""" Simple Chatbot @author: Nigel Gebodh @email: nigel.gebodh@gmail.com @website: https://ngebodh.github.io/ """ import numpy as np import streamlit as st from openai import OpenAI import os import sys from dotenv import load_dotenv, dotenv_values load_dotenv() # #=========================================== # updates = ''' # Updates # + 02/06/2026 # - Updated inference endpoints for HF models # - Added Kimi model # + 01/10/2026 # - Updated cooldown # + 01/08/2026 # - Updated logging info # + 10/10/2025 # - Update the model options since Gemma-2-9B-it # is no longer supported. Replaced with GPT-OSS-120B # + 04/20/2025 # - Changed the inference from HF b/c # API calls are not very limted. # - Added API call limiting to allow for demoing # - Added support for adding your own API token. # + 04/16/2025 # - Changed the inference points on HF b/c # older points no longer supported. # ''' # #------------------------------------------- #========================================================== # Logging # -------------------------------------------- import requests from datetime import datetime try: LOGGER_TOOL_WEBHOOK = os.environ.get("LOGGER_TOOL_URL") except Exception as e: print(f"❌ Error in loading LOGGER_TOOL_WEBHOOK") def log_to_webhook( *, session_info: dict, model: str, prompt: str, response: str, temperature: float, ): if not LOGGER_TOOL_WEBHOOK: return payload = { #Session info **session_info, #Model info "model": model, "temperature": temperature, #Content "user_prompt": prompt, "assistant_response": response, #Usage "api_call_count": st.session_state.api_call_count, "api_call_limit": API_CALL_LIMIT, "remaining_calls": API_CALL_LIMIT - st.session_state.api_call_count, #Timestamp "timestamp": datetime.utcnow().isoformat(), } try: requests.post(LOGGER_TOOL_WEBHOOK, json=payload, timeout=3) except Exception as e: print("Logging failed") # -------------------------------------------- #========================================================== # Unique Users / Session Info # -------------------------------------------- import uuid import time import hashlib import json import sys from datetime import datetime def get_session_info(): data = { "timezone": time.tzname, "platform": sys.platform, "rand": uuid.uuid4().hex, } raw = json.dumps(data, sort_keys=True) return hashlib.sha256(raw.encode()).hexdigest()[:12] if "session_info" not in st.session_state: st.session_state.session_info = { "session_id": str(uuid.uuid4()), "session_start": datetime.utcnow().isoformat(), "conversation_id": str(uuid.uuid4()), "run_count": 0, "fingerprint": get_session_info(), "platform": sys.platform, "timezone": time.tzname, } st.session_state.session_info["run_count"] += 1 def reset_conversation(): st.session_state.conversation = [] st.session_state.messages = [] st.session_state.session_info["conversation_id"] = str(uuid.uuid4()) # -------------------------------------------- #========================================================== # Limits # -------------------------------------------- API_CALL_LIMIT = 20 # Define the limit if 'api_call_count' not in st.session_state: st.session_state.api_call_count = 0 st.session_state.remaining_calls = API_CALL_LIMIT REQUEST_COOLDOWN = 3 # seconds between requests if "last_request_time" not in st.session_state: st.session_state.last_request_time = 0 # -------------------------------------------- model_links_hf ={ "Gemma-3-27B-it":{ "inf_point":"/static-proxy?url=https%3A%2F%2Frouter.huggingface.co%2Fv1", "link":"google/gemma-3-27b-it:scaleway", }, "Meta-Llama-3.1-8B":{ "inf_point":"/static-proxy?url=https%3A%2F%2Frouter.huggingface.co%2Fv1", "link":"meta-llama/Meta-Llama-3.1-8B-Instruct:scaleway", }, "DeepSeek-R1-Distill-Llama-70B":{ "inf_point":"/static-proxy?url=https%3A%2F%2Frouter.huggingface.co%2Fv1", "link":"deepseek-ai/DeepSeek-R1-Distill-Llama-70B:scaleway", }, "Qwen2.5-Coder-32B-Instruct":{ "inf_point":"/static-proxy?url=https%3A%2F%2Frouter.huggingface.co%2Fv1", "link":"Qwen/Qwen3-235B-A22B-Instruct-2507:scaleway", }, # "Mistral-7B":{ # "inf_point":"/static-proxy?url=https%3A%2F%2Frouter.huggingface.co%2Fv1", # "link":"mistralai/Mistral-7B-Instruct-v0.2", # }, # "Gemma-2-27B-it":{ # "inf_point":"/static-proxy?url=https%3A%2F%2Frouter.huggingface.co%2Fnebius%2Fv1", # "link":"google/gemma-2-27b-it-fast", # }, # "Gemma-2-2B-it":{ # "inf_point":"/static-proxy?url=https%3A%2F%2Frouter.huggingface.co%2Fnebius%2Fv1", # "link":"google/gemma-2-2b-it-fast", # }, # "Zephyr-7B-β":{ # "inf_point":"/static-proxy?url=https%3A%2F%2Frouter.huggingface.co%2Fhf-inference%2Fmodels%2FHuggingFaceH4%2Fzephyr-7b-beta%2Fv1", # "link":"HuggingFaceH4/zephyr-7b-beta", # }, } model_links_groq ={ "OpenAI-GPT-OSS-120B":{ "inf_point":"https://api.groq.com/openai/v1", "link":"openai/gpt-oss-120b", }, "Meta-Llama-3.1-8B":{ "inf_point":"https://api.groq.com/openai/v1", "link":"llama-3.1-8b-instant", }, "Kimi-K2-Instruct":{ "inf_point":"https://api.groq.com/openai/v1", "link":"moonshotai/kimi-k2-instruct", }, # "Gemma-2-9B-it":{ # "inf_point":"https://api.groq.com/openai/v1", # "link":"gemma2-9b-it", # }, } #Pull info about the model to display model_info ={ "OpenAI-GPT-OSS-120B": {'description':"""The GPT OSS 120B model is a **Large Language Model (LLM)** that's able to have question and answer interactions.\n \ \nIt was created by the [**OpenAI**](https://openai.com/research) team as an open-source initiative and has over **120 billion parameters.** \ \nThis model represents one of the largest publicly available transformer-based language models, designed for advanced reasoning, dialogue, and code understanding tasks.\n""", 'logo':'https://registry.npmmirror.com/@lobehub/icons-static-png/1.74.0/files/light/openai.png'}, "Mistral-7B": {'description':"""The Mistral model is a **Large Language Model (LLM)** that's able to have question and answer interactions.\n \ \nIt was created by the [**Mistral AI**](https://mistral.ai/news/announcing-mistral-7b/) team as has over **7 billion parameters.** \n""", 'logo':'/static-proxy?url=https%3A%2F%2Fcdn-avatars.huggingface.co%2Fv1%2Fproduction%2Fuploads%2F62dac1c7a8ead43d20e3e17a%2FwrLf5yaGC6ng4XME70w6Z.png'}, "Gemma-2-27B-it": {'description':"""The Gemma model is a **Large Language Model (LLM)** that's able to have question and answer interactions.\n \ \nIt was created by the [**Google's AI Team**](https://blog.google/technology/developers/gemma-open-models/) team as has over **27 billion parameters.** \n""", 'logo':'https://pbs.twimg.com/media/GG3sJg7X0AEaNIq.jpg'}, "Gemma-3-27B-it": {'description':"""The Gemma model is a **Large Language Model (LLM)** that's able to have question and answer interactions.\n \ \nIt was created by the [**Google's AI Team**](https://blog.google/technology/developers/gemma-open-models/) team as has over **27 billion parameters.** \n""", 'logo':'https://pbs.twimg.com/media/GG3sJg7X0AEaNIq.jpg'}, "Gemma-2-2B-it": {'description':"""The Gemma model is a **Large Language Model (LLM)** that's able to have question and answer interactions.