Spaces:
				
			
			
	
			
			
		Running
		
			on 
			
			Zero
	
	
	
			
			
	
	
	
	
		
		
		Running
		
			on 
			
			Zero
	Update app.py
Browse files
    	
        app.py
    CHANGED
    
    | 
         @@ -1,230 +1,106 @@ 
     | 
|
| 1 | 
         
             
            import os
         
     | 
| 2 | 
         
            -
            import  
     | 
| 
         | 
|
| 
         | 
|
| 3 | 
         
             
            import torch
         
     | 
| 4 | 
         
             
            import gradio as gr
         
     | 
| 5 | 
         
            -
            from  
     | 
| 6 | 
         
            -
            from transformers import (
         
     | 
| 7 | 
         
            -
                AutoTokenizer,
         
     | 
| 8 | 
         
            -
                AutoModelForCausalLM,
         
     | 
| 9 | 
         
            -
                TextIteratorStreamer,
         
     | 
| 10 | 
         
            -
            )
         
     | 
| 11 | 
         
             
            from huggingface_hub import login
         
     | 
| 12 | 
         
            -
            import threading
         
     | 
| 13 | 
         
             
            import spaces
         
     | 
| 14 | 
         | 
| 15 | 
         
            -
            """
         
     | 
| 16 | 
         
            -
            Gradio chat app for facebook/MobileLLM-Pro
         
     | 
| 17 | 
         
            -
            - Uses the model's chat template when using the "instruct" subfolder
         
     | 
| 18 | 
         
            -
            - Streams tokens to the Gradio UI
         
     | 
| 19 | 
         
            -
            - Minimal controls: max_new_tokens, temperature, top_p
         
     | 
| 20 | 
         
            -
            - Optional HF_TOKEN login via env var or textbox
         
     | 
| 21 | 
         
            -
             
     | 
| 22 | 
         
            -
            To run locally:
         
     | 
| 23 | 
         
            -
              pip install -U gradio transformers accelerate sentencepiece huggingface_hub
         
     | 
| 24 | 
         
            -
              HF_TOKEN=xxxx python app.py
         
     | 
| 25 | 
         
            -
             
     | 
| 26 | 
         
            -
            On Hugging Face Spaces:
         
     | 
| 27 | 
         
            -
              - Remove explicit login() call or set HF_TOKEN as a secret
         
     | 
| 28 | 
         
            -
            """
         
     | 
| 29 | 
         
            -
             
     | 
| 30 | 
         
             
            MODEL_ID = "facebook/MobileLLM-Pro"
         
     | 
| 31 | 
         
            -
             
     | 
| 32 | 
         
            -
             
     | 
| 33 | 
         
            -
             
     | 
| 34 | 
         
            -
             
     | 
| 35 | 
         | 
| 36 | 
         
            -
            #  
     | 
| 37 | 
         
            -
            HF_TOKEN = os.getenv("HF_TOKEN")
         
     | 
| 38 | 
         
             
            if HF_TOKEN:
         
     | 
| 39 | 
         
             
                try:
         
     | 
| 
         | 
|
| 40 | 
         
             
                    login(token=HF_TOKEN)
         
     | 
| 41 | 
         
            -
             
     | 
| 42 | 
         
            -
             
     | 
| 43 | 
         
            -
                     
     | 
| 44 | 
         
            -
             
     | 
| 45 | 
         
            -
             
     | 
| 46 | 
         
            -
             
     | 
| 47 | 
         
            -
             
     | 
| 48 | 
         
            -
             
     | 
| 49 | 
         
            -
             
     | 
| 50 | 
         
            -
             
     | 
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 51 | 
         
             
                )
         
     | 
| 52 | 
         
            -
                 
     | 
| 53 | 
         
             
                    MODEL_ID,
         
     | 
| 54 | 
         
             
                    trust_remote_code=True,
         
     | 
| 55 | 
         
            -
                    subfolder= 
     | 
| 56 | 
         
             
                    torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
         
     | 
| 57 | 
         
             
                    low_cpu_mem_usage=True,
         
     | 
| 58 | 
         
             
                    device_map="auto" if torch.cuda.is_available() else None,
         
     | 
| 59 | 
         
             
                )
         
     | 
| 60 | 
         
            -
             
     | 
| 61 | 
         
            -
             
     | 
| 62 | 
         
            -
                 
     | 
| 63 | 
         
            -
             
     | 
| 64 | 
         
            -
             
     | 
| 65 | 
         
            -
                model.eval()
         
     | 
| 66 | 
         
            -
                print("[INFO] Model loaded.")
         
