pysfi 0.1.10__py3-none-any.whl → 0.1.11__py3-none-any.whl

sfi/llmclient/llmclient.py

@@ -0,0 +1,592 @@
+"""LLM Chat client application.
+
+Provides a graphical interface client for streaming conversations with LLM servers.
+Supports real-time streaming response display, connection testing, and parameter adjustment.
+"""
+
+from __future__ import annotations
+
+import atexit
+import json
+import logging
+import sys
+from codecs import getincrementaldecoder
+from pathlib import Path
+from types import SimpleNamespace
+from typing import ClassVar
+from urllib.error import URLError
+from urllib.request import Request, urlopen
+
+from PySide2.QtCore import Qt, QThread, Signal
+from PySide2.QtGui import QMoveEvent, QResizeEvent, QTextCursor
+from PySide2.QtWidgets import (
+    QApplication,
+    QDoubleSpinBox,
+    QGroupBox,
+    QHBoxLayout,
+    QLabel,
+    QLineEdit,
+    QMainWindow,
+    QPushButton,
+    QSpinBox,
+    QTextEdit,
+    QVBoxLayout,
+    QWidget,
+)
+
+CONFIG_FILE = Path.home() / ".sfi" / "llmclient.json"
+logging.basicConfig(level=logging.INFO, format="%(levelname)s: %(message)s")
+logger = logging.getLogger(__name__)
+
+CONNECTION_TIMEOUT = 5
+
+
+class LLMClientConfig(SimpleNamespace):
+    """LLM Chat client configuration."""
+
+    TITLE: str = "Llama Local Model Tool"
+    WIN_SIZE: ClassVar[list[int]] = [800, 600]
+    WIN_POS: ClassVar[list[int]] = [100, 100]
+    SERVER_URL: str = "http://localhost:8080"
+    MAX_TOKENS: int = 256
+    TEMPERATURE: float = 0.7
+    TOP_P: float = 0.9
+    TOP_K: int = 40
+
+    MAX_TOKENS_RANGE: ClassVar[list[int]] = [1, 4096]
+    TEMPERATURE_RANGE: ClassVar[list[float]] = [0.0, 2.0]
+    TOP_P_RANGE: ClassVar[list[float]] = [0.0, 1.0]
+    TOP_K_RANGE: ClassVar[list[int]] = [1, 100]
+
+    def __init__(self) -> None:
+        if CONFIG_FILE.exists():
+            logger.info("Loading configuration from %s", CONFIG_FILE)
+            try:
+                self.__dict__.update(json.loads(CONFIG_FILE.read_text()))
+            except (json.JSONDecodeError, TypeError) as e:
+                logger.warning("Failed to load configuration: %s", e)
+                logger.info("Using default configuration")
+        else:
+            logger.info("Using default configuration")
+
+    def save(self) -> None:
+        """Save configuration."""
+        CONFIG_FILE.parent.mkdir(parents=True, exist_ok=True)
+        CONFIG_FILE.write_text(json.dumps(vars(self), indent=4))
+
+
+conf = LLMClientConfig()
+atexit.register(conf.save)
+
+
+class LLMWorker(QThread):
+    """LLM server communication worker thread.
+
+    Handles HTTP streaming requests in background thread to avoid blocking UI main thread.
+    Uses incremental UTF-8 decoder to correctly handle multi-byte characters across lines,
+    preventing garbled text from truncated multi-byte characters.
+
+    Signals:
+        response_received: Emitted when response content is received, carries response text
+        error_occurred: Emitted when error occurs, carries error message
+        is_finished: Emitted when request completes
+    """
+
+    response_received = Signal(str)
+    error_occurred = Signal(str)
+    is_finished = Signal()
+
+    def __init__(
+        self,
+        prompt: str,
+        server_url: str,
+        max_tokens: int,
+        temperature: float,
+        top_p: float,
+        top_k: int,
+    ) -> None:
+        """Initialize LLM worker thread.
