lemonade-sdk 8.1.10__py3-none-any.whl → 8.1.12__py3-none-any.whl

lemonade/tools/server/utils/macos_tray.py

@@ -0,0 +1,226 @@
+import platform
+import subprocess
+from typing import Callable, Optional
+
+# Check if we're on macOS and import accordingly
+if platform.system() == "Darwin":
+    try:
+        import rumps
+
+        RUMPS_AVAILABLE = True
+    except ImportError:
+        RUMPS_AVAILABLE = False
+        print("Warning: rumps not available. Install with: pip install rumps")
+else:
+    RUMPS_AVAILABLE = False
+
+
+class MenuItem:
+    """
+    Cross-platform menu item representation.
+    """
+
+    def __init__(
+        self,
+        text: str,
+        callback: Optional[Callable] = None,
+        enabled: bool = True,
+        submenu=None,
+        checked: bool = False,
+    ):
+        self.text = text
+        self.callback = callback
+        self.enabled = enabled
+        self.submenu = submenu
+        self.checked = checked
+
+
+class Menu:
+    """
+    Cross-platform menu representation.
+    """
+
+    SEPARATOR = "SEPARATOR"
+
+    def __init__(self, *items):
+        self.items = list(items)
+
+
+class MacOSSystemTray:
+    """
+    macOS-specific system tray implementation using rumps.
+    """
+
+    def __init__(self, app_name: str, icon_path: str):
+        self._check_rumps_availability()
+
+        self.app_name = app_name
+        self.icon_path = icon_path
+        self.app = None
+        self.menu_callbacks = {}
+        self._menu_update_timer = None
+
+    def _check_rumps_availability(self):
+        """Check if rumps is available and raise error if not."""
+        if not RUMPS_AVAILABLE:
+            raise ImportError("rumps library is required for macOS tray support")
+
+    def create_menu(self):
+        """
+        Create the context menu based on current state. Override in subclass.
+        """
+        return Menu(MenuItem("Exit", self.exit_app))
+
+    def build_rumps_menu(self, menu_items):
+        """
+        Convert our menu structure to rumps menu items.
+        """
+        rumps_items = []
+
+        for item in menu_items:
+            if item == Menu.SEPARATOR:
+                rumps_items.append(rumps.separator)
+            elif isinstance(item, MenuItem):
+                if item.submenu:
+                    # Create submenu
+                    submenu_items = self.build_rumps_menu(item.submenu.items)
+                    submenu = rumps.MenuItem(item.text)
+                    for sub_item in submenu_items:
+                        submenu.add(sub_item)
+                    rumps_items.append(submenu)
+                else:
+                    # Create regular menu item
+                    menu_item = rumps.MenuItem(
+                        item.text,
+                        callback=(
+                            self._create_callback_wrapper(item)
+                            if item.callback
+                            else None
+                        ),
+                    )
+
+                    # Set enabled state
+                    if not item.enabled:
+                        menu_item.set_callback(None)
+
+                    # Set checked state
+                    if item.checked:
+                        menu_item.state = 1
+                    else:
+                        menu_item.state = 0
+
+                    rumps_items.append(menu_item)
+
+        return rumps_items
+
+    def _create_callback_wrapper(self, item):
+        """Create a callback wrapper that matches our interface."""
+
+        def wrapper(sender):  # pylint: disable=unused-argument
+            if item.callback:
+                item.callback(None, item)
+
+        return wrapper
+
+    def show_balloon_notification(
+        self, title, message, timeout=5000
+    ):  # pylint: disable=unused-argument
+        """
+        Show a notification on macOS using the Notification Center.
+        Falls back to console output if AppleScript fails.
+        """
+        try:
+            # Escape quotes in message and title for AppleScript
+            escaped_title = title.replace('"', '\\"')
+            escaped_message = message.replace('"', '\\"')
+            escaped_app_name = self.app_name.replace('"', '\\"')
+
+            # Use AppleScript to show notification
+            script = (
+                f'display notification "{escaped_message}" '
+                f'with title "{escaped_title}" subtitle "{escaped_app_name}"'
+            )
+            subprocess.run(
+                ["osascript", "-e", script], check=True, capture_output=True, text=True
+            )
+        except FileNotFoundError:
+            # osascript not available, fallback to console
+            print(f"[{self.app_name}] {title}: {message}")
+        except subprocess.CalledProcessError as e:
+            # AppleScript failed, fallback to console
+            print(f"[{self.app_name}] {title}: {message}")
+            print(f"Warning: Failed to show notification via AppleScript: {e}")
+        except Exception as e:  # pylint: disable=broad-exception-caught
+            # Any other error, fallback to console
+            print(f"[{self.app_name}] {title}: {message}")
+            print(f"Warning: Failed to show notification: {e}")
+
+    def exit_app(self, _, __):
+        """Exit the application."""
+        if self.app:
+            rumps.quit_application()
+
+    def run(self):
+        """
+        Run the tray application.
+        """
+        self._check_rumps_availability()
+
+        try:
+            # Create the rumps app
+            self.app = rumps.App(self.app_name, icon=self.icon_path, quit_button=None)
+
+            # Build the initial menu
+            self.refresh_menu()
+
+            # Set up a timer to refresh menu periodically (every 3 seconds)
+            # This provides a good balance between responsiveness and performance
+            self._setup_menu_refresh_timer()
+
+            # Call the on_ready hook if available (for compatibility with tray.py)
+            if hasattr(self, "on_ready") and callable(getattr(self, "on_ready", None)):
+                getattr(self, "on_ready")()
+
+            # Start the app
+            self.app.run()
+        except Exception as e:
+            raise RuntimeError(f"Failed to start macOS tray application: {e}") from e
+
+    def refresh_menu(self):
+        """
+        Refresh the menu by rebuilding it with current state.
+        """
+        if not self.app:
+            return
+
+        # Clear existing menu
+        self.app.menu.clear()
+
+        # Build fresh menu with current state
+        menu = self.create_menu()
+        menu_items = self.build_rumps_menu(menu.items)
+
+        # Add updated menu items
+        for item in menu_items:
+            self.app.menu.add(item)
+
+    def _setup_menu_refresh_timer(self):
+        """
+        Set up a timer to periodically refresh the menu.
+        """
+        if not self.app:
+            return
+
+        # Create a timer that refreshes the menu every 3 seconds
+        @rumps.timer(3)
+        def refresh_menu_timer(sender):  # pylint: disable=unused-argument
+            self.refresh_menu()
+
+        # Store reference to prevent garbage collection
+        self._menu_update_timer = refresh_menu_timer
+
+    def update_menu(self):
+        """
+        Update the menu by rebuilding it.
+        """
+        self.refresh_menu()

