├── .streamlit └── config.toml ├── README.md ├── __init__.py ├── functions.json ├── launch.bat ├── llmon.py ├── llmon_logo.png ├── main.py └── page_icon.png /.streamlit/config.toml: -------------------------------------------------------------------------------- 1 | [client] 2 | toolbarMode = "minimal" 3 | 4 | [theme] 5 | primaryColor="#171717" 6 | backgroundColor="#212121" 7 | secondaryBackgroundColor="#171717" 8 | textColor="#e4dede" 9 | font="monospace" 10 | 11 | [runner] 12 | fastReruns = true 13 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | ![image](https://github.com/3eeps/llmon-py/assets/55860052/5603c6b4-6b68-4814-96b1-bd46bff1c78e) 2 | 3 | llmon-py is a WIP multimodal webui for inferencing Llama3-8B via text, voice and function calling. 4 | 5 | ![image](https://github.com/3eeps/llmon-py/assets/55860052/d41e04e1-e6a9-4070-93c2-686a7fe38cf2) 6 | -------------------------------------------------------------------------------- /__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/3eeps/llmon-py/20321828351440b778736400707f82435804b5a1/__init__.py -------------------------------------------------------------------------------- /functions.json: -------------------------------------------------------------------------------- 1 | [ 2 | { 3 | "name": "no_function_message", 4 | "description": "Reply to user message when the function calls below do not matches the users query.", 5 | "parameters": { 6 | "type": "object", 7 | "properties": { 8 | "user_message": { 9 | "type": "string", 10 | "description": "Message from the user." 11 | } 12 | }, 13 | "required": ["user_message"] 14 | } 15 | }, 16 | { 17 | "name": "change_voice_style", 18 | "description": "Allows the user to change the tone and style of the current text to speech model based on a prompt.", 19 | "parameters": { 20 | "type": "object", 21 | "properties": { 22 | "voice_style": { 23 | "type": "string", 24 | "description": "Style description from the user. Example: A male voice that is calm but stern." 25 | } 26 | }, 27 | "required": ["voice_style"] 28 | } 29 | }, 30 | { 31 | "name": "describe_image", 32 | "description": "Answer user query about an image.", 33 | "parameters": { 34 | "type": "object", 35 | "properties": { 36 | "user_query": { 37 | "type": "string", 38 | "description": "The query provided by the user." 39 | } 40 | }, 41 | "required": ["user_query"] 42 | } 43 | }, 44 | { 45 | "name": "create_image", 46 | "description": "Generate an image from a prompt.", 47 | "parameters": { 48 | "type": "object", 49 | "properties": { 50 | "image_prompt": { 51 | "type": "string", 52 | "description": "The prompt required to generate image." 53 | } 54 | }, 55 | "required": ["image_prompt"] 56 | } 57 | }, 58 | { 59 | "name": "video_player", 60 | "description": "Retrieve YouTube query and play the video in the chat.", 61 | "parameters": { 62 | "type": "object", 63 | "properties": { 64 | "youtube_query": { 65 | "type": "string", 66 | "description": "The YouTube query provided by the user." 67 | } 68 | }, 69 | "required": ["youtube_query"] 70 | } 71 | } 72 | ] 73 | -------------------------------------------------------------------------------- /launch.bat: -------------------------------------------------------------------------------- 1 | cd *where main.py is* 2 | streamlit run main.py 3 | -------------------------------------------------------------------------------- /llmon.py: -------------------------------------------------------------------------------- 1 | import streamlit as st 2 | import os 3 | import sounddevice 4 | import psutil 5 | import json 6 | import base64 7 | from time import sleep 8 | import keyboard 9 | from scipy.io.wavfile import write as write_wav 10 | from googlesearch import search 11 | from transformers import AutoTokenizer 12 | import torch 13 | import torchaudio 14 | from diffusers import StableDiffusion3Pipeline 15 | import soundfile as sf 16 | from