+
+        Args:
+            prompt: User input prompt text
+            server_url: LLM server address
+            max_tokens: Maximum number of tokens to generate
+            temperature: Temperature parameter controlling randomness (0.0-2.0)
+            top_p: Nucleus sampling parameter (0.0-1.0)
+            top_k: Number of candidate tokens to retain
+        """
+        super().__init__()
+        self.prompt = prompt
+        self.server_url = server_url
+        self.max_tokens = max_tokens
+        self.temperature = temperature
+        self.top_p = top_p
+        self.top_k = top_k
+        self._is_running = True
+
+    def run(self) -> None:
+        """Execute streaming HTTP request and process response.
+
+        Receives streaming response using Server-Sent Events (SSE) format.
+        Uses incremental UTF-8 decoder to avoid garbled text from truncated multi-byte characters,
+        ensuring full-chain character encoding consistency (request/response both use UTF-8).
+        """
+        try:
+            headers = {"Content-Type": "application/json; charset=utf-8"}
+            data = {
+                "prompt": self.prompt,
+                "max_tokens": self.max_tokens,
+                "temperature": self.temperature,
+                "top_p": self.top_p,
+                "top_k": self.top_k,
+                "stream": True,
+            }
+
+            request = Request(
+                f"{self.server_url}/completion",
+                data=json.dumps(data, ensure_ascii=False).encode("utf-8"),
+                headers=headers,
+            )
+
+            with urlopen(request) as response:
+                if response.status != 200:
+                    error_text = response.read().decode("utf-8")
+                    self.error_occurred.emit(
+                        f"Error: {response.status} - {error_text}",
+                    )
+                    return
+
+                decoder = getincrementaldecoder("utf-8")(errors="replace")
+                buffer = ""
+
+                for line in response:
+                    if not self._is_running:
+                        break
+
+                    if line:
+                        try:
+                            decoded_line = decoder.decode(line, False).strip()
+                            if not decoded_line:
+                                continue
+
+                            if decoded_line.startswith("data: "):
+                                json_str = decoded_line[6:]
+                                try:
+                                    json_data = json.loads(json_str)
+                                    content = json_data.get("content", "")
+                                    if content:
+                                        buffer += content
+                                        self.response_received.emit(buffer)
+                                except json.JSONDecodeError as e:
+                                    logger.debug(
+                                        f"JSON parsing failed: {json_str}, error: {e}"
+                                    )
+                                    continue
+                        except Exception as e:
+                            logger.debug(f"Line processing failed: {e}")
+                            continue
+
+                decoder.decode(b"", True)
+
+        except URLError as e:
+            logger.error(f"Connection error: {e.reason}")
+            self.error_occurred.emit(f"Connection error: {e.reason}")
+        except Exception as e:
+            logger.error(f"Request error: {e!s}")
+            self.error_occurred.emit(f"Request error: {e!s}")
+        finally:
+            self.is_finished.emit()
+
+    def stop(self) -> None:
+        """Stop worker thread execution."""
+        self._is_running = False
+
+
+class ConnectionTestWorker(QThread):
+    """Server connection test worker thread.
+
+    Tests whether the LLM server's health check endpoint is accessible in background thread.
+
+    Signals:
+        result_ready: Emitted when test completes, passes (success_flag, message_text)
+    """
+
+    result_ready = Signal(bool, str)
+
+    def __init__(self, server_url: str) -> None:
+        """Initialize connection test worker thread.
+
+        Args:
+            server_url: LLM server address
+        """
+        super().__init__()
+        self.server_url = server_url
+
+    def run(self) -> None:
+        """Execute connection test by accessing server health check endpoint."""
+        try:
+            request = Request(f"{self.server_url}/health")
+            with urlopen(request, timeout=CONNECTION_TIMEOUT) as response:
+                if response.status == 200:
+                    self.result_ready.emit(True, "Connection successful!")
+                else:
+                    self.result_ready.emit(
+                        False, f"Connection failed: {response.status}"
+                    )
+        except URLError as e:
+            logger.error(f"Connection error: {e.reason}")
+            self.result_ready.emit(False, f"Connection error: {e.reason}")
+        except Exception as e:
+            logger.error(f"Request error: {e!s}")
+            self.result_ready.emit(False, f"Request error: {e!s}")
+
+
+class LLMChatApp(QMainWindow):
+    """LLM Chat client main window.
+
+    Provides graphical interface for interacting with LLM server, supports:
+    - Real-time streaming response display (uses incremental updates to avoid repeated rendering)
+    - Connection testing
+    - Model parameter adjustment (temperature, top-p, top-k, etc.)