lemonade/tools/server/utils/{system_tray.py → windows_tray.py}

@@ -25,12 +25,14 @@ class MenuItem:
         enabled: bool = True,
         submenu=None,
         bitmap_path: Optional[str] = None,
+        checked: bool = False,
     ):
         self.text = text
         self.callback = callback
         self.enabled = enabled
         self.submenu = submenu
         self.bitmap_path = bitmap_path
+        self.checked = checked
         self.id = None  # Will be set when menu is created
         self.bitmap_handle = None
 
@@ -358,9 +360,20 @@ class SystemTray:
         self.register_window_class()
         self.create_window()
 
+        # Set up Windows console control handler for CTRL+C
+        self.console_handler = self.setup_console_control_handler(self.logger)
+
         # Add tray icon
         self.add_tray_icon()
 
+        # Notify subclasses that the tray is ready (hwnd and icon created)
+        # Allows showing initial notifications after initialization
+        try:
+            if hasattr(self, "on_ready") and callable(getattr(self, "on_ready")):
+                self.on_ready()
+        except Exception:
+            pass
+
         # Run the message loop in the main thread
         self.message_loop()
 

lemonade/tools/server/webapp.py

@@ -1,5 +1,6 @@
 from pathlib import Path
 import json
+import platform
 from fastapi.responses import HTMLResponse
 from lemonade_server.model_manager import ModelManager
 
@@ -14,10 +15,11 @@ def get_webapp_html(port=8000):
     # Use shared filter function from model_manager.py
     filtered_models = ModelManager().filter_models_by_backend(server_models)
 
-    # Pass filtered server_models to JS
+    # Pass filtered server_models and platform info to JS
     server_models_js = (
         f"<script>window.SERVER_MODELS = {json.dumps(filtered_models)};</script>"
     )
+    platform_js = f"<script>window.PLATFORM = '{platform.system()}';</script>"
 
     # Load HTML template
     template_path = Path(__file__).parent / "static" / "webapp.html"
@@ -27,5 +29,6 @@ def get_webapp_html(port=8000):
     # Replace template variables
     html_content = html_template.replace("{{SERVER_PORT}}", str(port))
     html_content = html_content.replace("{{SERVER_MODELS_JS}}", server_models_js)
+    html_content = html_content.replace("{{PLATFORM_JS}}", platform_js)
 
     return HTMLResponse(content=html_content)

