use ollama with intel gpu.

[ShafinSiKDER802]

i am using intel arcA770. with normal ollama download from website it useage cpu. i tried using ollama-ipex-llm and it worked on intel gpu. But when i tried to use it on Vscode it giving me this error :

Error parsing Ollama response: Error: POST predict: Post "http://127.0.0.1:58023/completion": read tcp 127.0.0.1:58025->127.0.0.1:58023: wsarecv: An existing connection was forcibly closed by the remote host. {"error":"POST predict: Post "http://127.0.0.1:58023/completion\": read tcp 127.0.0.1:58025-\u003e127.0.0.1:58023: wsarecv: An existing connection was forcibly closed by the remote host."}

[bdougie]

This feels like a specific issue to triage via Ollama. The best I can do is provide some triaging ideas. Based on my research (in claude):

Summary of the Problem

This is a connection error when trying to use ollama-ipex-llm (Intel's GPU-accelerated Ollama) with the Continue VS Code extension.

The Core Issue Port 58023

The error shows port 58023 instead of Ollama's standard port 11434. This is highly unusual and suggests:

1. Wrong port configuration - The Continue extension might be trying to connect to the wrong port
2. Port conflict - Something else might be occupying port 11434
3. Incorrect IPEX-LLM setup - The ollama-ipex-llm service may not be configured to listen on the correct port

Fix: Ollama-IPEX-LLM + Continue Connection Error

Problem

Your error shows port 58023, but Ollama uses port 11434. This is a configuration mismatch.

Quick Fix

1. Start Ollama Correctly

Windows:

set OLLAMA_HOST=0.0.0.0:11434
set OLLAMA_NUM_GPU=999
ollama serve

Linux:

export OLLAMA_HOST=0.0.0.0:11434
export OLLAMA_NUM_GPU=999
./ollama serve

Leave this terminal open.

2. Fix Continue Config

Open VS Code → Ctrl+Shift+P → "Continue: Open Config"

Set this in config.json:

{
  "models": [{
    "title": "IPEX-LLM",
    "provider": "ollama",
    "model": "AUTODETECT",
    "apiBase": "http://localhost:11434"
  }]
}

Critical: Make sure apiBase is http://localhost:11434 (NOT 58023)

3. Reload VS Code

Ctrl+Shift+P → "Developer: Reload Window"

Verify It Works

New terminal:

curl http://localhost:11434/api/tags

Should return JSON with your models.

Still Not Working?

1. Kill all Ollama processes first:

   • Windows: taskkill /F /IM ollama.exe
   • Linux: pkill ollama

2. Check what's on port 11434:

   • Windows: netstat -ano | findstr :11434
   • Linux: netstat -tuln | grep 11434

3. Pull a model if needed:

   ollama pull llama3.2

---

Links:

• IPEX-LLM Docs
• Continue IPEX-LLM Setup