# Bad request error when using inference endpoints: Cannot find backend for CPU

**URL:** <https://discuss.huggingface.co/t/bad-request-error-when-using-inference-endpoints-cannot-find-backend-for-cpu/92451>\
**Category:** Inference Endpoints on the Hub\
**Created:** [June 16, 2024, 5:55pm UTC](https://discuss.huggingface.co/t/bad-request-error-when-using-inference-endpoints-cannot-find-backend-for-cpu/92451 "2024-06-16T17:55:22Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![dordonezc](https://avatars.discourse-cdn.com/v4/letter/d/a698b9/32.png) [@dordonezc](https://discuss.huggingface.co/u/dordonezc)\
**Post date:** [June 16, 2024, 5:55pm UTC](https://discuss.huggingface.co/t/bad-request-error-when-using-inference-endpoints-cannot-find-backend-for-cpu/92451/1 "2024-06-16T17:55:22Z")

</div>

Hey,

I wrote a custom inference handler to deploy the phi-mini-128k model using langchain here [dordonezc/Phi-3-small-128k-instruct-4-endpoints · Hugging Face](https://huggingface.co/dordonezc/Phi-3-small-128k-instruct-4-endpoints)

However, when I try to send a request to the endpoint I get the following error

BadRequestError: (Request ID: 7I0Wm6)  
Bad request:  
Cannot find backend for CPU

I am using an Nvidia T4 · 4x GPU · 64 GB image

I think something related to the flash attention is not working up correctly, does anyone know what might be happening?