Repository navigation
fix(paper-gf16): честность — FPGA-synth вместо 'verified on silicon', shuttle TTSKY26b - #1340
Merged
Merged
Conversation
Contributor
|
📓 NotebookLM Notebook linked to this PR
This notebook contains session context, decisions, and artifacts for this work. |
Contributor
gHashTag
force-pushed
the
fix/gf16-paper-honesty-silicon-shuttle
branch
from
July 4, 2026 14:23
2d60082 to
b806ec0
Compare
Contributor
PR DashboardGenerated at: 2026-07-04 14:23:49 UTC
Summary
Seal Status
|
Contributor
|
📓 NotebookLM Notebook linked to this PR
This notebook contains session context, decisions, and artifacts for this work. |
… on silicon', shuttle TTSKY26b Статья GF16 (arXiv:2606.05017, черновик + tex-исходник) заявляла '4x4 matmul verified on silicon' при доказательстве через RTL/FPGA (Artix-7 XC7A100T) — противоречие honesty-правилу (encoding != compute != FPGA; sim-bitexact != HW; 'отправлен на изготовление' != 'измерено на кристалле'). Правки (обе: docs/arxiv-submission/trinity-gf16.tex + docs/arxiv-trinity-gf16-draft.md): - 'verified on silicon, with 35/35 RTL tests' -> 'verified in FPGA synthesis and RTL simulation, with 35/35 tests'. - shuttle 'TTSKY26a (May 2026)' -> 'TTSKY26b TT4913 Gamma (submission closed May 2026)' по SSOT frozen_silicon_anchor (conformance/FORMAT-SPEC-001.json) + добавлено 'silicon not yet returned (expected late 2026), so no on-chip measurement is claimed'. - 'actual hardware runs' -> 'actual FPGA hardware runs (Artix-7 XC7A100T), not ASIC silicon'. - заголовок и \label §5 ASIC Path: TTSKY26a -> TTSKY26b TT4913 Gamma. Не тронуто: цифры 323 MHz / 40350 LUT / 64 DSP48E1 / 35/35 / 12.8-41.2 GOPS, спека 1/6/9 bias=31. Closes #1341
gHashTag
force-pushed
the
fix/gf16-paper-honesty-silicon-shuttle
branch
from
July 4, 2026 14:24
b806ec0 to
ed0cd1c
Compare
Contributor
|
📓 NotebookLM Notebook linked to this PR
This notebook contains session context, decisions, and artifacts for this work. |
Contributor
PR DashboardGenerated at: 2026-07-04 14:25:04 UTC
Summary
Seal Status
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Суть
Honesty-правка формулировок статьи GF16 (arXiv:2606.05017): текст заявлял «4x4 matmul verified on silicon» при доказательстве через RTL/FPGA-synth (Artix-7 XC7A100T). Это противоречит правилу разграничения encoding ≠ compute ≠ FPGA и sim/synth ≠ HW-кремний: «отправлен на изготовление» ≠ «измерено на кристалле».
Что изменено (оба файла согласованно)
docs/arxiv-submission/trinity-gf16.tex+docs/arxiv-trinity-gf16-draft.md:conformance/FORMAT-SPEC-001.json(frozen_silicon_anchor.tapeout = "TTSKY26b TT4913 Gamma", RTLtt-trinity-gamma/src/gf16_v2_mul.v). Добавлено: «silicon has not yet been returned (expected late 2026), so no on-chip measurement is claimed» — по данным TinyTapeout чипы TTSKY26a/b ожидаются ноя–дек 2026.\label§5 ASIC Path:TTSKY26a→TTSKY26b TT4913 Gamma.Что НЕ тронуто
Цифры производительности (323 MHz, 40 350 LUT, 64 DSP48E1, 35/35 tests, 12.8–41.2 GOPS — все с FPGA-прогонов), спека 1/6/9 bias=31, φ-anchor. Diff: 12 insertions / 10 deletions в 2 файлах.
Верификация фактов
frozen_silicon_anchor(FORMAT-SPEC-001.json): tapeout = TTSKY26b TT4913 Gamma.Связано с erratum-треком каталожной статьи (arXiv:2606.09686, 84→83).
Closes #1341