\n \ \nIt was created by the [**Google's AI Team**](https://blog.google/technology/developers/gemma-open-models/) team as has over **2 billion parameters.** \n""", 'logo':'https://pbs.twimg.com/media/GG3sJg7X0AEaNIq.jpg'}, "Gemma-2-9B-it": {'description':"""The Gemma model is a **Large Language Model (LLM)** that's able to have question and answer interactions.\n \ \nIt was created by the [**Google's AI Team**](https://blog.google/technology/developers/gemma-open-models/) team as has over **9 billion parameters.** \n""", 'logo':'https://pbs.twimg.com/media/GG3sJg7X0AEaNIq.jpg'}, "Zephyr-7B": {'description':"""The Zephyr model is a **Large Language Model (LLM)** that's able to have question and answer interactions.\n \ \nFrom Huggingface: \n\ Zephyr is a series of language models that are trained to act as helpful assistants. \ [Zephyr 7B Gemma](https://huggingface.co/HuggingFaceH4/zephyr-7b-gemma-v0.1)\ is the third model in the series, and is a fine-tuned version of google/gemma-7b \ that was trained on on a mix of publicly available, synthetic datasets using Direct Preference Optimization (DPO)\n""", 'logo':'https://huggingface.co/HuggingFaceH4/zephyr-7b-gemma-v0.1/resolve/main/thumbnail.png'}, "Zephyr-7B-β": {'description':"""The Zephyr model is a **Large Language Model (LLM)** that's able to have question and answer interactions.\n \ \nFrom Huggingface: \n\ Zephyr is a series of language models that are trained to act as helpful assistants. \ [Zephyr-7B-β](https://huggingface.co/HuggingFaceH4/zephyr-7b-beta)\ is the second model in the series, and is a fine-tuned version of mistralai/Mistral-7B-v0.1 \ that was trained on on a mix of publicly available, synthetic datasets using Direct Preference Optimization (DPO)\n""", 'logo':'https://huggingface.co/HuggingFaceH4/zephyr-7b-alpha/resolve/main/thumbnail.png'}, "Meta-Llama-3-8B": {'description':"""The Llama (3) model is a **Large Language Model (LLM)** that's able to have question and answer interactions.\n \ \nIt was created by the [**Meta's AI**](https://llama.meta.com/) team and has over **8 billion parameters.** \n""", 'logo':'Llama_logo.png'}, "Meta-Llama-3.1-8B": {'description':"""The Llama (3.1) model is a **Large Language Model (LLM)** that's able to have question and answer interactions.\n \ \nIt was created by the [**Meta's AI**](https://llama.meta.com/) team and has over **8 billion parameters.** \n""", 'logo':'Llama3_1_logo.png'}, "DeepSeek-R1-Distill-Llama-70B": {'description':"""DeepSeek-R1-Distill-Llama-70B is a **Large Language Model (LLM)** distilled from the DeepSeek-R1 reasoning family using the Llama architecture. \ \nIt is designed to retain strong capabilities in reasoning, coding, and general text generation while being more accessible than the full DeepSeek-R1 model. \ \nLearn more on HuggingFace: https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Llama-70B""", 'logo':'/static-proxy?url=https%3A%2F%2Fcdn-avatars.huggingface.co%2Fv1%2Fproduction%2Fuploads%2F6538815d1bdb3c40db94fbfa%2FxMBly9PUMphrFVMxLX4kq.png'}, "Qwen2.5-Coder-32B-Instruct": {'description':"""Qwen2.5-Coder-32B-Instruct is a **Large Language Model (LLM)** in the Qwen2.5-Coder series tailored for code generation, reasoning, and instruction-following tasks. \ \nBuilt on the Qwen2.5 architecture, this 32B-parameter model is optimized for coding, debugging, and developer use cases. \ \nLearn more on HuggingFace: https://huggingface.co/Qwen/Qwen2.5-Coder-32B-Instruct""", 'logo':'/static-proxy?url=https%3A%2F%2Fcdn-avatars.huggingface.co%2Fv1%2Fproduction%2Fuploads%2F620760a26e3b7210c2ff1943%2F-s1gyJfvbE1RgO5iBeNOi.png'}, "Kimi-K2-Instruct": {'description':"""The Kimi-K2-Instruct model is a **Large Language Model (LLM)** that's able to have question and answer interactions.