     | 
| 67 | 
         
            -
                return tokenizer, model
         
     | 
| 68 | 
         
            -
             
     | 
| 69 | 
         | 
| 70 | 
         
             
            def _history_to_messages(history: List[Tuple[str, str]]) -> List[Dict[str, str]]:
         
     | 
| 71 | 
         
            -
                 
     | 
| 72 | 
         
            -
                messages: List[Dict[str, str]] = []
         
     | 
| 73 | 
         
             
                for user_msg, bot_msg in history:
         
     | 
| 74 | 
         
             
                    if user_msg:
         
     | 
| 75 | 
         
            -
                         
     | 
| 76 | 
         
             
                    if bot_msg:
         
     | 
| 77 | 
         
            -
                         
     | 
| 78 | 
         
            -
                return  
     | 
| 79 | 
         
            -
             
     | 
| 80 | 
         | 
| 81 | 
         
             
            @spaces.GPU(duration=120)
         
     | 
| 82 | 
         
            -
            def generate_stream(
         
     | 
| 83 | 
         
            -
                message: str,
         
     | 
| 84 | 
         
            -
                history: List[Tuple[str, str]],
         
     | 
| 85 | 
         
            -
                version: str,
         
     | 
| 86 | 
         
            -
                max_new_tokens: int,
         
     | 
| 87 | 
         
            -
                temperature: float,
         
     | 
| 88 | 
         
            -
                top_p: float,
         
     | 
| 89 | 
         
            -
                use_chat_template: bool,
         
     | 
| 90 | 
         
            -
                state: Dict[str, Any],
         
     | 
| 91 | 
         
            -
            ):
         
     | 
| 92 | 
         
            -
                """Streaming text generator compatible with gr.ChatInterface.
         
     | 
| 93 | 
         
            -
             
     | 
| 94 | 
         
            -
                Args map to UI controls. `state` holds tokenizer/model between calls.
         
     | 
| 95 | 
         
             
                """
         
     | 
| 96 | 
         
            -
                 
     | 
| 97 | 
         
            -
                 
     | 
| 98 | 
         
            -
             
     | 
| 99 | 
         
            -
                 
     | 
| 100 | 
         
            -
                if (
         
     | 
| 101 | 
         
            -
                    tokenizer is None
         
     | 
| 102 | 
         
            -
                    or model is None
         
     | 
| 103 | 
         
            -
                    or state.get("version") != version
         
     | 
| 104 | 
         
            -
                ):
         
     | 
| 105 | 
         
            -
                    tokenizer, model = load_model(version)
         
     | 
| 106 | 
         
            -
                    state["tokenizer"], state["model"], state["version"] = tokenizer, model, version
         
     | 
| 107 | 
         
            -
             
     | 
| 108 | 
         
            -
                device = next(model.parameters()).device
         
     | 
| 109 | 
         
            -
             
     | 
| 110 | 
         
            -
                if use_chat_template and version == "instruct":
         
     | 
| 111 | 
         
            -
                    messages = _history_to_messages(history) + [
         
     | 
| 112 | 
         
            -
                        {"role": "user", "content": message}
         
     | 
| 113 | 
         
            -
                    ]
         
     | 
| 114 | 
         
            -
                    inputs = tokenizer.apply_chat_template(
         
     | 
| 115 | 
         
            -
                        messages,
         
     | 
| 116 | 
         
            -
                        return_tensors="pt",
         
     | 
| 117 | 
         
            -
                        add_generation_prompt=True,
         
     | 
| 118 | 
         
            -
                    ).to(device)
         
     | 
| 119 | 
         
            -
                    input_ids = inputs if isinstance(inputs, torch.Tensor) else inputs["input_ids"]
         
     | 
| 120 | 
         
            -
                else:
         
     | 
| 121 | 
         
            -
                    input_ids = tokenizer(
         
     | 
| 122 | 
         
            -
                        message,
         
     | 
| 123 | 
         
            -
                        return_tensors="pt",
         
     | 
| 124 | 
         
            -
                        add_special_tokens=True,
         
     | 
| 125 | 
         
            -
                    )["input_ids"].to(device)
         
     | 
| 126 | 
         | 
| 127 | 
         
            -
                 
     | 
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 128 | 
         | 
| 
         | 
|
| 129 | 
         
