lemonade-sdk 8.0.5__py3-none-any.whl → 8.1.0__py3-none-any.whl

lemonade/tools/report/table.py

@@ -758,15 +758,18 @@ class LemonadePerfTable(Table):
             data[key] = model_stats.get(key, "")
 
         # Create a new entry with Driver Versions and relevant Python Packages
-        sw_versions = [
-            key + ": " + value
-            for key, value in data[fs.Keys.SYSTEM_INFO]["Driver Versions"].items()
-        ]
-        sw_versions += [
-            pkg
-            for pkg in data[fs.Keys.SYSTEM_INFO]["Python Packages"]
-            if any(name in pkg for name in PYTHON_PACKAGES)
-        ]
+        sw_versions = []
+        if "Driver Versions" in data[fs.Keys.SYSTEM_INFO]:
+            sw_versions += [
+                key + ": " + value
+                for key, value in data[fs.Keys.SYSTEM_INFO]["Driver Versions"].items()
+            ]
+        if "Python Packages" in data[fs.Keys.SYSTEM_INFO]:
+            sw_versions += [
+                pkg
+                for pkg in data[fs.Keys.SYSTEM_INFO]["Python Packages"]
+                if any(name in pkg for name in PYTHON_PACKAGES)
+            ]
         if isinstance(data[Keys.RYZEN_AI_VERSION_INFO], dict):
             sw_versions += [
                 "Ryzen AI: " + value

lemonade/tools/server/llamacpp.py

@@ -1,13 +1,11 @@
-import sys
 import os
+import sys
 import logging
 import time
 import subprocess
-import zipfile
 import re
 import threading
 import platform
-import shutil
 
 import requests
 from tabulate import tabulate
@@ -18,12 +16,18 @@ from openai import OpenAI
 
 from lemonade_server.pydantic_models import (
     ChatCompletionRequest,
+    CompletionRequest,
     PullConfig,
     EmbeddingsRequest,
     RerankingRequest,
 )
 from lemonade_server.model_manager import ModelManager
 from lemonade.tools.server.utils.port import find_free_port
+from lemonade.tools.llamacpp.utils import (
+    get_llama_server_exe_path,
+    install_llamacpp,
+    download_gguf,
+)
 
 LLAMA_VERSION = "b5787"
 
@@ -80,39 +84,6 @@ def get_binary_url_and_filename(version):
     return url, filename
 
 
-def validate_platform_support():
-    """
-    Validate platform support before attempting download
-    """
-    system = platform.system().lower()
-
-    if system not in ["windows", "linux"]:
-        raise HTTPException(
-            status_code=status.HTTP_422_UNPROCESSABLE_ENTITY,
-            detail=(
-                f"Platform {system} not supported for llamacpp. "
-                "Supported: Windows, Ubuntu Linux"
-            ),
-        )
-
-    if system == "linux":
-        # Check if we're actually on Ubuntu/compatible distro and log a warning if not
-        try:
-            with open("/etc/os-release", "r", encoding="utf-8") as f:
-                os_info = f.read().lower()
-                if "ubuntu" not in os_info and "debian" not in os_info:
-                    logging.warning(
-                        "llamacpp binaries are built for Ubuntu. "
-                        "Compatibility with other Linux distributions is not guaranteed."
-                    )
-        except (FileNotFoundError, PermissionError, OSError) as e:
-            logging.warning(
-                "Could not determine Linux distribution (%s). "
-                "llamacpp binaries are built for Ubuntu.",
-                str(e),
-            )
-
-
 class LlamaTelemetry:
     """
     Manages telemetry data collection and display for llama server.
@@ -283,7 +254,7 @@ def _launch_llama_subprocess(
     """
 
     # Get the current executable path (handles both Windows and Ubuntu structures)
-    _, exe_path = get_llama_server_paths()
+    exe_path = get_llama_server_exe_path()
 
     # Build the base command
     base_command = [exe_path, "-m", snapshot_files["variant"]]
@@ -350,68 +321,23 @@ def _launch_llama_subprocess(
 
 
 def server_load(model_config: PullConfig, telemetry: LlamaTelemetry):
-    # Validate platform support before proceeding
-    validate_platform_support()
+    # Install and/or update llama.cpp if needed
+    try:
+        install_llamacpp()
+    except NotImplementedError as e:
+        raise HTTPException(
+            status_code=status.HTTP_422_UNPROCESSABLE_ENTITY, detail=str(e)
+        )
 
