""" SAM-AI Cloud Interface (ZeroGPU Space Edition) ============================================== Dual-mode Gradio application offering SAM-AI Chat and SAM-AI Code, powered exclusively by sovereign open-weight foundation model DeepSeek-R1-Distill-Qwen-7B. """ import os import sys import time import gradio as gr from pathlib import Path # Add current directory to path sys.path.insert(0, str(Path(__file__).parent.resolve())) from model_loader import engine, ModelLifecycleStatus, SAM_AI_SYSTEM_PROMPT # ZeroGPU requires model loading at module level print("[*] Starting SAM-AI ZeroGPU Space...") try: engine.load() except Exception as exc: print(f"[!] Warning: Initial model loading encountered exception: {exc}") SYSTEM_PROMPT_CHAT = SAM_AI_SYSTEM_PROMPT SYSTEM_PROMPT_CODE = ( "You are SAM-AI Code, a specialized sovereign software engineering assistant developed " "by Parallax Intelligence Lab. You assist with coding, debugging, refactoring, and code review. " "When suggesting changes, present code clearly and use standard unified diff format where appropriate. " "Never invent fake library APIs or pretend unexecuted commands ran successfully." ) try: import spaces except ImportError: class _MockSpaces: def GPU(self, duration=60): def decorator(fn): return fn return decorator spaces = _MockSpaces() @spaces.GPU(duration=20) def chat_respond(user_message, chat_history, temperature, top_p, max_tokens): """Handles multi-turn conversation streaming in SAM-AI Chat.""" if not user_message or not user_message.strip(): yield chat_history, "" return # Build history in OpenAI-style format messages = [] for turn in (chat_history or []): if turn[0]: messages.append({"role": "user", "content": turn[0]}) if turn[1]: messages.append({"role": "assistant", "content": turn[1]}) messages.append({"role": "user", "content": user_message}) # Append pending message to history (with deduplication guard) history_list = [list(turn) for turn in (chat_history or [])] if history_list and history_list[-1][0] == user_message and not history_list[-1][1]: new_history = history_list else: new_history = history_list + [[user_message, ""]] for partial_response in engine.generate_stream( messages=messages, temperature=temperature, top_p=top_p, max_new_tokens=int(max_tokens), system_prompt=SYSTEM_PROMPT_CHAT ): new_history[-1][1] = partial_response yield new_history, "" @spaces.GPU(duration=20) def code_respond(user_prompt, code_context, task_type, temperature, max_tokens): """Handles coding tasks, explanations, and diff generation in SAM-AI Code.""" if not user_prompt or not user_prompt.strip(): yield "Please provide a coding prompt or question." return full_content = f"### Task: {task_type}\n\n" if code_context and code_context.strip(): full_content += f"```\n{code_context.strip()}\n```\n\n" full_content += f"**User Request:**\n{user_prompt.strip()}" messages = [{"role": "user", "content": full_content}] for partial_response in engine.generate_stream( messages=messages, temperature=temperature, top_p=0.95, max_new_tokens=int(max_tokens), system_prompt=SYSTEM_PROMPT_CODE ): yield partial_response def get_status_markdown(): """Generates real-time status banner for header.""" diag = engine.get_diagnostics() status = diag["status"] is_loaded = diag["is_loaded"] device = diag["device"].upper() adapter_name = Path(diag["adapter_dir"]).name if diag.get("adapter_dir") else "None (Pure Base)" adapter_label = f" + {adapter_name}" if diag.get("adapter_dir") else " (Pure Foundation Model)" if is_loaded: badge = "🟢 **READY (SOVEREIGN ZERO-GPU)**" color = "green" elif status == ModelLifecycleStatus.LOADING_BASE_MODEL or status == ModelLifecycleStatus.ATTACHING_ADAPTER: badge = "🟡 **LOADING WEIGHTS...**" color = "orange" else: badge = f"🔴 **NOT READY: {status}**" color = "red" return ( f"
" f"
" f"
Model: {diag['base_model']}{adapter_label}
" f"
Status: {badge} | Device: {device}
" f"
" f"
{diag['status_detail']}
" f"