             
                gen_kwargs = dict(
         
     | 
| 130 | 
         
             
                    input_ids=input_ids,
         
     | 
| 131 | 
         
            -
                    max_new_tokens= 
     | 
| 132 | 
         
            -
                    do_sample= 
     | 
| 133 | 
         
            -
                    temperature= 
     | 
| 134 | 
         
            -
                    top_p=float( 
     | 
| 135 | 
         
            -
                    pad_token_id= 
     | 
| 136 | 
         
            -
                    eos_token_id= 
     | 
| 137 | 
         
             
                    streamer=streamer,
         
     | 
| 138 | 
         
             
                )
         
     | 
| 139 | 
         | 
| 140 | 
         
            -
                thread = threading.Thread(target= 
     | 
| 141 | 
         
             
                thread.start()
         
     | 
| 142 | 
         | 
| 143 | 
         
            -
                 
     | 
| 144 | 
         
             
                for new_text in streamer:
         
     | 
| 145 | 
         
            -
                     
     | 
| 146 | 
         
            -
                    yield  
     | 
| 147 | 
         
            -
             
     | 
| 148 | 
         
            -
             
     | 
| 149 | 
         
            -
             
     | 
| 150 | 
         
            -
                gr. 
     | 
| 151 | 
         
            -
                 
     | 
| 152 | 
         
            -
                 
     | 
| 153 | 
         
            -
             
     | 
| 154 | 
         
            -
                """)
         
     | 
| 155 | 
         
            -
                gr.Markdown(
         
     | 
| 156 | 
         
            -
                    "<div style='text-align: center;'>Built with <a href='https://huggingface.co/spaces/akhaliq/anycoder'>anycoder</a></div>",
         
     | 
| 157 | 
         
            -
                    elem_id="anycoder_attribution"
         
     | 
| 158 | 
         
            -
                )
         
     | 
| 159 | 
         
            -
             
     | 
| 160 | 
         
            -
                with gr.Row():
         
     | 
| 161 | 
         
            -
                    with gr.Column(scale=3):
         
     | 
| 162 | 
         
            -
                        chatbot = gr.Chatbot(height=420, label="MobileLLM-Pro")
         
     | 
| 163 | 
         
            -
                        msg = gr.Textbox(placeholder="Ask me anything…", scale=1)
         
     | 
| 164 | 
         
            -
                        submit = gr.Button("Send", variant="primary")
         
     | 
| 165 | 
         
            -
                        clear_btn = gr.Button("Clear chat")
         
     | 
| 166 | 
         
            -
                    with gr.Column(scale=2):
         
     | 
| 167 | 
         
            -
                        version = gr.Dropdown(["base", "instruct"], value=DEFAULT_VERSION, label="Subfolder (version)")
         
     | 
| 168 | 
         
            -
                        use_chat_template = gr.Checkbox(value=True, label="Use chat template (instruct only)")
         
     | 
| 169 | 
         
            -
                        max_new = gr.Slider(32, 1024, value=DEFAULT_MAX_NEW_TOKENS, step=8, label="Max new tokens")
         
     | 
| 170 | 
         
            -
                        temperature = gr.Slider(0.0, 1.5, value=DEFAULT_TEMPERATURE, step=0.05, label="Temperature")
         
     | 
| 171 | 
         
            -
                        top_p = gr.Slider(0.1, 1.0, value=DEFAULT_TOP_P, step=0.01, label="Top-p")
         
     | 
| 172 | 
         
            -
                        hf_token_box = gr.Textbox(value=os.getenv("HF_TOKEN", ""), label="HF_TOKEN (optional)")
         
     | 
| 173 | 
         
            -
             
     | 
| 174 | 
         
            -
                        state = gr.State({"tokenizer": None, "model": None, "version": None})
         
     | 
| 175 | 
         
            -
             
     | 
| 176 | 
         
            -
                def _maybe_login(token: str):
         
     | 
| 177 | 
         
            -
                    token = (token or "").strip()
         
     | 
| 178 | 
         
            -
                    if not token:
         
     | 
| 179 | 
         
            -
                        return "(No token provided; skipping login)"
         
     | 
| 180 | 
         
            -
                    try:
         
     | 
| 181 | 
         
            -
                        login(token=token)
         
     | 
| 182 | 
         
            -
                        return "Logged in to Hugging Face Hub."
         