+    - Conversation history tracking
+    """
+
+    def __init__(self) -> None:
+        """Initialize LLM Chat main window."""
+        super().__init__()
+        self.setWindowTitle(conf.TITLE)
+        self.setGeometry(*conf.WIN_POS, *conf.WIN_SIZE)
+
+        self.init_ui()
+
+        self.worker_thread: LLMWorker | None = None
+        self.test_thread: ConnectionTestWorker | None = None
+        self.current_ai_start_pos = -1
+
+    def init_ui(self) -> None:
+        """Initialize user interface components."""
+        main_widget = QWidget()
+        main_layout = QVBoxLayout()
+        main_widget.setLayout(main_layout)
+        self.setCentralWidget(main_widget)
+
+        # 构建界面各部分
+        main_layout.addWidget(self._create_server_group())
+        main_layout.addWidget(self._create_params_group())
+        main_layout.addWidget(self._create_chat_display())
+        main_layout.addLayout(self._create_input_layout())
+
+        self.statusBar().showMessage("Ready")
+
+    def _create_server_group(self) -> QGroupBox:
+        """Create server settings group.
+
+        Returns:
+            Group box containing server address input and test connection button
+        """
+        server_group = QGroupBox("Server Settings")
+        server_layout = QHBoxLayout()
+
+        self.server_url_input = QLineEdit(conf.SERVER_URL)
+        self.server_url_input.setPlaceholderText("Enter llama-server address")
+        self.server_url_input.textChanged.connect(self.on_config_changed)
+
+        self.test_connection_btn = QPushButton("Test Connection")
+        self.test_connection_btn.clicked.connect(self.test_connection)
+
+        server_layout.addWidget(QLabel("Server Address:"))
+        server_layout.addWidget(self.server_url_input)
+        server_layout.addWidget(self.test_connection_btn)
+        server_group.setLayout(server_layout)
+
+        return server_group
+
+    def _create_params_group(self) -> QGroupBox:
+        """Create model parameter settings group.
+
+        Returns:
+            Group box containing all model parameter adjustment controls
+        """
+        params_group = QGroupBox("Model Parameters")
+        params_layout = QHBoxLayout()
+
+        self.max_tokens_spin = QSpinBox()
+        self.max_tokens_spin.setRange(*conf.MAX_TOKENS_RANGE)
+        self.max_tokens_spin.setValue(conf.MAX_TOKENS)
+        self.max_tokens_spin.valueChanged.connect(self.on_config_changed)
+
+        self.temperature_spin = QDoubleSpinBox()
+        self.temperature_spin.setRange(*conf.TEMPERATURE_RANGE)
+        self.temperature_spin.setSingleStep(0.1)
+        self.temperature_spin.setValue(conf.TEMPERATURE)
+        self.temperature_spin.valueChanged.connect(self.on_config_changed)
+
+        self.top_p_spin = QDoubleSpinBox()
+        self.top_p_spin.setRange(*conf.TOP_P_RANGE)
+        self.top_p_spin.setSingleStep(0.05)
+        self.top_p_spin.setValue(conf.TOP_P)
+        self.top_p_spin.valueChanged.connect(self.on_config_changed)
+
+        self.top_k_spin = QSpinBox()
+        self.top_k_spin.setRange(*conf.TOP_K_RANGE)
+        self.top_k_spin.setValue(conf.TOP_K)
+        self.top_k_spin.valueChanged.connect(self.on_config_changed)
+
+        params_layout.addWidget(QLabel("Max Tokens:"))
+        params_layout.addWidget(self.max_tokens_spin)
+        params_layout.addWidget(QLabel("Temperature:"))
+        params_layout.addWidget(self.temperature_spin)
+        params_layout.addWidget(QLabel("Top P:"))
+        params_layout.addWidget(self.top_p_spin)
+        params_layout.addWidget(QLabel("Top K:"))
+        params_layout.addWidget(self.top_k_spin)
+        params_group.setLayout(params_layout)
+
+        return params_group
+
+    def _create_chat_display(self) -> QTextEdit:
+        """Create chat display area.