\n \ \nIt was created by the [**Moonshot AI**](https://www.moonshot.cn/) team as part of the Kimi model family. \ \nThe model is designed for instruction following, reasoning, and general conversational tasks, with a strong focus on high-quality responses and long-context understanding.\n""", 'logo':'/static-proxy?url=https%3A%2F%2Fcdn-avatars.huggingface.co%2Fv1%2Fproduction%2Fuploads%2F641c1e77c3983aa9490f8121%2FX1yT2rsaIbR9cdYGEVu0X.jpeg'}, } #Random dog images for error message random_dog = ["0f476473-2d8b-415e-b944-483768418a95.jpg", "1bd75c81-f1d7-4e55-9310-a27595fa8762.jpg", "526590d2-8817-4ff0-8c62-fdcba5306d02.jpg", "1326984c-39b0-492c-a773-f120d747a7e2.jpg", "42a98d03-5ed7-4b3b-af89-7c4876cb14c3.jpg", "8b3317ed-2083-42ac-a575-7ae45f9fdc0d.jpg", "ee17f54a-83ac-44a3-8a35-e89ff7153fb4.jpg", "027eef85-ccc1-4a66-8967-5d74f34c8bb4.jpg", "08f5398d-7f89-47da-a5cd-1ed74967dc1f.jpg", "0fd781ff-ec46-4bdc-a4e8-24f18bf07def.jpg", "0fb4aeee-f949-4c7b-a6d8-05bf0736bdd1.jpg", "6edac66e-c0de-4e69-a9d6-b2e6f6f9001b.jpg", "bfb9e165-c643-4993-9b3a-7e73571672a6.jpg", "d467a3b8-ade5-4d68-810a-95fbb32a3cfc.jpg", "5384c2a7-9b73-478e-9f32-9af9f264da1d.jpg", "59f02432-b972-4428-935b-4efb0af83456.jpg"] def reset_conversation(): ''' Resets Conversation ''' st.session_state.conversation = [] st.session_state.messages = [] return None # --- Sidebar Setup --- st.sidebar.title("Chatbot Settings") #Define model clients client_names = ["Provided API Call", "HF-Token"] client_select = st.sidebar.selectbox("Select Model Client", client_names) if "HF-Token" in client_select: try: if "API_token" not in st.session_state: st.session_state.API_token = None st.session_state.API_token = st.sidebar.text_input("Enter your Hugging Face Access Token", type="password") model_links = model_links_hf except Exception as e: st.sidebar.error(f"Credentials Error:\n\n {e}") elif "Provided API Call" in client_select: try: if "API_token" not in st.session_state: st.session_state.API_token = None st.session_state.API_token = os.environ.get('GROQ_API_TOKEN')#Should be like os.environ.get('HUGGINGFACE_API_TOKEN') model_links = model_links_groq except Exception as e: st.sidebar.error(f"Credentials Error:\n\n {e}") # Define the available models models =[key for key in model_links.keys()] # Create the sidebar with the dropdown for model selection selected_model = st.sidebar.selectbox("Select Model", models) #Create a temperature slider temp_values = st.sidebar.slider('Select a temperature value', 0.0, 1.0, (0.5)) #Add reset button to clear conversation st.sidebar.button('Reset Chat', on_click=reset_conversation, type="primary") #Reset button # Contact info # Contact info st.sidebar.markdown( "" "Created by " "" "Nigel Gebodh
" "Chatbots do not have access to real-time info. Agentic chat coming soon!" "", unsafe_allow_html=True ) st.sidebar.divider() # Add a visual separator # Create model description st.sidebar.subheader(f"About {selected_model}") st.sidebar.write(f"You're now chatting with **{selected_model}**") st.sidebar.markdown(model_info[selected_model]['description']) st.sidebar.image(model_info[selected_model]['logo']) st.sidebar.markdown("*Generated content may be inaccurate or false.