从陪读蛙回到沉浸式翻译后，发现沉浸式翻译漏翻厉害
- 内容介绍
- 文章标签
- 相关推荐
沉浸式翻译版本：v1.26.6 哈雷酱
陪读蛙版本：v1.32.1
测试页面：Documentation | NestJS - A progressive Node.js framework
模型：LongCat-Flash-Chat-2602-Exp
从最顶部一路读下来
沉浸式翻译（默认提示词）
image1804×1913 320 KB
陪读蛙（默认提示词）
image1753×1906 317 KB
网友解答：--【壹】--：
尝试了 Aloxaf 酱 的自定义专家，似乎解决了。
因为今天遇到陪读蛙会无限打满接口的 BUG，
fix(translate): honor 429 rate limit and add queue cooldown (fixes #1289) (#1290)
main ← fix/1289-rate-limit-cooldown
已打开 06:11AM - 07 Apr 26 UTC
frogGuaGuaGuaGua
+202
-11
This PR fixes the issue where the automatic retry mechanism bypasses the rate li…mit by indiscriminately retrying 429 Too Many Requests errors. Changes: - Removed re-throwing of `Error` strings in `api/ai.ts` to preserve the original error objects, keeping the `statusCode` and `responseHeaders` accessible. - Added 429 and rate-limit detection in the `RequestQueue` using `isRateLimitError`. - Implemented a queue-wide `cooldownUntil` that pauses the entire queue (not just the individual task) upon encountering a 429 error. - Updated backoff calculation to respect upstream `Retry-After` and `Retry-After-Ms` headers when present, rather than blindly applying the default exponential backoff. - Added relevant unit tests to cover 429 cooldown and non-retryable API errors. Fixes #1289. --- ## Summary by cubic Respect upstream rate limits by pausing the request queue on 429 and honoring Retry-After headers, preventing aggressive retries and API throttling. Fixes #1289. - **Bug Fixes** - Preserve original error objects in `api/ai.ts` so `statusCode` and `responseHeaders` are retained. - Detect 429 and rate-limit messages; apply a queue-wide cooldown via `cooldownUntil`. - Use `Retry-After`/`Retry-After-Ms` for backoff; disable jitter on rate-limited retries. - Retry only retryable errors (e.g., 408/409/429/5xx); added tests for 429 cooldown and non-retryable cases. <sup>Written for commit f6e6217af112ba1ff0bcf9646133aafe7be18717. Summary will update on new commits.</sup>
先用沉浸式翻译吧
--【贰】--：
这个模型好用吗，没有找到合适的高 rpm 模型
--【叁】--：
不会啊
image1680×1025 113 KB
--【肆】--：
确实，沉浸式翻译在有些网页上是会漏掉，不知道为啥
--【伍】--：
沉浸式翻译没问题啊，我看你的陪读蛙反而把格式弄坏了
image1312×1301 98.9 KB
--【陆】--：
为啥都喜欢双语对照，既然要看中翻，我觉得仅译文模式才舒服。双语对照不是没翻译，是没地方显示翻译了吧
image1920×869 154 KB
沉浸式翻译版本：v1.26.6 哈雷酱
陪读蛙版本：v1.32.1
测试页面：Documentation | NestJS - A progressive Node.js framework
模型：LongCat-Flash-Chat-2602-Exp
从最顶部一路读下来
沉浸式翻译（默认提示词）
image1804×1913 320 KB
陪读蛙（默认提示词）
image1753×1906 317 KB
网友解答：--【壹】--：
尝试了 Aloxaf 酱 的自定义专家，似乎解决了。
因为今天遇到陪读蛙会无限打满接口的 BUG，
fix(translate): honor 429 rate limit and add queue cooldown (fixes #1289) (#1290)
main ← fix/1289-rate-limit-cooldown
已打开 06:11AM - 07 Apr 26 UTC
frogGuaGuaGuaGua
+202
-11
This PR fixes the issue where the automatic retry mechanism bypasses the rate li…mit by indiscriminately retrying 429 Too Many Requests errors. Changes: - Removed re-throwing of `Error` strings in `api/ai.ts` to preserve the original error objects, keeping the `statusCode` and `responseHeaders` accessible. - Added 429 and rate-limit detection in the `RequestQueue` using `isRateLimitError`. - Implemented a queue-wide `cooldownUntil` that pauses the entire queue (not just the individual task) upon encountering a 429 error. - Updated backoff calculation to respect upstream `Retry-After` and `Retry-After-Ms` headers when present, rather than blindly applying the default exponential backoff. - Added relevant unit tests to cover 429 cooldown and non-retryable API errors. Fixes #1289. --- ## Summary by cubic Respect upstream rate limits by pausing the request queue on 429 and honoring Retry-After headers, preventing aggressive retries and API throttling. Fixes #1289. - **Bug Fixes** - Preserve original error objects in `api/ai.ts` so `statusCode` and `responseHeaders` are retained. - Detect 429 and rate-limit messages; apply a queue-wide cooldown via `cooldownUntil`. - Use `Retry-After`/`Retry-After-Ms` for backoff; disable jitter on rate-limited retries. - Retry only retryable errors (e.g., 408/409/429/5xx); added tests for 429 cooldown and non-retryable cases. <sup>Written for commit f6e6217af112ba1ff0bcf9646133aafe7be18717. Summary will update on new commits.</sup>
先用沉浸式翻译吧
--【贰】--：
这个模型好用吗，没有找到合适的高 rpm 模型
--【叁】--：
不会啊
image1680×1025 113 KB
--【肆】--：
确实，沉浸式翻译在有些网页上是会漏掉，不知道为啥
--【伍】--：
沉浸式翻译没问题啊，我看你的陪读蛙反而把格式弄坏了
image1312×1301 98.9 KB
--【陆】--：
为啥都喜欢双语对照，既然要看中翻，我觉得仅译文模式才舒服。双语对照不是没翻译，是没地方显示翻译了吧
image1920×869 154 KB