+
+        Returns:
+            Read-only text display widget
+        """
+        self.chat_display = QTextEdit()
+        self.chat_display.setReadOnly(True)
+        self.chat_display.setStyleSheet("font-family: monospace;")
+        return self.chat_display
+
+    def _create_input_layout(self) -> QHBoxLayout:
+        """Create user input area layout.
+
+        Returns:
+            Layout containing input box and send/stop buttons
+        """
+        input_layout = QHBoxLayout()
+
+        self.user_input = QLineEdit()
+        self.user_input.setPlaceholderText("Enter your prompt...")
+        self.user_input.returnPressed.connect(self.send_prompt)
+
+        self.send_btn = QPushButton("Send")
+        self.send_btn.clicked.connect(self.send_prompt)
+
+        self.stop_btn = QPushButton("Stop")
+        self.stop_btn.clicked.connect(self.stop_generation)
+        self.stop_btn.setEnabled(False)
+
+        input_layout.addWidget(self.user_input)
+        input_layout.addWidget(self.send_btn)
+        input_layout.addWidget(self.stop_btn)
+
+        return input_layout
+
+    def on_config_changed(self) -> None:
+        """Configuration changed handler."""
+        conf.SERVER_URL = self.server_url_input.text().strip()
+        conf.MAX_TOKENS = self.max_tokens_spin.value()
+        conf.TEMPERATURE = self.temperature_spin.value()
+        conf.TOP_P = self.top_p_spin.value()
+        conf.TOP_K = self.top_k_spin.value()
+
+    def test_connection(self) -> None:
+        """Test connection with LLM server.
+
+        Sends health check request in background thread to avoid blocking UI.
+        """
+        server_url = self.server_url_input.text().strip()
+        if not server_url:
+            self.statusBar().showMessage("Please enter server address")
+            return
+
+        if self.test_thread and self.test_thread.isRunning():
+            self.statusBar().showMessage("Test in progress...")
+            return
+
+        self.test_thread = ConnectionTestWorker(server_url)
+        self.test_thread.result_ready.connect(self.on_connection_test_result)
+        self.test_thread.finished.connect(self.on_test_thread_finished)
+
+        self.statusBar().showMessage("Testing connection...")
+        self.test_connection_btn.setEnabled(False)
+
+        self.test_thread.start()
+
+    def on_test_thread_finished(self) -> None:
+        """Handle test thread completion event and clean up resources."""
+        if self.test_thread:
+            self.test_thread.quit()
+            self.test_thread.wait()
+            self.test_thread = None
+
+    def on_connection_test_result(self, success: bool, message: str) -> None:
+        """Handle connection test result.
+
+        Args:
+            success: Whether connection succeeded
+            message: Result message text
+        """
+        self.statusBar().showMessage(message)
+        self.test_connection_btn.setEnabled(True)
+
+    def send_prompt(self) -> None:
+        """Send user input prompt to LLM server.
+
+        Creates worker thread to handle streaming response and updates UI to display conversation.
+        Resets AI reply start position to avoid repeated rendering issues.
+        """
+        if self.worker_thread and self.worker_thread.isRunning():
+            self.statusBar().showMessage("Please wait for current request to complete")
+            return
+
+        prompt = self.user_input.text().strip()
+        if not prompt:
+            self.statusBar().showMessage("Please enter prompt")
+            return
+
+        self.current_ai_start_pos = -1
+
+        server_url = self.server_url_input.text().strip()
+        max_tokens = self.max_tokens_spin.value()
+        temperature = self.temperature_spin.value()
+        top_p = self.top_p_spin.value()
+        top_k = self.top_k_spin.value()
+
+        self._display_user_input(prompt, server_url)
+        self.user_input.clear()
+
+        logger.info(f"Sending prompt: {prompt}")
+        logger.info(
+            f"Parameters: max_tokens={max_tokens}, temperature={temperature}, top_p={top_p}, top_k={top_k}"
+        )
+
+        self.worker_thread = LLMWorker(
+            prompt=prompt,
+            server_url=server_url,
+            max_tokens=max_tokens,
+            temperature=temperature,
+            top_p=top_p,
+            top_k=top_k,
+        )
+
+        self.worker_thread.response_received.connect(self.update_response)
+        self.worker_thread.error_occurred.connect(self.handle_error)
+        self.worker_thread.is_finished.connect(self.on_finished)
+
+        self.send_btn.setEnabled(False)
+        self.stop_btn.setEnabled(True)
+        self.statusBar().showMessage("Generating response...")
