-none-any.whl.

File metadata

  • Download URL: tt_lang_t27-0.2.0-py3-none-any.whl
  • Upload date:
  • Size: 13.6 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/6.1.0 CPython/3.13.12

File hashes

Hashes for tt_lang_t27-0.2.0-py3-none-any.whl
Algorithm Hash digest
SHA256 7ad1eb5a52fefba127a2003eaf34ccc5452bcd37b626613157285ae5b345f406
MD5 3bb780f8177b47fdd96f36489e4926cf
BLAKE2b-256 771c71bb9dfbf5b6f6572c807c550691aefaf36032a130ff0a06f599a80c8139

See more details on using hashes here.

Provenance

The following attestation bundles were made for tt_lang_t27-0.2.0-py3-none-any.whl:

Publisher: publish.yml on gHashTag/tt-lang-t27

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page