lemonade/tools/server/wrapped_server.py

@@ -58,8 +58,22 @@ class WrappedServerTelemetry(ABC):
         telemetry = [
             ["Input tokens", self.input_tokens],
             ["Output tokens", self.output_tokens],
-            ["TTFT (s)", f"{self.time_to_first_token:.2f}"],
-            ["TPS", f"{self.tokens_per_second:.2f}"],
+            [
+                "TTFT (s)",
+                (
+                    f"{self.time_to_first_token:.2f}"
+                    if self.time_to_first_token is not None
+                    else "N/A"
+                ),
+            ],
+            [
+                "TPS",
+                (
+                    f"{self.tokens_per_second:.2f}"
+                    if self.tokens_per_second is not None
+                    else "N/A"
+                ),
+            ],
         ]
 
         table = tabulate(
@@ -83,7 +97,7 @@ class WrappedServer(ABC):
         self.telemetry: WrappedServerTelemetry = telemetry
         self.log_thread_exception = None
 
-    def choose_port(self):
+    def _choose_port(self):
         """
         Users probably don't care what port we start the wrapped server on, so let's
         search for an empty port
@@ -318,18 +332,44 @@ class WrappedServer(ABC):
         if chat_completion_request.stream:
 
             def event_stream():
-                try:
-                    # Enable streaming
-                    # pylint: disable=missing-kwoa
-                    for chunk in client.chat.completions.create(**openai_client_params):
-                        yield f"data: {chunk.model_dump_json()}\n\n"
-                    yield "data: [DONE]\n\n"
-
-                    # Show telemetry after completion
-                    self.telemetry.show_telemetry()
+                # Ensure streaming is enabled in params
+                stream_params = dict(openai_client_params)
+                stream_params["stream"] = True
 
-                except Exception as e:  # pylint: disable=broad-exception-caught
-                    yield f'data: {{"error": "{str(e)}"}}\n\n'
+                # Use streaming context so we can explicitly close on cancellation
+                with client.chat.completions.with_streaming_response.create(
+                    # pylint: disable=missing-kwoa
+                    **stream_params,
+                ) as response:
+                    try:
+                        for line in response.iter_lines():
+                            # Preserve SSE event boundaries: blank line separates events
+                            if line == b"" or line == "":
+                                yield "\n"
+                                continue
+                            if isinstance(line, bytes):
+                                try:
+                                    line = line.decode("utf-8", errors="ignore")
+                                except (UnicodeDecodeError, LookupError):
+                                    # Skip lines that fail decoding due to encoding issues
+                                    continue
+                            # Forward SSE lines as-is
+                            if not line.endswith("\n"):
+                                line += "\n"
+                            yield line
+
+                        # Show telemetry after completion
+                        self.telemetry.show_telemetry()
+
+                    except GeneratorExit:
+                        # Client disconnected/cancelled; close upstream stream and stop
+                        try:
+                            response.close()
+                        except Exception:  # pylint: disable=broad-exception-caught
+                            pass
+                        raise
+                    except Exception as e:  # pylint: disable=broad-exception-caught
+                        yield f'data: {{"error": "{str(e)}"}}\n\n'
 
             return StreamingResponse(
                 event_stream(),
@@ -387,18 +427,44 @@ class WrappedServer(ABC):
         if completion_request.stream:
 
             def event_stream():
-                try:
-                    # Enable streaming
-                    # pylint: disable=missing-kwoa
-                    for chunk in client.completions.create(**openai_client_params):
-                        yield f"data: {chunk.model_dump_json()}\n\n"
-                    yield "data: [DONE]\n\n"
-
-                    # Show telemetry after completion
-                    self.telemetry.show_telemetry()
+                # Ensure streaming is enabled in params
+                stream_params = dict(openai_client_params)
+                stream_params["stream"] = True
 
-                except Exception as e:  # pylint: disable=broad-exception-caught
-                    yield f'data: {{"error": "{str(e)}"}}\n\n'
+                # Use streaming context so we can explicitly close on cancellation
+                with client.completions.with_streaming_response.create(
+                    # pylint: disable=missing-kwoa
+                    **stream_params,
+                ) as response:
+                    try:
+                        for line in response.iter_lines():
+                            # Preserve SSE event boundaries: blank line separates events
+                            if line == b"" or line == "":
+                                yield "\n"
+                                continue
+                            if isinstance(line, bytes):
+                                try:
+                                    line = line.decode("utf-8", errors="ignore")
+                                except (UnicodeDecodeError, LookupError):
+                                    # Skip lines that fail decoding due to encoding issues
+                                    continue
+                            # Forward SSE lines as-is
+                            if not line.endswith("\n"):
+                                line += "\n"
+                            yield line
+
+                        # Show telemetry after completion
+                        self.telemetry.show_telemetry()
+
+                    except GeneratorExit:
+                        # Client disconnected/cancelled; close upstream stream and stop
+                        try:
+                            response.close()
+                        except Exception:  # pylint: disable=broad-exception-caught
+                            pass
+                        raise
+                    except Exception as e:  # pylint: disable=broad-exception-caught
+                        yield f'data: {{"error": "{str(e)}"}}\n\n'
 
             return StreamingResponse(
                 event_stream(),

lemonade/version.py

@@ -1 +1 @@
-__version__ = "8.1.10"
+__version__ = "8.1.12"

lemonade_install/install.py

@@ -447,6 +447,12 @@ class Install:
             choices=["rocm", "vulkan"],
         )
 