     | 
| 183 | 
         
            -
                    except Exception as e:
         
     | 
| 184 | 
         
            -
                        return f"Login failed: {e}"
         
     | 
| 185 | 
         
            -
             
     | 
| 186 | 
         
            -
                login_btn = gr.Button("Login to HF (optional)")
         
     | 
| 187 | 
         
            -
                login_status = gr.Markdown()
         
     | 
| 188 | 
         
            -
                login_btn.click(_maybe_login, inputs=[hf_token_box], outputs=[login_status])
         
     | 
| 189 | 
         
            -
             
     | 
| 190 | 
         
            -
                def user_submit(user_message, chat_history):
         
     | 
| 191 | 
         
            -
                    # Immediately append the user's message so the stream shows inline
         
     | 
| 192 | 
         
            -
                    return "", chat_history + [(user_message, None)]
         
     | 
| 193 | 
         
            -
             
     | 
| 194 | 
         
            -
                def bot_respond(chat_history, version, max_new, temperature, top_p, use_chat_template, state):
         
     | 
| 195 | 
         
            -
                    # The last tuple is (user, None)
         
     | 
| 196 | 
         
            -
                    user_message = chat_history[-1][0] if chat_history else ""
         
     | 
| 197 | 
         
            -
                    partials = generate_stream(
         
     | 
| 198 | 
         
            -
                        user_message,
         
     | 
| 199 | 
         
            -
                        chat_history[:-1],
         
     | 
| 200 | 
         
            -
                        version,
         
     | 
| 201 | 
         
            -
                        int(max_new),
         
     | 
| 202 | 
         
            -
                        float(temperature),
         
     | 
| 203 | 
         
            -
                        float(top_p),
         
     | 
| 204 | 
         
            -
                        bool(use_chat_template),
         
     | 
| 205 | 
         
            -
                        state,
         
     | 
| 206 | 
         
            -
                    )
         
     | 
| 207 | 
         
            -
                    # Stream tokens to the last assistant message slot
         
     | 
| 208 | 
         
            -
                    for chunk in partials:
         
     | 
| 209 | 
         
            -
                        chat_history[-1] = (chat_history[-1][0], chunk)
         
     | 
| 210 | 
         
            -
                        yield chat_history
         
     | 
| 211 | 
         
            -
             
     | 
| 212 | 
         
            -
                msg.submit(user_submit, [msg, chatbot], [msg, chatbot]).then(
         
     | 
| 213 | 
         
            -
                    bot_respond,
         
     | 
| 214 | 
         
            -
                    [chatbot, version, max_new, temperature, top_p, use_chat_template, state],
         
     | 
| 215 | 
         
            -
                    [chatbot],
         
     | 
| 216 | 
         
            -
                )
         
     | 
| 217 | 
         
            -
                submit.click(user_submit, [msg, chatbot], [msg, chatbot]).then(
         
     | 
| 218 | 
         
            -
                    bot_respond,
         
     | 
| 219 | 
         
            -
                    [chatbot, version, max_new, temperature, top_p, use_chat_template, state],
         
     | 
| 220 | 
         
            -
                    [chatbot],
         
     | 
| 221 | 
         
            -
                )
         
     | 
| 222 | 
         
            -
             
     | 
| 223 | 
         
            -
                def clear_chat():
         
     | 
| 224 | 
         
            -
                    return []
         
     | 
| 225 | 
         
            -
             
     | 
| 226 | 
         
            -
                clear_btn.click(clear_chat, outputs=[chatbot])
         
     | 
| 227 | 
         | 
| 228 | 
         
             
            if __name__ == "__main__":
         
     | 
| 229 | 
         
            -
                 
     | 
| 230 | 
         
            -
                demo.launch(server_name="0.0.0.0", server_port=int(os.getenv("PORT", 7860)))
         