parler_tts import ParlerTTSForConditionalGeneration 17 | from TTS.tts.configs.xtts_config import XttsConfig 18 | from TTS.tts.models.xtts import Xtts 19 | 20 | def model_inference(prompt=""): 21 | if st.session_state.function_calling: 22 | st.session_state['model_temperature'] = 0.85 23 | st.session_state['model_min_p'] = 0.06 24 | st.session_state['model_top_p'] = 0.0 25 | st.session_state['model_top_k'] = 0 26 | st.session_state['repeat_penalty'] = 1.1 27 | 28 | if st.session_state['model_select'] == 'DeepSeek-Coder-V2-Lite-Instruct-Q5_K_M.gguf': 29 | st.session_state['model_temperature'] = 0.1 30 | st.session_state['model_min_p'] = 0.06 31 | st.session_state['model_top_p'] = 0.0 32 | st.session_state['model_top_k'] = 0 33 | st.session_state['repeat_penalty'] = 1.1 34 | 35 | model_output = st.session_state["chat_model"](prompt=prompt, 36 | max_tokens=st.session_state['max_context'] - st.session_state.token_count, 37 | repeat_penalty=float(st.session_state['repeat_penalty']), 38 | top_k=int(st.session_state['model_top_k']), 39 | top_p=float(st.session_state['model_top_p']), 40 | min_p=float(st.session_state['model_min_p']), 41 | temperature=float(st.session_state['model_temperature'])) 42 | return model_output['choices'][0]['text'], model_output['usage']['total_tokens'] 43 | 44 | def clear_vram(): 45 | try: 46 | del st.session_state['sd3_medium'] 47 | except: pass 48 | del st.session_state['chat_model'] 49 | 50 | def stream_text(text="", delay=0.03): 51 | for word in text.split(" "): 52 | yield word + " " 53 | sleep(delay) 54 | 55 | def init_state(): 56 | default_settings_state = { 57 | 'max_context': 8192, 58 | 'gpu_layer_count': -1, 59 | 'cpu_core_count': 8, 60 | 'cpu_batch_count': 8, 61 | 'batch_size': 256, 62 | 'model_temperature': 0.85, 63 | 'model_top_p': 0.0, 64 | 'model_top_k': 0, 65 | 'model_min_p': 0.05, 66 | 'repeat_penalty': 1.1, 67 | 'message_list': [], 68 | 'init_app': False, 69 | 'model_select': None} 70 | 71 | st.session_state.model_list = ['Meta-Llama-3-8B-Instruct.Q6_K.gguf', 'DeepSeek-Coder-V2-Lite-Instruct-Q5_K_M.gguf'] 72 | st.session_state.show_generated_image = False 73 | st.session_state.show_uploaded_image = False 74 | st.session_state.bytes_data = "" 75 | st.session_state.sdxl_base64 = "" 76 | st.session_state.model_param_settings = False 77 | st.session_state.token_count = 0 78 | st.session_state.start_app = False 79 | st.session_state.function_calling = False 80 | st.session_state.function_results = "" 81 | st.session_state.custom_template = "" 82 | st.session_state.video_link = "" 83 | st.session_state.user_chat = False 84 | st.session_state.new_voice_reply = "" 85 | 86 | with open("functions.json", "r") as file: 87 | st.session_state.functions = json.load(file) 88 | 89 | for key, value in default_settings_state.items(): 90 | if key not in st.session_state: 91 | st.session_state[key] = value 92 | 93 | def hide_deploy_button(): 94 | st.markdown( 95 | r""" 96 | 101 | """, unsafe_allow_html=True) 102 | 103 | def load_conversation(): 104 | with open('messsage_list.json', 'r') as file: 105 | st.session_state['message_list'] = json.load(file) 106 | with open('streamlit_list.json', 'r') as file: 107 | st.session_state.messages = json.load(file) 108 | 109 | def save_conversation(message_list, streamlit_message_list): 110 | with open('messsage_list.json', 'w') as file: 111 | json.dump(message_list, file) 112 | with open('streamlit_list.json', 'w') as file: 113 | json.dump(streamlit_message_list, file) 114 | 115 | def unload_model(): 116 | sleep(0.5) 117 | del st.session_state['chat_model'] 118 | 119 | def sidebar(): 120 | col1, col2, col3 = st.columns([3,1,1]) 121 | with col1: 122 | st.title('🍋 llmon-py') 123 | with col2: 124 | if st.button(label="📂", help='load previous conversation'): 125 | load_conversation() 126 | with col3: 127 | if st.button(label='✨', help='start a new chat'): 128 | save_conversation(message_list=st.session_state['message_list'], streamlit_message_list=st.session_state.messages) 129 | st.session_state['message_list'] = [] 130 | st.session_state.messages = [] 131 | st.session_state.bytes_data = None 132 | st.session_state.sdxl_base64 = None 133 | st.session_state.function_calling = False 134 | st.session_state.custom_template = "" 135 | st.session_state.token_count = 0 136 | st.session_state.video_link = None 137 | 138 | loaded_model_title = "" 139 | if st.session_state['model_select'] == 'DeepSeek-Coder-V2-Lite-Instruct-Q5_K_M.gguf': 140 | loaded_model_title = "deepseek-coder-v2" 141 | if st.session_state['model_select'] == 'Meta-Llama-3-8B-Instruct.Q6_K.gguf': 142 | loaded_model_title = "meta-llama-3" 143 | st.caption(f"running :green[{loaded_model_title}]") 144 | st.caption("") 145 | 146 | #st.session_state['model_select'] = st.selectbox(label='model list', options=st.session_state.model_list, label_visibility='hidden', help='load your model of choice', on_change=unload_model) 147 | uploaded_file = st.file_uploader(label='file uploader', label_visibility='collapsed', type=['png', 'jpeg'], disabled=True) 148 | if uploaded_file: 149 | st.session_state.bytes_data = uploaded_file.getvalue() 150 | with open("ocr_upload_image.png", 'wb') as file: 151 | file.write(st.session_state.bytes_data) 152 | 153 | with open("ocr_upload_image.png", "rb") as f: 154 | st.session_state.bytes_data = base64.b64encode(f.read()).decode() 155 | os.remove('ocr_upload_image.png') 156 | 157 | st.caption(body="custom model template") 158 | st.session_state.custom_template = st.text_area(label='custom prompt', value="", label_visibility='collapsed') 159 | disable_function_call = False 160 | if st.session_state['model_select'] == 'DeepSeek-Coder-V2-Lite-Instruct-Q5_K_M.gguf': 161 | disable_function_call = True 162 | st.session_state.function_calling = st.checkbox(':orange[enable function calling] :green[(beta)]', value=st.session_state.function_calling, help='currently allows user to :green[generate images and find youtube videos]. may not produce desired output.', disabled=disable_function_call) 163 | st.session_state.model_param_settings = st.checkbox(':orange[model parameters]', value=st.session_state.model_param_settings, help='tweak model parameters to steer llama-3s output.') 164 | if st.session_state.model_param_settings: 165 | st.caption(f"ctx:{st.session_state.token_count}/{st.session_state['max_context']}") 166 | temp_help = "determinines whether the output is more random and creative or more predictable. :green[a higher temperature will result in lower probability], i.e more creative outputs." 167 | top_p_help = "controls the diversity of the generated text by only considering tokens with the highest probability mass. :green[top_p = 0.1: only tokens within the top 10% probability are considered. 0.9: considers tokens within the top 90% probability]." 168 | top_k_help = "limits the model's output to the top-k most probable tokens at each step. This can help reduce incoherent or nonsensical output by restricting the model's vocabulary. :green[a top-K of 1 means the next selected token is the most probable among all tokens in the model's vocabulary]." 169 | min_p_help = "different from top k or top p, sets a minimum percentage requirement to consider tokens relative to the largest token probability. :green[for example, min p = 0.1 is equivalent to only considering tokens at least 1/10th the top token probability]." 170 | rep_help = "helps the model generate more diverse content instead of repeating previous phrases. Repetition is prevented by applying a high penalty to phrases or words that tend to be repeated. :green[a higher penalty generally results in more diverse outputs, whilst a lower value might lead to more repetition]." 