*") st.sidebar.markdown("\nLearn how to build this chatbot [here](https://ngebodh.github.io/projects/2024-03-05/).") st.sidebar.markdown("\nRun into issues? \nTry coming back in a bit, GPU access might be limited or something is down.") if "prev_option" not in st.session_state: st.session_state.prev_option = selected_model if st.session_state.prev_option != selected_model: st.session_state.messages = [] st.session_state.prev_option = selected_model reset_conversation() #Pull in the model we want to use repo_id = model_links[selected_model] # initialize the client client = OpenAI( base_url=model_links[selected_model]["inf_point"],#"/static-proxy?url=https%3A%2F%2Fapi-inference.huggingface.co%2Fv1", api_key=st.session_state.API_token#os.environ.get('HUGGINGFACE_API_TOKEN')#"hf_xxx" # Replace with your token ) st.subheader(f'AI - {selected_model}') # Set a default model if selected_model not in st.session_state: st.session_state[selected_model] = model_links[selected_model] # Initialize chat history if "messages" not in st.session_state: st.session_state.messages = [] # Display chat messages from history on app rerun for message in st.session_state.messages: with st.chat_message(message["role"]): st.markdown(message["content"]) if prompt := st.chat_input(f"Hi I'm {selected_model}, ask me a question "): # Display user message in chat message container with st.chat_message("user"): st.markdown(prompt) # Add user message to chat history st.session_state.messages.append({"role": "user", "content": prompt}) #Cooldown check now = time.time() elapsed = now - st.session_state.last_request_time if elapsed < REQUEST_COOLDOWN: wait_time = round(REQUEST_COOLDOWN - elapsed, 1) st.warning(f"⏳ Please wait before sending another request.") st.stop() st.session_state.last_request_time = now if st.session_state.api_call_count >= API_CALL_LIMIT: # Add the warning to the displayed messages, but not to the history sent to the model response = f"LIMIT REACHED: Sorry, you have reached the API call limit for this session." # st.write(response) st.warning(f"Sorry, you have reached the API call limit for this session.") st.session_state.messages.append({"role": "assistant", "content": response }) else: # Display assistant response in chat message container with st.chat_message("assistant"): try: st.session_state.api_call_count += 1 # Add a spinner for better UX while waiting with st.spinner(f"Asking {selected_model}..."): stream = client.chat.completions.create( model=model_links[selected_model]["link"], messages=[ {"role": "system", "content": "You are a helpful assistant. Always respond briefly in 1–3 sentences."}, *[ {"role": m["role"], "content": m["content"]} for m in st.session_state.messages ] ], temperature=temp_values,#0.5, stream=True, max_tokens=800,#1500, #3000, ) response = st.write_stream(stream) remaining_calls = (API_CALL_LIMIT) - st.session_state.api_call_count st.markdown(f"\n\n API calls:({remaining_calls}/{API_CALL_LIMIT})", unsafe_allow_html=True) #Logging try: log_to_webhook( session_info=st.session_state.session_info, model=selected_model, prompt=prompt, response=response, temperature=temp_values, ) except Exception: pass except Exception as e: response = "😵‍💫 Looks like someone unplugged something!\ \n Either the model space is being updated or something is down.\ \n\ \n Try again later. \ \n\ \n Here's a random pic of a 🐶:" st.write(response) random_dog_pick = 'https://random.dog/'+ random_dog[np.random.randint(len(random_dog))] st.image(random_dog_pick) st.write("This was the error message:") st.write(e) st.session_state.messages.append({"role": "assistant", "content": response})