Building a Lightweight Streamlit Client for Local Ollama LLM Interaction

Ollama 提供端點串接服務，可由程式管理及使用本地大語言模型(LLM, Large Language Model)。 以下程式碼展示如何以 Streamlit 套件，建立一個輕量級的網頁介面，供呼叫本地端安裝的 Ollama 大語言模型。 Ollama 預設的服務端點為 http://localhost:11434。 使用者透過簡單的網頁操作，即可輸入提示詞（prompt），即時獲得模型回應。 此應用支援模型選擇、選字機率均化溫度（temperature）、選字累加機率門檻值(top_p)、問答最大符號數（max_tokens）等參數調整，並允許限定特定使用者（如 "john"）。 Ollama Streamlit為一個實用的範例，適合想要快速整合本地 LLM ，建立互動式介面的開發者。

######################################
#  ollama_client_streamlit.py
#   an streamlit client to call ollama LLMs.
#
#   # python 3.7.7 above recommended
#   > conda activate your_streamlit_environment
#   > streamlit run ollama_client_streamlit.py --server.port 8080
#
#   http://localhost:8080?name=john
#
# Reference:
#   GitHub: Ollama API Documentation
#
import streamlit as st
import requests
import time
import json

###################################
# Parameters for the Streamlit App
baseurl = 'http://localhost:11434'
generate_url = baseurl + '/api/generate'
tags_url = baseurl + '/api/tags'
# only names in pass_list are allowed to use LLM
pass_name_list = ["john"]   
default_prompt = 'Where is the capital of Japan?'
# api_key is optional for endpoints with bearer authentication
api_key = ''

# returns a list of llm names installed in ollama
def get_ollama_models():
    """Fetches available models from the Ollama API and returns a list."""
    ##url = "http://localhost:11434/api/tags"  # Ollama API endpoint
    try:
        headers={'Authorization': f'Bearer {api_key}'}
        response = requests.get(tags_url, headers=headers)
        
        # Raises an error for HTTP failures
        response.raise_for_status()
        data = response.json()

        # Extract model names
        return [model["name"] for model in data.get("models", [])]  
    except requests.exceptions.RequestException as e:
        print(f"Error fetching models: {e}")
        return []


# Calls /api/generate to request LLM in stream=true mode
#   model: ollama installed llm
#   temperature: 0 ~ 1 (decide how creative the answer is generated)
#   top_p: 0 ~ 1 (decide how many most probable words to select)
#   max_tokens: 0 ~ 2048 tokens
def generate_stream(prompt, model, temperature=0.7, top_p=0.95, max_tokens=1024):
    args = {
        'prompt': prompt,
        'model': model,
        'options': {
            'temperature': temperature,
            'top_p': top_p,
            'max_tokens': max_tokens},
        'stream': True
    }

    headers = {'Authorization': f'Bearer {api_key}'}
    r = requests.post(generate_url, headers=headers, json=args, stream=True)
    if r.status_code == 200:
    # {"model":"gemma3:12b-it-qat",
    #  "created_at":"2025-05-24T09:18:43.944815912Z",
    #  "response":"The","done":false}
        # print("Request successful!")
        # with requests.post(url, headers=headers, 
        #    data=json.dumps(data), stream=True) as response:
        for line in r.iter_lines():
            if line:
                try:
                    chunk = json.loads(line.decode("utf-8"))
                    message = chunk.get("response", {})
                    # print(message)
                    yield message
                except json.JSONDecodeError:
                    continue

        # message = r.json()['response']
    else:
        message = f"Error: {r.status_code}, Message: {r.text}"

    # response_json = response.json()
    # message = response_json['response']

    yield message


#  no cache resource is needed with ollama endpoints
@st.cache_resource
def _load():
    return

#  calls generate_stream() to collect stream chunk response 
#    in _result.markdown(full_response)
@st.cache_data(ttl='1d')  # cache _result for one day
def _generate(prompt, model, temperature=0.7, top_p=0.95, max_tokens=1024):
    _result = st.empty()
    full_response = ''

    start = time.time()
    for chunk in generate_stream(prompt, model, temperature, top_p, max_tokens):
        full_response += chunk
        _result.markdown(full_response)
        # st.write_stream(response)
    end = time.time()

    st.write(f"generate text in {end - start:.3f} seconds\n\n")
    # return response_placeholder


start = time.time()
_load()
end = time.time()

#############
# filter user
# only names in pass_name_list are allowed to use the app
params = st.query_params
name = params.get("name", "")
if name in pass_name_list:
    result = f"Hello, {name}!"
else:
    #result = "Unsupported user"
    st.write(f"Sorry, you are not supported user.")
    exit(0)

###############
# Streamlit UI
st.sidebar.write(f"Hello, {name}!")

# title for the web page
st.title("Ollama LLM Demo ")

# let sliders be located in left hand side of ui
# slider for temperature
temperature = st.sidebar.slider("temperature", 0.0, 1.0, 0.7, 0.01)

# slider for top_p
top_p = st.sidebar.slider("top_p", 0.0, 1.0, 0.95, 0.01)

# slider for max_tokens
max_tokens = st.sidebar.slider("max_tokens", 0, 2048, 1024, 1)

# selectbox for model_engine
models = get_ollama_models()
#print(models)  # ["llama3.1", "gemma3:12b-it-qat"]
model_engine = st.sidebar.selectbox("model_engine", models)

# input text
prompt = st.text_area("Your Query for LLM", default_prompt)

# output area
_generate(prompt, model_engine, temperature, top_p, max_tokens)