     | 
| 
         | 
|
| 1 | 
         
             
            import os
         
     | 
| 2 | 
         
            +
            import threading
         
     | 
| 3 | 
         
            +
            from typing import List, Tuple, Dict
         
     | 
| 4 | 
         
            +
             
     | 
| 5 | 
         
             
            import torch
         
     | 
| 6 | 
         
             
            import gradio as gr
         
     | 
| 7 | 
         
            +
            from transformers import AutoTokenizer, AutoModelForCausalLM, TextIteratorStreamer
         
     | 
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 8 | 
         
             
            from huggingface_hub import login
         
     | 
| 
         | 
|
| 9 | 
         
             
            import spaces
         
     | 
| 10 | 
         | 
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 11 | 
         
             
            MODEL_ID = "facebook/MobileLLM-Pro"
         
     | 
| 12 | 
         
            +
            SUBFOLDER = "instruct"           # use the chat template
         
     | 
| 13 | 
         
            +
            MAX_NEW_TOKENS = 256
         
     | 
| 14 | 
         
            +
            TEMPERATURE = 0.7
         
     | 
| 15 | 
         
            +
            TOP_P = 0.95
         
     | 
| 16 | 
         | 
| 17 | 
         
            +
            # --- Silent Hub auth via env/Space Secret (no UI) ---
         
     | 
| 18 | 
         
            +
            HF_TOKEN = os.getenv("HF_TOKEN") or os.getenv("HUGGINGFACEHUB_API_TOKEN") or os.getenv("HUGGINGFACE_TOKEN")
         
     | 
| 19 | 
         
             
            if HF_TOKEN:
         
     | 
| 20 | 
         
             
                try:
         
     | 
| 21 | 
         
            +
                    # No prints; stays silent if token works or fails
         
     | 
| 22 | 
         
             
                    login(token=HF_TOKEN)
         
     | 
| 23 | 
         
            +
                except Exception:
         
     | 
| 24 | 
         
            +
                    # Stay silent to avoid exposing anything to the UI/logs
         
     | 
| 25 | 
         
            +
                    pass
         
     | 
| 26 | 
         
            +
             
     | 
| 27 | 
         
            +
            # Globals so we only load once
         
     | 
| 28 | 
         
            +
            _tokenizer = None
         
     | 
| 29 | 
         
            +
            _model = None
         
     | 
| 30 | 
         
            +
            _device = None
         
     | 
| 31 | 
         
            +
             
     | 
| 32 | 
         
            +
            def _ensure_loaded():
         
     | 
| 33 | 
         
            +
                global _tokenizer, _model, _device
         
     | 
| 34 | 
         
            +
                if _tokenizer is not None and _model is not None:
         
     | 
| 35 | 
         
            +
                    return
         
     | 
| 36 | 
         
            +
                _tokenizer = AutoTokenizer.from_pretrained(
         
     | 
| 37 | 
         
            +
                    MODEL_ID, trust_remote_code=True, subfolder=SUBFOLDER
         
     | 
| 38 | 
         
             
                )
         
     | 
| 39 | 
         
            +
                _model = AutoModelForCausalLM.from_pretrained(
         
     | 
| 40 | 
         
             
                    MODEL_ID,
         
     | 
| 41 | 
         
             
                    trust_remote_code=True,
         
     | 
| 42 | 
         
            +
                    subfolder=SUBFOLDER,
         
     | 
| 43 | 
         
             
                    torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
         
     | 
| 44 | 
         
             
                    low_cpu_mem_usage=True,
         
     | 
| 45 | 
         
             
                    device_map="auto" if torch.cuda.is_available() else None,
         
     | 
| 46 | 
         
             
                )
         
     | 
| 47 | 
         
            +
                if _tokenizer.pad_token_id is None and _tokenizer.eos_token_id is not None:
         
     | 
| 48 | 
         
            +
                    _tokenizer.pad_token = _tokenizer.eos_token
         
     | 
| 49 | 
         
            +
                _model.eval()
         
     | 
| 50 | 
         
            +
                _device = next(_model.parameters()).device
         
     | 
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 51 | 
         | 
| 52 | 
         