+        parser.add_argument(
+            "--flm",
+            action="store_true",
+            help="Install FLM (FastFlowLM) for running local language models",
+        )
+
         parser.add_argument(
             "--override",
             action="store_true",
@@ -727,19 +733,33 @@ class Install:
 
         install_llamacpp(backend)
 
+    @staticmethod
+    def _install_flm():
+        """
+        Install FLM (FastFlowLM) for running local language models.
+        """
+
+        # Check if the processor is supported before proceeding
+        check_ryzen_ai_processor()
+
+        from lemonade.tools.flm.utils import install_flm
+
+        install_flm()
+
     def run(
         self,
         ryzenai: Optional[str] = None,
         build_model: Optional[str] = None,
         llamacpp: Optional[str] = None,
+        flm: Optional[bool] = None,
         yes: bool = False,
         token: Optional[str] = None,
         override: bool = False,
     ):
-        if ryzenai is None and llamacpp is None:
+        if ryzenai is None and llamacpp is None and flm is None:
             raise ValueError(
                 "You must select something to install, "
-                "for example `--ryzenai` or `--llamacpp`"
+                "for example `--llamacpp`, `--flm`, or `--ryzenai`"
             )
 
         if ryzenai is not None:
@@ -748,6 +768,9 @@ class Install:
         if llamacpp is not None:
             self._install_llamacpp(llamacpp)
 
+        if flm:
+            self._install_flm()
+
 
 def main():
     installer = Install()

{lemonade_sdk-8.1.10.dist-info → lemonade_sdk-8.1.12.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: lemonade-sdk
-Version: 8.1.10
+Version: 8.1.12
 Summary: Lemonade SDK: Your LLM Aide for Validation and Deployment
 Author-email: lemonade@amd.com
 Requires-Python: >=3.10, <3.14
@@ -29,6 +29,7 @@ Requires-Dist: tabulate
 Requires-Dist: sentencepiece
 Requires-Dist: huggingface-hub[hf_xet]==0.33.0
 Requires-Dist: python-dotenv
+Requires-Dist: rumps>=0.4.0; sys_platform == "darwin"
 Provides-Extra: oga-ryzenai
 Requires-Dist: onnxruntime-genai-directml-ryzenai==0.7.0.2.1; extra == "oga-ryzenai"
 Requires-Dist: protobuf>=6.30.1; extra == "oga-ryzenai"
@@ -65,6 +66,8 @@ Dynamic: summary
     <img src="https://img.shields.io/badge/Windows-11-0078D6?logo=windows&logoColor=white" alt="Windows 11" /></a>
   <a href="https://lemonade-server.ai/#linux" title="Ubuntu 24.04 & 25.04 Supported">
     <img src="https://img.shields.io/badge/Ubuntu-24.04%20%7C%2025.04-E95420?logo=ubuntu&logoColor=white" alt="Ubuntu 24.04 | 25.04" /></a>
+  <a href="https://lemonade-server.ai/" title="macOS 14+ with Apple Silicon">
+    <img src="https://img.shields.io/badge/macOS-14%2B-000000?logo=apple&logoColor=white" alt="macOS 14+" /></a>
   <a href="docs/README.md#installation" title="Check out our instructions">
     <img src="https://img.shields.io/badge/Python-3.10--3.13-blue?logo=python&logoColor=white" alt="Made with Python" /></a>
   <a href="https://github.com/lemonade-sdk/lemonade/blob/main/docs/contribute.md" title="Contribution Guide">
@@ -152,11 +155,11 @@ You can also import custom GGUF and ONNX models from Hugging Face by using our [
 
 Lemonade supports the following configurations, while also making it easy to switch between them at runtime. Find more information about it [here](./docs/README.md#software-and-hardware-overview).
 
