Repository navigation
Merged
Conversation
…loses #10789) specs/t27t/t27t.t27 is t27t's law: the RTX 4090 bench of #10360 with every ratio recomputed from its times, the saving f x (1 - 1/s) recomputed cell by cell (16% to 24%, f still an estimate), and pick_path, which sends forward GEMMs to int8 tensor cores via torch._int_mm, keeps gradients in bf16 and never picks dp4a. Runner: gHashTag/igla-coder-gpu. `tri t27t` runs it. AGENTS.md section 8 points at it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This was referenced Oct 19, 2026
Closed
Contributor
gHashTag
enabled auto-merge (squash)
October 19, 2026 03:02
This was referenced Oct 19, 2026
scripts/tri: give the t27t arm its own line, so the docs check finds tri t27t (Closes #10807)
#10809
Merged
gHashTag
added a commit
that referenced
this pull request
Oct 19, 2026
…tri t27t (Closes #10807) (#10809) PR #10790 put t27t) after factory) on one line. The documented-commands check reads an arm only when it starts a line, so tri t27t counted as a dead mention (138 -> 139). The command is unchanged. The blank line above the case goes, so the diff adds no foreign line. Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Closes #10789
t27t is the t27 trainer. This PR gives it a law in t27,
specs/t27t/t27t.t27, and a front door,tri t27t. Its runner is the PyTorch trainer gHashTag/igla-coder-gpu, which keeps its language; t27t says what it must do.What the spec holds
pick_path.torch._int_mm.F_MEASURED = falseandRUNNER_USES_INT_MM = false, plus aNOT_MEASUREDlist.Other files
AGENTS.mdsection 8 has a new "t27t: the t27 trainer" entry that points at the spec.scripts/tri:tri t27twas added on existing lines, so the foreign line count stays the same.docs/reports/t27b_expectations.jsonhas a row for the new spec.Checked on the Railway lab (t27c 0.5.2, zig 0.16.0)
t27c test-report specs/t27t/t27t.t27: 8 of 8 tests pass, 3 invariants proved, 0 vacuous.pass, so the t27b-native-ratchet in CI is the check.🤖 Generated with Claude Code