171 | st.session_state['model_temperature'] = st.text_input(label=':orange[temperature]', value=st.session_state['model_temperature'], help=temp_help) 172 | st.session_state['model_top_p'] = st.text_input(label=':orange[top p]', value=st.session_state['model_top_p'], help=top_p_help) 173 | st.session_state['model_top_k'] = st.text_input(label=':orange[top k]', value=st.session_state['model_top_k'], help=top_k_help) 174 | st.session_state['model_min_p'] = st.text_input(label=':orange[min p]', value=st.session_state['model_min_p'], help=min_p_help) 175 | st.session_state['repeat_penalty'] = st.text_input(label=':orange[repetition penalty]', value=st.session_state['repeat_penalty'], help=rep_help) 176 | 177 | bottom_col1, bottom_col2, bottom_col3 = st.columns([1,1,1]) 178 | with bottom_col2: 179 | if st.button(":orange[shutdown]", help='shut down app on server side'): 180 | shut_down_app() 181 | 182 | def shut_down_app(): 183 | save_conversation(message_list=st.session_state['message_list'], streamlit_message_list=st.session_state.messages) 184 | if st.session_state.start_app: 185 | clear_vram() 186 | try: 187 | os.remove('image.txt') 188 | except: pass 189 | try: 190 | os.remove('.google-cookie') 191 | except: pass 192 | keyboard.press_and_release('ctrl+w') 193 | llmon_process_id = os.getpid() 194 | process = psutil.Process(llmon_process_id) 195 | process.terminate() 196 | 197 | class ChatTemplate: 198 | def chat_template(prompt="", function_result=""): 199 | system_message = "" 200 | template = "" 201 | if st.session_state['model_select'] == 'Meta-Llama-3-8B-Instruct.Q6_K.gguf': 202 | system_message = f"""You are an helpful AI assistant, answer any request the user may have. Share postive or negative considerations when appropriate, but keep them concise. Conversation history: {st.session_state['message_list']}""" 203 | template = f"""<|begin_of_text|<|start_header_id|>system<|end_header_id|>{system_message}<|eot_id|><|start_header_id|>user<|end_header_id|>{prompt}<|eot_id|><|start_header_id|>assistant<|end_header_id|>""" 204 | if st.session_state['model_select'] == 'DeepSeek-Coder-V2-Lite-Instruct-Q5_K_M.gguf': 205 | system_message = f"""You are an helpful AI coding assistant, answer any request the user may have. Chat history: {st.session_state['message_list']}""" 206 | template = f"""<｜begin▁of▁sentence｜>{system_message} 207 | 208 | User: {prompt} 209 | 210 | Assistant: <｜end▁of▁sentence｜>Assistant:""" 211 | 212 | #if st.session_state.function_calling: 213 | # system_message = f"""As an AI model with function calling support, you are provided with the following functions: {json.dumps(st.session_state.functions)} 214 | # When the user asks a question that can be answered with a function, do not describe the function or wrap the function in "||", output the function filled with the appropriate data required as a Python dictionary.""" 215 | # system_message_plus_example = """Example message: 'Hello there!' Your reply: '{'function_name': 'user_chat', 'parameters': {'user_message': 'Hello there!'}}'""" 216 | 217 | if st.session_state.function_calling: 218 | system_message = f"""As an AI model with function calling support, you are provided with the following functions: {json.dumps(st.session_state.functions)} 219 | When the user asks a question that can be answered with a function, please do not describe the function. Only output the function filled with the appropriate data required as a Python style dictionary.""" 