-| Hardware | Engine: OGA | Engine: llamacpp | Engine: HF | Windows | Linux |
-|----------|-------------|------------------|------------|---------|-------|
-| **🧠 CPU** | All platforms | All platforms | All platforms | ✅ | ✅ |
-| **🎮 GPU** | — | Vulkan: All platforms<br>ROCm: Selected AMD platforms* | — | ✅ | ✅ |
-| **🤖 NPU** | AMD Ryzen™ AI 300 series | — | — | ✅ | — |
+| Hardware | Engine: OGA | Engine: llamacpp | Engine: FLM | Windows | Linux | macOS |
+|----------|-------------|------------------|------------|---------|-------|-------|
+| **🧠 CPU** | All platforms | All platforms | - | ✅ | ✅ | ✅ |
+| **🎮 GPU** | — | Vulkan: All platforms<br>ROCm: Selected AMD platforms*<br>Metal: Apple Silicon | — | ✅ | ✅ | ✅ |
+| **🤖 NPU** | AMD Ryzen™ AI 300 series | — | Ryzen™ AI 300 series | ✅ | — | — |
 
 <details>
 <summary><small><i>* See supported AMD ROCm platforms</i></small></summary>
@@ -261,6 +264,7 @@ This project is:
   - [OnnxRuntime GenAI](https://github.com/microsoft/onnxruntime-genai)
   - [Hugging Face Hub](https://github.com/huggingface/huggingface_hub)
   - [OpenAI API](https://github.com/openai/openai-python)
+  - [IRON/MLIR-AIE](https://github.com/Xilinx/mlir-aie)
   - and more...
 - Accelerated by mentorship from the OCV Catalyst program.
 - Licensed under the [Apache 2.0 License](https://github.com/lemonade-sdk/lemonade/blob/main/LICENSE).

