gpt.ntnu.no – High Performance Computing

Table of Contents

Why it exists?

gpt.ntnu.no and gpt.uio.no is a collaboration between NTNU and UiO to provide access to LLMs (Large Language Model).

Access to LLMs is provided via Web Interface and API.

With local LLMs - data is transferred only inside university networks.

What models are available?

Models available (updated 2025-12-02)

Model	Purpose	Comment	Hosted
GPT-OSS 120B	general purpose (recommended)	by OpenAI, for tasks which requires advanced analytics.	IDUN HPC cluster (local)
GLM 4.5 Air	AI coding assistant	by Chinese A.Ai, Sutable for task that requires deeper context and reasoning. IDUN HPC cluster	IDUN HPC cluster (local)
GPT-4o mini	general purpose	By OpenAI, the fastest of the smaller models – when you need with notes, brainstorming, drafting emails.	Microsoft Azur Cloud
Multilingual-E5-large-instruct	embedding	by Infloat, Microsoft Research Asia. The Multilingual-E5-Large-Instruct is an embedding model designed to handle multilingual text data (supports 100 languages).	IDUN HPC cluster (local)

Web interface

Create AI assistant:

API

Do you need API token then send support email to: help@hpc.ntnu.no

Test you API token

with curl command:

$ curl \
-H "Content-Type: application/json" \
-H "model: gpt-oss-120b" \
-H "Authorization: Bearer ...API...TOKEN......." \
-d '{"model": "gpt-oss-120b", "messages":[{"content":"Test, please respond ok", "role":"user"}]}' \
https://gpt.ntnu.no/api/models/chat-completion

Output example:

{"id":"8a5f1bd56d934fb69605bb0bf37c2c1f","object":"chat.completion","created":1762253304,"model":"gpt-oss-120b","choices":[{"index":0,"message":{"role":"assistant","content":"ok","reasoning_content":"User says \"Test, please respond ok\". Simple. Respond \"ok\".","tool_calls":null},"logprobs":null,"finish_reason":"stop","matched_stop":200002}],"usage":{"prompt_tokens":72,"total_tokens":98,"completion_tokens":26,"prompt_tokens_details":null,"reasoning_tokens":0},"metadata":{"weight_version":"default"}}

Example with Python

Create Python script chat.py to use API:

#!/usr/bin/python3
import sys
import requests
 
def chat_with_model(token,model,question):
    url = 'https://gpt.ntnu.no/api/models/chat-completion'
    headers = {
        'Model': model,
        'Authorization': f'Bearer {token}',
        'Content-Type': 'application/json'
    }
    data = {
      "model": model,
      "messages": [
        {
          "role": "user",
          "content": question
        }
      ]
    }
    response = requests.post(url, headers=headers, json=data)
    return response.json()
 
my_api_key = sys.argv[1]
my_model = sys.argv[2]
my_question = sys.argv[3]
answer = chat_with_model(my_api_key, my_model, my_question)
print(answer)

Output example:

$ python3 chat.py '…..API…TOKEN………' 'gpt-oss-120b' 'What is the tallest mountain in Norway? Answer with name only.'
. . .
. . .
{'id': '1cb7f9dd6c63422e9c183a5659926a46', 'object': 'chat.completion', 'created': 1762253012, 'model': 'gpt-oss-120b', 'choices': [{'index': 0, 'message': {'role': 'assistant', 'content': 'Galdhøpiggen', 'reasoning_content': 'User asks: "What is the tallest mountain in Norway? Answer with name only." So just the name: Galdhøpiggen. Provide only name.', 'tool_calls': None}, 'logprobs': None, 'finish_reason': 'stop', 'matched_stop': 200002}], 'usage': {'prompt_tokens': 80, 'total_tokens': 129, 'completion_tokens': 49, 'prompt_tokens_details': None, 'reasoning_tokens': 0}, 'metadata': {'weight_version': 'default'}}

Text embedding model

Example:

curl -X POST "https://gpt.ntnu.no/api/models/chat/completions" \
-H "Authorization: Bearer ...API...TOKEN......." \
-H "model: multilingual-e5-large-instruct" \
-H "Content-Type: application/json" \
-d '{ "inputs": "testing", "normalize": true, "truncate": true, "truncation_direction": "Right" }'

Visual Studio Code extensions

We tested extensions Roo Code, Cline other should work as well.

Roo Code configuration:

Cline configuration