Azure Custom Translator训练会话无法完成，请求排查问题

Hey there, sorry to hear you're stuck getting your Custom Translator training session to complete. Let's walk through some practical checks and fixes that might resolve this issue:

• Validate Your Training Data Integrity
  Double-check all uploaded datasets (training, tuning, test) for common issues. Even small problems like mismatched sentence pairs, blank lines, corrupted entries, or unsupported file formats (stick to .txt, .tmx, .xliff as recommended) can halt training entirely. A single faulty entry in a TMX file, for example, can cause the whole job to fail silently.

• Confirm Subscription Health & Quotas
  Even though you've added a subscription, it's worth verifying a few details:

  • Head to the Azure Portal, locate your Translator resource, and check the Usage + Quotas section to ensure you haven't hit limits on training hours or concurrent jobs.
  • Make sure your subscription is active—expired trials or billing hold-ups can block training without obvious alerts.

• Review Project & Model Settings
  Ensure your project's source and target languages exactly match the languages in your datasets. A language mismatch is one of the most common causes of failed training. Also, verify you've selected a base model that supports your specific language pair—some models have limited language support.

• Check Detailed Training Logs
  In the Custom Translator portal, navigate to your project's Training Sessions tab. Click into the failed session to access detailed logs. These logs typically spell out the exact failure cause (e.g., "invalid entry in line 42 of training file" or "quota exceeded")—this is the fastest way to pinpoint the issue.

• Test with a Minimal Project
  If you've made multiple edits to datasets or settings, try creating a small test project with a tiny, verified set of sentence pairs (10-20 pairs should work). If this test training completes successfully, the problem is likely with your original project's data or configuration rather than the platform itself.

If none of these steps resolve the issue, reaching out to Azure Support with your project ID (7f1b6f9a-d4f4-4ea1-8492-69075339422f) is your best next step—they can access backend logs that aren't visible to users to diagnose the failure.

内容的提问来源于stack exchange，提问作者Eric Baudouin