{lemonade_sdk-8.1.10.dist-info → lemonade_sdk-8.1.12.dist-info}/RECORD

@@ -1,10 +1,10 @@
 lemonade/__init__.py,sha256=W1Qk7r0rnQqFhPNHp6BIBT_q-OH3s-8Q_POoVfAmKW0,117
 lemonade/api.py,sha256=Oc4yBA3LZg8FrTsbuDq1p9-XE74pqNnIEUhXyKa7qg8,5786
-lemonade/cache.py,sha256=5iZbk273TiTMqK_vdzPOPYTo6VsWW2gNByOISA9zi1w,3002
+lemonade/cache.py,sha256=fUjtHYkRdHTULjNMrDNOFDGW_QMVUg54ZE1NukBP2oM,3314
 lemonade/cli.py,sha256=qU5bW7RQAUKNSpvrhVyzn68NMxyi-336Ke_JU4bsv1Q,5708
 lemonade/sequence.py,sha256=KSH7BPsiyDKsOsg_ziQKEGsDwMmuO_YbgPRBxkZd0pw,13267
 lemonade/state.py,sha256=sdSezla7Cd7KYL90xY3p9kcNV4ndSyN6UvNLOr3vBMA,5261
-lemonade/version.py,sha256=BSj3P5N0EwzL0-jahgtLzMLFGfZHooaYa76BeUiW2wc,23
+lemonade/version.py,sha256=IzX_CD-LCNW8UO9z0HHoCrMMPsTBn-2q8Vqxoe3H6hQ,23
 lemonade/common/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 lemonade/common/build.py,sha256=zTb0m1-kuUx6zw5QHp2SNnVuN6jOTMQ2FCdj9iH374U,6140
 lemonade/common/cli_helpers.py,sha256=hjBfXrTtFl8gmCFlL-ksviXR0mOcdPtTWVNKoEp3PG4,4993
@@ -13,7 +13,7 @@ lemonade/common/filesystem.py,sha256=QV3cHhKNu-7W2rr8wZ4JQfD2rP_5T2Js7jiDQBYWHVQ
 lemonade/common/inference_engines.py,sha256=3bUGQe9wtfTiwt8kvI_ry077uyc9lid2G1fJX95kN1A,12969
 lemonade/common/network.py,sha256=qXpUjDYQEYM_gH3JwTtU-pu_yCKcaa1IeohJRPy91-A,2903
 lemonade/common/printing.py,sha256=GFFzrXIineIOMa9yu0lo5sL4j6A5BBg_T9aUCdP-juw,3229
-lemonade/common/status.py,sha256=xSOZN508cdRtrs1HVyr9zmASYg69EsZBLSs0lroLoCM,16519
+lemonade/common/status.py,sha256=dxAahChPGVmfT91DJW949Xjgm9r5E-Y9KOLPEw7BMh8,16562
 lemonade/common/system_info.py,sha256=Msa0pCSj3ZN3nerjY8wdqjjJLg6GPhbWf2htSNcFIHc,49607
 lemonade/common/test_helpers.py,sha256=Gwk-pa_6xYAo2oro-2EJNfuouAfw8k_brCbcMC-E-r0,758
 lemonade/profilers/__init__.py,sha256=JKVonvJ4XZ9_6sKXPWsiMLQCNyzQOxhQw5BEHR1qOfU,31
@@ -24,54 +24,59 @@ lemonade/profilers/profiler.py,sha256=Y5FSbc386bMlTVbqCuya9pYrso5aTthxahR1V_ZKQ9
 lemonade/tools/__init__.py,sha256=_6xRc-FHxmujoLjLjWtpYrWYEXtCSneSy-5ya01kyPk,53
 lemonade/tools/accuracy.py,sha256=9HCmczDngkBUuUrt49d2CkRo4J0qyWoFYs5cj20bGkg,11714
 lemonade/tools/adapter.py,sha256=Ex63Y1SPCOSV4M_QtzEn3YVd39d3yew0lpmEFgp8aH4,3169
-lemonade/tools/bench.py,sha256=aN5LMA_EH6-ZhAH3Gf26JYL7s0eKpUd3j-bReRhzvEY,10016
+lemonade/tools/bench.py,sha256=AZP1UGn2dZD61XshxO67tMUgOKTLEBqO3C872lKLzNQ,10903
 lemonade/tools/humaneval.py,sha256=JbxuoOzvR4iyxZv4R6MI7a3gUt5ef_Jj6Ie-9VP2wzY,9531
 lemonade/tools/management_tools.py,sha256=HQBcr7LYuMqVRYQtvnkNpfutBTA7lblszyoAjjVGu1Y,10201
 lemonade/tools/mmlu.py,sha256=c2QaIMDzjqxCvgHlMXmy_dP1sAFkwkDxL7RO2nogI6s,11071
 lemonade/tools/perplexity.py,sha256=eiaTZ3yhqF2pfwOffVbKKJLwjSri7Im2pC-tBJr7LLU,5638
 lemonade/tools/prompt.py,sha256=PyLksp1k8jsZsU7XBRK61k1DUHhbdLa20h-AP8Noh3w,9011
 lemonade/tools/tool.py,sha256=UsxVYukfm_iM3BfeGYPZxQlTK5UfDfDOl3RIyLr8A1Y,13256
+lemonade/tools/flm/__init__.py,sha256=NQ4CEzJZGS_VvxPMlfrK4Dcx48bQSoUR4iG8e7yZjas,46
+lemonade/tools/flm/utils.py,sha256=HOfw2ohjNE69S6WL3XVgVI0MfO4hQ50U32Cxb8vnjYI,8126
 lemonade/tools/huggingface/bench.py,sha256=-mTfldCtquL4mspq8ykVwDc9Mut5Ecv_jHJnSb0CYGE,6734
 lemonade/tools/huggingface/load.py,sha256=KsSGOBBD-tNEIfYC8mCWV_jpnkjHMhN3juVmC1Ln4uQ,7745
 lemonade/tools/huggingface/utils.py,sha256=j1S-IgjDsznUIVwkHSqqChmFyqIx9f3WcEelzohWwvU,13955
