[gt50]

Title

Issue: relates to #606 and #728

Summary

Two things about replies Strata ends because they repeat:

• Opt-in: a guard for a short phrase repeated back to back.
• Always: the stop is named.

Split out of #1789 (which now only carries the Windows AMD images), as AGENTS.md asks: one change per PR, a new feature opt-in.

What changed

• serve/server.py, "repeat_stop_period": N (opt-in; default 1 = only After a 36,689-token request at 155k, every later request answers one repeated token until the engine reloads #606's one-token rule):
  • a run of repeat_stop_tokens tokens that repeats with a period of 2..N tokens ends the reply like After a 36,689-token request at 155k, every later request answers one repeated token until the engine reloads #606;
  • O(N) per token;
  • seen on a scanned page: [unclear] written until max_tokens ran out (12,000 tokens, 229 s), which the one-token rule never sees;
  • 16 is a good value.
• choices[0].stop_cause: a reply ended by either guard (or by Infinite loop issue #728's reasoning_loop_recovery: "stop") keeps finish_reason: "length" and adds Strata's own "stop_cause": "repetition" / "reasoning_loop". On the OpenAI chat API, streamed and not.
• serve/structured.py: a structured answer ended that way fails with a message that says the reply repeated itself, instead of "increase the output budget", which a larger max_tokens cannot fix. Responses API too.
  • Seen with a one-field json_schema on a scanned rent roll: the model wrote the escaped newline 256 times, After a 36,689-token request at 155k, every later request answers one repeated token until the engine reloads #606 stopped it, and the client was told to raise the budget.
• Docs: docs/DETAILS.md.

Extra Notes

• Default path: with the default repeat_stop_period 1, the guard code is skipped. Replies that end normally, or at max_tokens, are unchanged. stop_cause and the new message appear only when After a 36,689-token request at 155k, every later request answers one repeated token until the engine reloads #606 or Infinite loop issue #728 already end a reply, which are replies that were already cut short.
• Tests:
  • serve/test_server.py RepeatStop: off by default, a phrase ended with period 16, period 1, a long varied reply runs to its end;
  • serve/test_structured.py: the message, stop_cause streamed and not, and no cause for a real budget stop.
  • All serve and setup test files pass on main 61b3fb5 (1158 tests; rebased).

🤖 Generated with Claude Code

[qiuzan2001]

Two points (serve.test_server and serve.test_structured pass here, 329 tests): (1) the new repetition message in serve/structured.py:197 tells users to set structured_plain_string, which exists only in the unmerged #1804; git grep structured_plain_string upstream/main finds nothing, so on its own this PR points users at an option that does not exist. (2) stop_cause is also set when the existing #606 one-token guard or the reasoning-loop stop fires, so default responses (and the default structured error text) change even with repeat_stop_period unset; AGENTS.md asks for the default to stay byte-identical, so gating it with the opt-in or splitting it out would match.

[gt50]

@qiuzan2001 thanks, both points were right. Fixed in c720feb:

1. The repetition error no longer mentions structured_plain_string. If structured: one-string-field json_schema answered as plain text (opt-in), progress in structured keep-alives #1804 lands, it can add that pointer itself.
2. stop_cause is now added only when repeat_stop_period is set (2 or more). With the default config, a reply ended by the After a 36,689-token request at 155k, every later request answers one repeated token until the engine reloads #606 one-token guard or by reasoning_loop_recovery is the same as on main, and so is the structured error text. New tests check both. They fail on the previous commit.

serve.test_server and serve.test_structured pass (330 tests). docs/DETAILS.md now says the field comes only with the opt-in.

[Niko1221]

Thanks for the pull request.

We're resetting the tracker for 0.2.0-beta. There are more than 1,300 open issues and pull requests, too many for us to review properly, so we're closing everything that isn't being worked on right now. Closed here does not mean rejected.

To reopen it takes about two minutes:

1. Rebase on main if it no longer applies, and re-run your checks on 0.2.0-beta.
2. Start the title with its area: [Engine], [GUI], [TUI], [Benchmark], [Docs] or [Other].
3. Fill in the pull request template: the problem and result, and the validation (what you ran, on which revision, what you saw).
4. Comment /reopen. A check runs and either reopens the pull request or lists what is still missing.

The rules are in CONTRIBUTING.md.