             
            def _history_to_messages(history: List[Tuple[str, str]]) -> List[Dict[str, str]]:
         
     | 
| 53 | 
         
            +
                msgs: List[Dict[str, str]] = []
         
     | 
| 
         | 
|
| 54 | 
         
             
                for user_msg, bot_msg in history:
         
     | 
| 55 | 
         
             
                    if user_msg:
         
     | 
| 56 | 
         
            +
                        msgs.append({"role": "user", "content": user_msg})
         
     | 
| 57 | 
         
             
                    if bot_msg:
         
     | 
| 58 | 
         
            +
                        msgs.append({"role": "assistant", "content": bot_msg})
         
     | 
| 59 | 
         
            +
                return msgs
         
     | 
| 
         | 
|
| 60 | 
         | 
| 61 | 
         
             
            @spaces.GPU(duration=120)
         
     | 
| 62 | 
         
            +
            def generate_stream(message: str, history: List[Tuple[str, str]]):
         
     | 
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 63 | 
         
             
                """
         
     | 
| 64 | 
         
            +
                Minimal streaming chat function for gr.ChatInterface.
         
     | 
| 65 | 
         
            +
                Uses instruct chat template. No token UI. No extra controls.
         
     | 
| 66 | 
         
            +
                """
         
     | 
| 67 | 
         
            +
                _ensure_loaded()
         
     | 
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 68 | 
         | 
| 69 | 
         
            +
                messages = _history_to_messages(history) + [{"role": "user", "content": message}]
         
     | 
| 70 | 
         
            +
                inputs = _tokenizer.apply_chat_template(
         
     | 
| 71 | 
         
            +
                    messages,
         
     | 
| 72 | 
         
            +
                    return_tensors="pt",
         
     | 
| 73 | 
         
            +
                    add_generation_prompt=True,
         
     | 
| 74 | 
         
            +
                )
         
     | 
| 75 | 
         
            +
                input_ids = inputs["input_ids"] if isinstance(inputs, dict) else inputs
         
     | 
| 76 | 
         
            +
                input_ids = input_ids.to(_device)
         
     | 
| 77 | 
         | 
| 78 | 
         
            +
                streamer = TextIteratorStreamer(_tokenizer, skip_special_tokens=True)
         
     | 
| 79 | 
         
             
                gen_kwargs = dict(
         
     | 
| 80 | 
         
             
                    input_ids=input_ids,
         
     | 
| 81 | 
         
            +
                    max_new_tokens=MAX_NEW_TOKENS,
         
     | 
| 82 | 
         
            +
                    do_sample=TEMPERATURE > 0.0,
         
     | 
| 83 | 
         
            +
                    temperature=float(TEMPERATURE),
         
     | 
| 84 | 
         
            +
                    top_p=float(TOP_P),
         
     | 
| 85 | 
         
            +
                    pad_token_id=_tokenizer.pad_token_id,
         
     | 
| 86 | 
         
            +
                    eos_token_id=_tokenizer.eos_token_id,
         
     | 
| 87 | 
         
             
                    streamer=streamer,
         
     | 
| 88 | 
         
             
                )
         
     | 
| 89 | 
         | 
| 90 | 
         
            +
                thread = threading.Thread(target=_model.generate, kwargs=gen_kwargs)
         
     | 
| 91 | 
         
             
                thread.start()
         
     | 
| 92 | 
         | 
| 93 | 
         
            +
                output = ""
         
     | 
| 94 | 
         
             
                for new_text in streamer:
         
     | 
| 95 | 
         
            +
                    output += new_text
         
     | 
| 96 | 
         
            +
                    yield output
         
     | 
| 97 | 
         
            +
             
     | 
| 98 | 
         
            +
            demo = gr.ChatInterface(
         
     | 
| 99 | 
         
            +
                fn=generate_stream,
         
     | 
| 100 | 
         
            +
                chatbot=gr.Chatbot(height=420, label="MobileLLM-Pro"),
         
     | 
| 101 | 
         
            +
                title="MobileLLM-Pro — Chat",
         
     | 
| 102 | 
         
            +
                description="Streaming chat with facebook/MobileLLM-Pro (instruct)",
         
     | 
| 103 | 
         
            +
            )
         
     | 
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 
         | 
|
| 104 | 
         | 
| 105 | 
         
             
            if __name__ == "__main__":
         
     | 
| 106 | 
         
            +
                demo.launch(server_name="0.0.0.0", server_port=int(os.getenv("PORT", 7860)))
         
     | 
| 
         |