-lemonade/tools/llamacpp/bench.py,sha256=1fkE02ecg-jRk92i5dTAXz6re14WH8bd-Z9l-m3lbDA,4844
-lemonade/tools/llamacpp/load.py,sha256=DFCvQN548Ch9H8U_rHOiYviinzw6vixb5-V7xLj7XE4,6499
-lemonade/tools/llamacpp/utils.py,sha256=WEjdGmVxl30rt0a62MNo_X8ndFQ2SIrqtA2uF3klE6g,33090
+lemonade/tools/llamacpp/bench.py,sha256=ubifZAdZatubfwuxgIY5IBzjW29lP-RwKaxhIZsXubQ,8533
+lemonade/tools/llamacpp/load.py,sha256=yVKiKdtIKG9DOcLLc1MECQys7TiBzal-bPbWrlqQbn0,7238
+lemonade/tools/llamacpp/utils.py,sha256=yQhAtE7z6F9adE1aGjGn0RVQBqiA-4RSJSGKdtFNXO4,40804
 lemonade/tools/oga/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-lemonade/tools/oga/bench.py,sha256=PJXv4UchcS2YPwijNzef8DY4DSAKYxIYY1ycHuH3T34,5005
+lemonade/tools/oga/bench.py,sha256=g8l_ssvi_FC5Xf9tIe-c0XpcPR9TIVALh0eza5scMjI,4156
 lemonade/tools/oga/load.py,sha256=x-A-nhoni-WyDpVCLcWRAMfs5ouac9MJzxT-rsnLPw8,34226
 lemonade/tools/oga/utils.py,sha256=F8UVLKlfYcLa2SUqlehar8-jaX2Aw4u58DjHNNvLdOA,17675
 lemonade/tools/report/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 lemonade/tools/report/llm_report.py,sha256=bVHhwCINA-Ok2EdSwAsLubsc83N3KWOVuwTguw7jDcE,6676
-lemonade/tools/report/table.py,sha256=Kv_Epd8a6KIrdzSC2EgIl6uTKw7E5eMq10Tg16O0WxM,27996
+lemonade/tools/report/table.py,sha256=_AzknkTXI9xTPem-3N16FPRhkKnLvhhsfnm_P3GN2sw,28241
 lemonade/tools/server/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-lemonade/tools/server/llamacpp.py,sha256=8HdTkrU2ht8L1ldXqkfYfYhXiA8TvySuaslinAMqr-c,9002
-lemonade/tools/server/serve.py,sha256=W6wugCEaRNsDpWKIcsACrKZRFEwt7H1qWr6kG07WCek,62739
+lemonade/tools/server/flm.py,sha256=NVpjFzcmX6YzW6wCWyMEvQEe_QpRK-Jlcaajxy3kWuo,4025
+lemonade/tools/server/llamacpp.py,sha256=dSab9hR_CcctmU8HeQYQ7U0XCbAQNd_QGZUt6q7cxHA,9952
+lemonade/tools/server/serve.py,sha256=rFv7N9ZW1gcRbw8ZXiXCiAOzPfJT5zbXt-b0Dljggmo,68780
 lemonade/tools/server/tool_calls.py,sha256=xrAlQwKG-nv2xLlf8f9CDSaUbyMn8ZtHkds9iZLG9K8,5230
-lemonade/tools/server/tray.py,sha256=2PQxoEWH-zzUGlveTW4heij4UC9SzxXcFlGs0JtFjF0,22226
-lemonade/tools/server/webapp.py,sha256=8Das5yXOaSBLZmSZ_eddJajQFxBhvl5D6GI_hHlGbE0,1040
-lemonade/tools/server/wrapped_server.py,sha256=DlzsGUwLQzjOFRfTTxfnhvmM_9lvAki96jWIAz7Czds,16713
+lemonade/tools/server/tray.py,sha256=EFnSc2Ra4owiHVz6ykoMhxi2fYqZAK1g21AynAYBiyk,24426
+lemonade/tools/server/webapp.py,sha256=GGSVIzN19C2ZaadOEPBg_D7Lt0PuF339NuWwjMPfZu8,1225
+lemonade/tools/server/wrapped_server.py,sha256=uh7ifrRX1Hx0IuRwZRCGPyQOukitE7kKQipCCz0bSGA,19844
 lemonade/tools/server/static/favicon.ico,sha256=hMmP9qGJNeZ0mFS86JIqPbZstXMZn0Z76_HfHQpREAU,126745
-lemonade/tools/server/static/styles.css,sha256=SYEK4rC-MdpkTj31gxNl9Kb3hCNd1Fpq-EGnRMTjVe8,45362
-lemonade/tools/server/static/webapp.html,sha256=j7A8SOwbY_GfOSkOMV3JvXhOKY1iG70JYYuA3WdoWSQ,17856
-lemonade/tools/server/static/js/chat.js,sha256=XpQSIn1TUra26tu2CtDyOayhXAbUEqzBK0oGtkCAu-s,39162
+lemonade/tools/server/static/logs.html,sha256=5TCpjL9XanRH5uiTEB0bJVuM-fhU8OaQrDrkZENkOMc,1543
+lemonade/tools/server/static/styles.css,sha256=GYJgRtlZSgz3pShDeuatu-J9TpVwGgOjOcIWwNnWYck,49100
+lemonade/tools/server/static/webapp.html,sha256=QQRMMMf8fbtJfGZYqBaRVd0-bENmdPfkVo8sdc4092Q,19151
+lemonade/tools/server/static/js/chat.js,sha256=jxyMyu4MfvI2YmsMbJQ8ZwDNBnLzu2nbjm-qLfgWSNI,42182