+
+        self.worker_thread.start()
+
+    def _display_user_input(self, prompt: str, server_url: str) -> None:
+        """Display user input and target in chat area.
+
+        Args:
+            prompt: User input prompt text
+            server_url: Target server address
+        """
+        cursor = self.chat_display.textCursor()
+        cursor.movePosition(QTextCursor.End)
+        self.chat_display.setTextCursor(cursor)
+
+        self.chat_display.setTextColor(Qt.blue)
+        self.chat_display.insertPlainText(f"You: {prompt}\n")
+
+        self.chat_display.setTextColor(Qt.darkGray)
+        self.chat_display.insertPlainText(f"[Sending to {server_url}]\n")
+
+        self.chat_display.setTextColor(Qt.black)
+        self.chat_display.insertPlainText("AI:")
+
+    def stop_generation(self) -> None:
+        """Stop current ongoing response generation."""
+        if self.worker_thread and self.worker_thread.isRunning():
+            self.worker_thread.stop()
+            self.statusBar().showMessage("Generation stopped")
+
+    def update_response(self, text: str) -> None:
+        """Update AI response content in chat display area.
+
+        Uses incremental update strategy to replace only AI reply portion of text,
+        avoiding performance issues and repeated rendering from frequent redrawing of entire text area.
+
+        Args:
+            text: Complete AI generated response text
+        """
+        cursor = self.chat_display.textCursor()
+
+        if self.current_ai_start_pos == -1:
+            cursor.movePosition(QTextCursor.End)
+            self.current_ai_start_pos = cursor.position()
+            self.chat_display.insertPlainText(f" {text}")
+        else:
+            cursor.setPosition(self.current_ai_start_pos)
+            cursor.movePosition(QTextCursor.End, QTextCursor.KeepAnchor)
+            cursor.removeSelectedText()
+            self.chat_display.setTextCursor(cursor)
+            self.chat_display.insertPlainText(f" {text}")
+
+        self.chat_display.ensureCursorVisible()
+
+    def append_to_chat(self, text: str, *, is_user: bool = False) -> None:
+        """Append text to chat area.
+
+        Args:
+            text: Text content to append
+            is_user: Whether this is a user message (for color setting)
+        """
+        cursor = self.chat_display.textCursor()
+        cursor.movePosition(QTextCursor.End)
+        self.chat_display.setTextCursor(cursor)
+
+        if is_user:
+            self.chat_display.setTextColor(Qt.blue)
+        else:
+            self.chat_display.setTextColor(Qt.black)
+
+        self.chat_display.insertPlainText(text + "\n")
+        self.chat_display.setTextColor(Qt.black)
+
+    def handle_error(self, error_msg: str) -> None:
+        """Handle error messages.
+
+        Args:
+            error_msg: Error message text
+        """
+        self.append_to_chat(f"Error: {error_msg}")
+        self.statusBar().showMessage(error_msg)
+
+    def on_finished(self) -> None:
+        """Handle response generation completion event, restore UI state and clean up resources."""
+        self.send_btn.setEnabled(True)
+        self.stop_btn.setEnabled(False)
+        self.statusBar().showMessage("Generation complete")
+        self.append_to_chat("")
+
+        if self.worker_thread:
+            self.worker_thread.quit()
+            self.worker_thread.wait()
+            self.worker_thread = None
+
+    def moveEvent(self, event: QMoveEvent) -> None:  # noqa: N802
+        """Handle window move event."""
+        top_left = self.geometry().topLeft()
+        conf.WIN_POS = [top_left.x(), top_left.y()]
+        return super().moveEvent(event)
+
+    def resizeEvent(self, event: QResizeEvent) -> None:  # noqa: N802
+        """Handle window resize event."""
+        geometry = self.geometry()
+        conf.WIN_SIZE = [geometry.width(), geometry.height()]
+        return super().resizeEvent(event)
+
+
+def main() -> None:
+    """Application entry point."""
+    app = QApplication(sys.argv)
+    app.setStyle("Fusion")
+
+    window = LLMChatApp()
+    window.show()
+
+    sys.exit(app.exec_())