220 | system_message_plus_example = """Example: User: 'play brother ali take me home' Your reply: '{'function_name': 'video_player', 'parameters': {'youtube_query': 'play brother ali take me home'}}'""" 221 | template = f"""<|begin_of_text|<|start_header_id|>system<|end_header_id|>{system_message + system_message_plus_example}<|eot_id|><|start_header_id|>user<|end_header_id|>{prompt}<|eot_id|><|start_header_id|>assistant<|end_header_id|>""" 222 | 223 | if len(function_result) > 1: 224 | system_message = f"""Using this data that is up to date, reply to the user using it: {function_result}. The question the user asked was:""" 225 | template = f"""<|begin_of_text|<|start_header_id|>system<|end_header_id|>{system_message}<|eot_id|><|start_header_id|>user<|end_header_id|>{prompt}<|eot_id|><|start_header_id|>assistant<|end_header_id|>""" 226 | 227 | if len(st.session_state.custom_template) > 1: 228 | system_message = f"""{st.session_state.custom_template} Chat history: {st.session_state['message_list']}""" 229 | template = f"""<|begin_of_text|<|start_header_id|>system<|end_header_id|>{system_message}<|eot_id|><|start_header_id|>user<|end_header_id|>{prompt}<|eot_id|><|start_header_id|>assistant<|end_header_id|>""" 230 | 231 | return template 232 | 233 | class Audio: 234 | def voice_to_text(): 235 | rec_user_voice = sounddevice.rec(int(st.session_state['user_audio_length']) * 44100, samplerate=44100, channels=2) 236 | sounddevice.wait() 237 | write_wav(filename='user_output.wav', rate=44100, data=rec_user_voice) 238 | user_voice_data = st.session_state['speech_tt_model'].transcribe('user_output.wav', speed_up=True) 239 | os.remove(f"user_output.wav") 240 | text_data = [] 241 | for voice in user_voice_data: 242 | text_data.append(voice.text) 243 | combined_text = ' '.join(text_data) 244 | return combined_text 245 | 246 | class Functions: 247 | def find_youtube_link(user_query): 248 | search_helper = 'youtube' 249 | search_query = user_query + search_helper 250 | for youtube_link in search(query=search_query, tld="co.in", num=1, stop=1, pause=2): 251 | print(youtube_link) 252 | return youtube_link 253 | return 'https://www.youtube.com/watch?v=dQw4w9WgXcQ' 254 | 255 | def change_llm_voice(voice_description): 256 | #load models 257 | device = "cuda:0" 258 | model = ParlerTTSForConditionalGeneration.from_pretrained("parler-tts/parler_tts_mini_v0.1").to(device, dtype=torch.float32) 259 | tokenizer = AutoTokenizer.from_pretrained("parler-tts/parler_tts_mini_v0.1") 260 | 261 | config = XttsConfig() 262 | config.load_json("./xtts_model/config.json") 263 | model = Xtts.init_from_config(config) 264 | model.load_checkpoint(config, checkpoint_dir="./xtts_model/") 265 | model.cuda() 266 | 267 | style_prompt = "It took me quite a long time to develop a voice and now that I have it I am not going to be silent." 268 | 269 | 270 | #### easy way!!!! make style voice here... use the voice with built in tts, even if its slow 271 | #### trick!!! when using tts... ask the llm to keep anwers very brief. less then a paragraph. should help inference time! 272 | 273 | 274 | 275 | #run parlor with user voice description 276 | input_ids = tokenizer(voice_description, return_tensors="pt").input_ids.to(device) 277 | prompt_input_ids = tokenizer(style_prompt, return_tensors="pt").input_ids.to(device) 278 | generation = model.generate(input_ids=input_ids, prompt_input_ids=prompt_input_ids).to(torch.float32) 279 | audio_arr = generation.cpu().numpy().squeeze() 280 | sf.write("new_voice_style.wav", audio_arr, model.config.sampling_rate) 281 | 282 | # send new voice style to xtts to say 'how is it?' trick :P 283 | language = 'en' 284 | #st.session_state.new_voice_reply = model_inference(prompt=f"The user has changed your text to speech voice! Here is the voice description from the user: {voice_description}. Say something fun to show off your new voice!") 285 | gpt_cond_latent, speaker_embedding = model.get_conditioning_latents(audio_path=["new_voice_style.wav"]) 286 | tts_chunks = model.inference_stream( 287 | prompt, 288 | language, 289 | gpt_cond_latent, 290 | speaker_embedding) 291 | 292 | chunk_list = [] 293 | for i, chunk in enumerate(tts_chunks): 294 | chunk_list.append(chunk) 295 | print(f"Received chunk {i} of audio length {chunk.shape[-1]}") 296 | wav_chunk = torch.cat(chunk_list, dim=0) 297 | torchaudio.save("new_llm_final_voice.wav", wav_chunk.squeeze().unsqueeze(0).cpu(), 24000) 298 | return st.session_state.new_voice_reply 299 | 300 | class SD3Medium: 301 | def init(): 302 | st.session_state['sd3_medium'] = StableDiffusion3Pipeline.from_pretrained( 303 | "stabilityai/stable-diffusion-3-medium-diffusers", torch_dtype=torch.float16, 304 | text_encoder_3=None, 305 | tokenizer_3=None 306 | ).to("cuda") 307 | 308 | def generate_image(prompt=""): 309 | image = st.session_state['sd3_medium']( 310 | prompt, 311 | negative_prompt="", 312 | num_inference_steps=28, 313 | guidance_scale=7.0 314 | ).images[0] 315 | image.save('image_sd3.png') 316 | 317 | with open("image_sd3.png", "rb") as f: 318 | st.session_state.sdxl_base64 = base64.b64encode(f.read()).decode() 319 | os.remove('image_sd3.png') -------------------------------------------------------------------------------- /llmon_logo.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/3eeps/llmon-py/20321828351440b778736400707f82435804b5a1/llmon_logo.png -------------------------------------------------------------------------------- /main.py: -------------------------------------------------------------------------------- 1 | import streamlit as st 2 | st.set_page_config(page_title="llmon-py", page_icon="page_icon.png", layout="centered", initial_sidebar_state="collapsed") 3 | lcol1, lcol2, lcol3 = st.columns([1,1,1]) 4 | with lcol2: 5 | with st.spinner('🍋 initializing...'): 6 | import llmon 7 | from pywhispercpp.model import Model 8 | from llama_cpp import Llama 9 | 10 | llmon.hide_deploy_button() 11 | if 'init_app' not in st.session_state: 12 | llmon.init_state() 13 | if st.session_state.init_app == False: 14 | col1, col2, col3 = st.columns([1,1,1]) 15 | with col2: 16 | st.write('🍋 welcome to llmon-py!') 17 | model_picked = st.selectbox(label='first model list', options=st.session_state.model_list, label_visibility='hidden') 18 | if st.button(":orange[click here to start!]"): 19 | st.session_state['model_select'] = model_picked 20 | st.session_state.start_app = True 21 | st.session_state.init_app = True 22 | st.rerun() 23 | 24 | lcol1, lcol2, lcol3 = st.columns([1,1,1]) 25 | with lcol2: 26 | with st.spinner('🍋 loading your model...'): 27 | if "messages" not in st.session_state and st.session_state.start_app: 28 | st.session_state.messages = [] 29 | 30 | if "chat_model" not in st.session_state and st.session_state.start_app: 31 | st.session_state["chat_model"] = Llama(model_path=f"./{st.session_state['model_select']}", 32 | n_batch=st.session_state['batch_size'], 33 | n_threads=st.session_state['cpu_core_count'], 34 | n_threads_batch=st.session_state['cpu_batch_count'], 35 | n_gpu_layers=int(st.session_state['gpu_layer_count']), 36 | n_ctx=st.session_state['max_context']) 37 | 38 | #if "sd3_medium" not in st.session_state and st.session_state.start_app and st.session_state['model_select'] != 'DeepSeek-Coder-V2-Lite-Instruct-Q5_K_M.gguf': 39 | #llmon.SD3Medium.init() 40 | 41 | if 'speech_tt_model' not in st.session_state and st.session_state.start_app: 42 | st.session_state['speech_tt_model'] = Model(models_dir='./speech models', n_threads=10) 43 | 44 | if st.session_state.start_app: 45 | with st.sidebar: 46 | llmon.sidebar() 47 | for message in st.session_state.messages: 48 | with st.chat_message(name=message['role']): 49 | try: 50 | st.markdown(f"""""", unsafe_allow_html=True) 51 | except: pass 52 | st.write(message["content"]) 53 | 54 | if user_text_input:= st.chat_input(placeholder=''): 55 | user_input = user_text_input 56 | if user_input == " ": 57 | try: user_input = llmon.Audio.voice_to_text() 58 | except: pass 59 | 60 | with st.chat_message(name="user"): 61 | st.write(user_input) 62 | if st.session_state.show_uploaded_image: 63 | st.session_state.messages.append({"role": "user", "content": user_input, "image": st.session_state.bytes_data}) 64 | elif st.session_state.show_uploaded_image == False: 65 | st.session_state.messages.append({"role": "user", "content": user_input}) 66 | 67 | st.session_state['message_list'].append(f"""user: {user_input}""") 68 | st.session_state.show_uploaded_image = False 69 | 70 | with st.spinner('🍋 generating...'): 71 | final_template = llmon.ChatTemplate.chat_template(prompt=user_input) 72 | model_output_text, st.session_state.token_count = llmon.model_inference(prompt=final_template) 73 | user_chat = False 74 | if st.session_state.function_calling: 75 | try: 76 | model_output_dict = eval(model_output_text) 77 | model_dict_values = [value for key, value in model_output_dict.items()] 78 | func_value_dict = model_dict_values.pop(1) 79 | output_function = model_dict_values.pop(0) 80 | value_name = [value for key, value in func_value_dict.items()] 81 | print(f"""{model_output_dict}\n{func_value_dict}\n{output_function}\n{value_name}""") 82 | 83 | if output_function == "no_function_message": 84 | user_chat = True 85 | text_output = True 86 | 87 | if output_function == "change_voice_style": 88 | llmon.Functions.change_llm_voice(voice_description=value_name[0]) 89 | voice_reply = True 90 | 91 | if output_function == "describe_image": 92 | st.session_state.show_uploaded_image = True 93 | text_output = True 94 | 95 | if output_function == "create_image": 96 | llmon.SD3Medium.generate_image(prompt=value_name[0]) 97 | st.session_state.show_generated_image = True 98 | 99 | if output_function == "video_player": 100 | st.session_state.video_link = llmon.Functions.find_youtube_link(user_query=value_name[0]) 101 | 102 | final_function_template = llmon.ChatTemplate.chat_template(prompt=user_input, function_result=st.session_state.function_results) 103 | model_output_text = "" 104 | if user_chat: 105 | st.session_state.function_calling = False 106 | final_template = llmon.ChatTemplate.chat_template(prompt=user_input) 107 | model_output_text, st.session_state.token_count = llmon.model_inference(prompt=final_template) 108 | st.session_state.function_calling = True 109 | if text_output and user_chat == False: 110 | model_output_text, st.session_state.token_count = llmon.model_inference(prompt=final_function_template) 111 | if voice_reply: 112 | st.session_state.function_calling = False 113 | voice_prompt = f"Pretend that I just gave you a new text to speech voice with this description: {value_name[0]}. Say something fun to show it off!" 114 | final_template = llmon.ChatTemplate.chat_template(prompt=voice_prompt) 115 | model_output_text, st.session_state.token_count = llmon.model_inference(prompt=final_function_template) 116 | st.session_state.function_calling = True 117 | except: pass 118 | 119 | with st.chat_message(name="assistant"): 120 | st.write_stream(llmon.stream_text(text=model_output_text)) 121 | if st.session_state.show_generated_image: 122 | st.markdown(f"""""", unsafe_allow_html=True) 123 | st.session_state.messages.append({"role": "assistant", "content": model_output_text, "image": st.session_state.sdxl_base64}) 124 | st.session_state['message_list'].append(f"You: Generated an image based on the users request.") 125 | 126 | if st.session_state.show_generated_image == False: 127 | st.session_state.messages.append({"role": "assistant", "content": model_output_text}) 128 | st.session_state['message_list'].append(f"You: {model_output_text}") 129 | 130 | if st.session_state.video_link: 131 | st.video(data=st.session_state.video_link) 132 | st.session_state.messages.append({"role": "assistant", "content": f"{st.session_state.video_link}"}) 133 | st.session_state['message_list'].append(f"You: Displayed a Youtube video based on the users search request.") 134 | 135 | st.session_state.show_generated_image = False 136 | st.session_state.video_link = None -------------------------------------------------------------------------------- /page_icon.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/3eeps/llmon-py/20321828351440b778736400707f82435804b5a1/page_icon.png --------------------------------------------------------------------------------