 lemonade/tools/server/static/js/model-settings.js,sha256=JXHeG7xVrRU181Hj7CZflERAi1Z6t-qwYFR4aH5nf5I,5820
-lemonade/tools/server/static/js/models.js,sha256=7bCJbvS8FWpj6f1ZOwM8pt6UOQueuiOGLG79qrs-C-A,35872
-lemonade/tools/server/static/js/shared.js,sha256=mD03xqyMH1iQwH4pOq4IpDDaAX0z7YZY71gD8gufRAg,17487
+lemonade/tools/server/static/js/models.js,sha256=es3LwrU49UtoC59e-AusUpdSXQnIRNsZvw8f05keAM0,37856
+lemonade/tools/server/static/js/shared.js,sha256=4l-P2PHvgN6ALw1FK4xz4_wLRl97gZGS-G3veW6s9oU,17741
+lemonade/tools/server/utils/macos_tray.py,sha256=xwHW44ZN5hDVlJcwIpHHfqn4VRXWxXHuDACaT-ZqdO8,7095
 lemonade/tools/server/utils/port.py,sha256=J7-g-Aqygb50jNoHLhhRfBZVM-uhGlcB5-oYBAehvgw,2263
-lemonade/tools/server/utils/system_tray.py,sha256=b9lvNv9chJKQxvmH7qzAuUe6H9HsLu7pdHFqGlAJaL0,12654
 lemonade/tools/server/utils/thread.py,sha256=Z-PDzGcpgfN2qxTmtlROWqrUN0B2fXdPrqo_J10fR_w,2772
+lemonade/tools/server/utils/windows_tray.py,sha256=2z5aTmUPlkT-QfkcfwHsyA6dv6nSNBT0gXUErarhac8,13170
 lemonade_install/__init__.py,sha256=26zohKg2jgr_5y7tObduWMYQg8zCTWMZHL8lfi2zZVQ,40
-lemonade_install/install.py,sha256=Dow7kt-K9WI4PH15hBwkKtOxede3dAaOmH4I1y_P5H4,27008
-lemonade_sdk-8.1.10.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
-lemonade_sdk-8.1.10.dist-info/licenses/NOTICE.md,sha256=RSca9LE5e6pvdWA_LXAUCcACIHPmINKqkRX-AVRqBGo,3499
-lemonade_server/cli.py,sha256=0Ht82D1z4Z2MuxTc07jF2glaegX4jT3Jh--sS6ZO6Qc,19302
-lemonade_server/model_manager.py,sha256=V8QRf1nlh3wAFtUHoSF_JeAXeR7sfaZE1uTfppcIfcw,20492
-lemonade_server/pydantic_models.py,sha256=49MyOlb5feLUlKsGcI75tWaflWckrItqcSVkdCY4e3A,3269
-lemonade_server/server_models.json,sha256=0H_G6Jw6Yuz6t0RZnFnq0SbBCsw_cQLe9j24TkyF2eI,12344
+lemonade_install/install.py,sha256=p3pYqhUnLQ9JJMcbjlSYDYqN-amnU_535O9Oj1yPbyM,27608
+lemonade_sdk-8.1.12.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+lemonade_sdk-8.1.12.dist-info/licenses/NOTICE.md,sha256=RSca9LE5e6pvdWA_LXAUCcACIHPmINKqkRX-AVRqBGo,3499
+lemonade_server/cli.py,sha256=mzlIqLeGkU96KHuib8lmZn0snU7XjQ9lspOy83dVplo,23401
+lemonade_server/model_manager.py,sha256=6W6_nQea6hLD82Il2o_EgQ7oNaLfICXPKNjZiY9Y1Xk,26331
+lemonade_server/pydantic_models.py,sha256=5U3PZ__UqcWQh-dNXVBc-vyJc6-In2vngZXP9VmiScM,3954
+lemonade_server/server_models.json,sha256=NQ-EgcZ-axyul3aEtlV82rbrTc1Y0gySNQoRpZx0D-8,14996
 lemonade_server/settings.py,sha256=JOlZmirUXO9rA6BCODVFwyXrrHtYoH_LiKYm49lGm_c,1260
-lemonade_sdk-8.1.10.dist-info/METADATA,sha256=EYovzTHGnvWEZI-v_Gg1X0ajXeXiQPydqOkisnh08ME,15023
-lemonade_sdk-8.1.10.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-lemonade_sdk-8.1.10.dist-info/entry_points.txt,sha256=7sRvpNhi1E7amnM7RZo57e8yFF9iA5uuRaIeJ1Xre6w,193
-lemonade_sdk-8.1.10.dist-info/top_level.txt,sha256=10ap5GNiPhalO4V50LRoxA1FqRT9g3Xkia6BITu880k,42
-lemonade_sdk-8.1.10.dist-info/RECORD,,
+lemonade_sdk-8.1.12.dist-info/METADATA,sha256=NzOwa4M6tbW-nHUx-IiDDsGlkAPyqX-pxYYso_GnTtg,15390
+lemonade_sdk-8.1.12.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+lemonade_sdk-8.1.12.dist-info/entry_points.txt,sha256=7sRvpNhi1E7amnM7RZo57e8yFF9iA5uuRaIeJ1Xre6w,193
+lemonade_sdk-8.1.12.dist-info/top_level.txt,sha256=10ap5GNiPhalO4V50LRoxA1FqRT9g3Xkia6BITu880k,42
+lemonade_sdk-8.1.12.dist-info/RECORD,,