Phi-4-mini-instruct deployment hangs indefinitely with 0 token generation in Azure AI Foundry

Faiz Delvi 20 Reputation points
2026-05-26T19:38:07.0666667+00:00

We are experiencing an issue with Phi-4-mini-instruct deployments in Azure AI Foundry.

Observed behavior:

  • Deployment succeeds successfully

Requests reach the endpoint

Playground stays on "Thinking..." indefinitely

No completion is ever returned

Metrics show:

Requests increasing

  Total token count = 0
  
     Completion token count = 0

Regions tested:

East US 2

Sweden Central

Additional findings:

Phi-4-mini-reasoning works correctly in the same subscription/resource

GPT models work correctly

Multiple redeployments tested

API integration is working for other models

This appears to be specific to Phi-4-mini-instruct preview deployments.

Has anyone else experienced this issue, or is there a known backend/runtime problem with Phi-4-mini-instruct currently?

Thank you.We are experiencing an issue with Phi-4-mini-instruct deployments in Azure AI Foundry.

Observed behavior:

Deployment succeeds successfully

Requests reach the endpoint

Playground stays on "Thinking..." indefinitely

No completion is ever returned

Metrics show:

Requests increasing

  Total token count = 0
  
     Completion token count = 0

Regions tested:

East US 2

Sweden Central

Additional findings:

Phi-4-mini-reasoning works correctly in the same subscription/resource

GPT models work correctly

Multiple redeployments tested

API integration is working for other models

This appears to be specific to Phi-4-mini-instruct preview deployments.

Has anyone else experienced this issue, or is there a known backend/runtime problem with Phi-4-mini-instruct currently?

Thank you.

Microsoft Foundry
Microsoft Foundry

A unified Azure platform for creating and managing AI models, agents, and applications with built‑in enterprise security, monitoring, and governance

Answer accepted by question author
Karnam Venkata Rajeswari 5,340 Reputation points Microsoft External Staff Moderator
2026-05-26T20:23:06.0066667+00:00

Hello @Faiz Delvi ,

Welcome to Microsoft Q&A .Thank you for reaching out to us.

The observed pattern is consistent with a potential model-specific inference and runtime condition affecting the Phi-4-mini-instruct deployment path, where the request is accepted but does not proceed to token generation.

Based on the consistent cross-region reproduction and the fact that other models operate correctly within the same subscription, the behavior is unlikely to be related to configuration, authentication, networking or quota limitations.

Quota or throttling scenarios typically result in explicit error responses (such as 429 or 5xx codes), rather than silent execution with zero token generation.

To ensure service continuity, the following alternatives can be used temporarily:

  • Phi-4-mini-reasoning for similar workloads
  • GPT-based deployments as fallback options
  • Optional routing logic to switch models when no completion tokens are generated

The following references might be helpful , please check them out

Azure OpenAI in Microsoft Foundry Models Quotas and Limits - Microsoft Foundry | Microsoft Learn

Please let us know if the response was helpful

 

Thank you

Was this answer helpful?

1 person found this answer helpful.
0 comments

3 additional answers

Sort by: Most helpful
Most helpful Newest Oldest
  1. Andrew Taylor - COREZENN 1,390 Reputation points Volunteer Moderator
    2026-06-02T19:34:18.68+00:00

    Hi @Faiz Delvi

    Thank you for bringing this to our attention and for providing such a detailed breakdown of the issue. Testing across different regions and isolating the behavior by verifying that Phi-4-mini-reasoning and GPT models work correctly are excellent troubleshooting steps.

    Currently, there are no official service advisories or documented known issues specifically regarding the Phi-4-mini-instruct preview model deployment hanging indefinitely with 0 token generation in Azure AI Foundry. Because this is a preview feature, it is highly likely you are encountering an undocumented backend anomaly or a container initialization bug specific to this model variant.

    Given that you have already exhausted standard troubleshooting steps (including multiple redeployments across East US 2 and Sweden Central), this issue requires deeper investigation by the product engineering team.

    Recommended Next Steps: Please open a technical support ticket so the Azure Support team can investigate the backend logs for your specific endpoint. You can create a support request directly from the Azure Portal Help + support or via the Azure Support Options page.

    For guidance on creating a support request, please refer to the official documentation: Create an Azure support request.

    Once you receive a response or resolution from the support engineers, we would greatly appreciate it if you could share those findings back here to help other community members who might be testing the same preview model!

    Please 'Upvote'(Thumbs-up) and 'Accept' as answer if the response was helpful. This will be benefitting other community members who face the same issue.

    Best regards,

    Andrew S Taylor

    Was this answer helpful?

    0 comments
  2. Oscar Bonilla Calderon 0 Reputation points
    2026-06-02T17:36:16.58+00:00

    Today is June 2nd, 8 days after the first report, and I'm still experiencing the exact same error. When will it be fixed?

    Was this answer helpful?

    0 comments
  3. kagiyama yutaka 5,570 Reputation points
    2026-05-27T00:12:45.4966667+00:00

    I think that Azure does not list any client‑side fix for Phi‑4‑mini‑instruct returning 0 tokens, and you can send a repro with the request id and time to Azure support.

    Was this answer helpful?

    0 comments

Your answer

Answers can be marked as 'Accepted' by the question author and 'Recommended' by moderators, which helps users know the answer solved the author's problem.