" ) # Build Gradio UI with matching warm light theme (Ivory / Terracotta / Charcoal) custom_theme = gr.themes.Soft( primary_hue="orange", secondary_hue="stone", neutral_hue="stone", ).set( body_background_fill="#FAF8F5", body_text_color="#28231F", block_background_fill="#FFFFFF", block_border_color="#E8E2DA", button_primary_background_fill="#D97745", button_primary_background_fill_hover="#C86332", button_primary_text_color="#FFFFFF", ) with gr.Blocks(theme=custom_theme, title="SAM-AI Sovereign Assistant") as demo: gr.Markdown( """ # 🧠 SAM-AI Sovereign Assistant **Self-Hosted Open-Weight Intelligence | DeepSeek-R1-Distill-Qwen-7B (Pure Sovereign Foundation Model)**\ *Running on Hugging Face Spaces ZeroGPU (NVIDIA RTX Pro 6000 Blackwell)* """ ) status_header = gr.HTML(value=get_status_markdown) with gr.Tabs() as tabs: # ===================================================================== # Tab 1: SAM-AI Chat # ===================================================================== with gr.Tab("💬 SAM-AI Chat", id="tab_chat"): chatbot = gr.Chatbot( height=520, show_copy_button=True, bubble_full_width=False, render_markdown=True, ) with gr.Row(): chat_input = gr.Textbox( placeholder="Ask a question, mathematical problem, or reasoning challenge...", show_label=False, scale=8, lines=2, ) with gr.Column(scale=2, min_width=120): send_btn = gr.Button("Send", variant="primary") clear_btn = gr.Button("Clear Chat") with gr.Accordion("⚙️ Inference Parameters", open=False): with gr.Row(): chat_temp = gr.Slider( minimum=0.0, maximum=1.0, value=0.6, step=0.05, label="Temperature (Creativity / Randomness)" ) chat_top_p = gr.Slider( minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-P (Nucleus Sampling)" ) chat_max_tokens = gr.Slider( minimum=256, maximum=4096, value=2048, step=128, label="Max New Tokens" ) # Event wiring chat_msg_event = chat_input.submit( chat_respond, inputs=[chat_input, chatbot, chat_temp, chat_top_p, chat_max_tokens], outputs=[chatbot, chat_input], api_name="chat_respond", ) send_btn.click( chat_respond, inputs=[chat_input, chatbot, chat_temp, chat_top_p, chat_max_tokens], outputs=[chatbot, chat_input], api_name="chat_respond", ) clear_btn.click(lambda: None, None, chatbot, queue=False) # ===================================================================== # Tab 2: SAM-AI Code # ===================================================================== with gr.Tab("💻 SAM-AI Code", id="tab_code"): with gr.Row(): with gr.Column(scale=5): code_task = gr.Radio( choices=["Generate Code", "Debug & Fix", "Refactor & Optimize", "Generate Unified Diff"], value="Generate Code", label="Coding Task Type" ) code_prompt = gr.Textbox( label="Coding Instructions / Problem Description", placeholder="e.g. Implement a thread-safe LRU cache with eviction telemetry in Python...", lines=4, ) code_context = gr.Code( label="Existing Code / Context (Optional)", language="python", lines=12, ) run_code_btn = gr.Button("Generate Solution", variant="primary") with gr.Column(scale=5): code_output = gr.Markdown( label="SAM-AI Code Response", value="*Click 'Generate Solution' to run SAM-AI Code on the assigned ZeroGPU.*" ) with gr.Accordion("⚙️ Code Generation Parameters", open=False): with gr.Row(): code_temp = gr.Slider( minimum=0.0, maximum=1.0, value=0.2, step=0.05, label="Temperature (Lower = More Deterministic)" ) code_max_tokens = gr.Slider( minimum=512, maximum=4096, value=2048, step=128, label="Max Tokens" ) run_code_btn.click( code_respond, inputs=[code_prompt, code_context, code_task, code_temp, code_max_tokens], outputs=[code_output], api_name="code_respond", ) # ===================================================================== # Tab 3: Diagnostics & Sovereign Architecture # ===================================================================== with gr.Tab("🛡️ Sovereign Diagnostics", id="tab_diag"): refresh_diag_btn = gr.Button("Refresh Telemetry") diag_json = gr.JSON(value=engine.get_diagnostics, label="Engine Diagnostic State") refresh_diag_btn.click(engine.get_diagnostics, outputs=[diag_json]) gr.Markdown( """ --- **Parallax Intelligence Lab | SAM-AI Project** *Zero commercial third-party inference APIs. Pure open-weight execution.* """ ) # Configure Gradio queue for ZeroGPU concurrency management demo.queue( max_size=20, default_concurrency_limit=1, ) if __name__ == "__main__": demo.launch( server_name="0.0.0.0", server_port=7860, show_error=True )