     # Get platform-specific paths at runtime
-    llama_server_exe_dir, llama_server_exe_path = get_llama_server_paths()
-
-    # Check whether the llamacpp install needs an upgrade
-    version_txt_path = os.path.join(llama_server_exe_dir, "version.txt")
-    if os.path.exists(version_txt_path):
-        with open(version_txt_path, "r", encoding="utf-8") as f:
-            llamacpp_installed_version = f.read()
-
-        if llamacpp_installed_version != LLAMA_VERSION:
-            # Remove the existing install, which will trigger a new install
-            # in the next code block
-            shutil.rmtree(llama_server_exe_dir)
-
-    # Download llama.cpp server if it isn't already available
-    if not os.path.exists(llama_server_exe_dir):
-        # Download llama.cpp server zip
-        llama_zip_url, filename = get_binary_url_and_filename(LLAMA_VERSION)
-        llama_zip_path = os.path.join(os.path.dirname(sys.executable), filename)
-        logging.info(f"Downloading llama.cpp server from {llama_zip_url}")
-
-        with requests.get(llama_zip_url, stream=True) as r:
-            r.raise_for_status()
-            with open(llama_zip_path, "wb") as f:
-                for chunk in r.iter_content(chunk_size=8192):
-                    f.write(chunk)
-
-        # Extract zip
-        logging.info(f"Extracting {llama_zip_path} to {llama_server_exe_dir}")
-        with zipfile.ZipFile(llama_zip_path, "r") as zip_ref:
-            zip_ref.extractall(llama_server_exe_dir)
-
-        # Make executable on Linux - need to update paths after extraction
-        if platform.system().lower() == "linux":
-            # Re-get the paths since extraction might have changed the directory structure
-            _, updated_exe_path = get_llama_server_paths()
-            if os.path.exists(updated_exe_path):
-                os.chmod(updated_exe_path, 0o755)
-                logging.info(f"Set executable permissions for {updated_exe_path}")
-            else:
-                logging.warning(
-                    f"Could not find llama-server executable at {updated_exe_path}"
-                )
-
-        # Save version.txt
-        with open(version_txt_path, "w", encoding="utf-8") as vf:
-            vf.write(LLAMA_VERSION)
-
-        # Delete zip file
-        os.remove(llama_zip_path)
-        logging.info("Cleaned up zip file")
+    llama_server_exe_path = get_llama_server_exe_path()
 
     # Download the gguf to the hugging face cache
-    model_manager = ModelManager()
-    snapshot_files = model_manager.download_gguf(model_config)
+    snapshot_files = download_gguf(model_config.checkpoint, model_config.mmproj)
     logging.debug(f"GGUF file paths: {snapshot_files}")
 
     # Check if model supports embeddings
-    supported_models = model_manager.supported_models
+    supported_models = ModelManager().supported_models
     model_info = supported_models.get(model_config.model_name, {})
     supports_embeddings = "embeddings" in model_info.get("labels", [])
     supports_reranking = "reranking" in model_info.get("labels", [])
@@ -523,6 +449,68 @@ def chat_completion(
             )
 
 
+def completion(completion_request: CompletionRequest, telemetry: LlamaTelemetry):
+    """
+    Handle text completions using the llamacpp server.
+
+    Args:
+        completion_request: The completion request containing prompt and parameters
+        telemetry: Telemetry object containing the server port
+
+    Returns:
+        Completion response from the llamacpp server
+    """
+    base_url = llamacpp_address(telemetry.port)
+    client = OpenAI(
+        base_url=base_url,
+        api_key="lemonade",
+    )
+
+    # Convert Pydantic model to dict and remove unset/null values
+    request_dict = completion_request.model_dump(exclude_unset=True, exclude_none=True)
+
+    # Check if streaming is requested
+    if completion_request.stream:
+
+        def event_stream():
+            try:
+                # Enable streaming
+                for chunk in client.completions.create(**request_dict):
+                    yield f"data: {chunk.model_dump_json()}\n\n"
+                yield "data: [DONE]\n\n"
+
+                # Show telemetry after completion
+                telemetry.show_telemetry()
+
+            except Exception as e:  # pylint: disable=broad-exception-caught
+                yield f'data: {{"error": "{str(e)}"}}\n\n'
+
+        return StreamingResponse(
+            event_stream(),
+            media_type="text/event-stream",
+            headers={
+                "Cache-Control": "no-cache",
+                "Connection": "keep-alive",
+            },
+        )
+    else:
+        # Non-streaming response
+        try:
+            # Disable streaming for non-streaming requests
+            response = client.completions.create(**request_dict)
+
+            # Show telemetry after completion
+            telemetry.show_telemetry()
+
+            return response
+
+        except Exception as e:  # pylint: disable=broad-exception-caught
+            raise HTTPException(
+                status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+                detail=f"Completion error: {str(e)}",
+            )
+
+
 def embeddings(embeddings_request: EmbeddingsRequest, telemetry: LlamaTelemetry):
     """
     Generate embeddings using the llamacpp server.

lemonade/tools/server/serve.py

@@ -284,7 +284,7 @@ class Server(ManagementTool):
     def _setup_server_common(
         self,
         port: int,
-        truncate_inputs: bool = False,
+        truncate_inputs: Optional[int] = None,
         log_level: str = DEFAULT_LOG_LEVEL,
         tray: bool = False,
         log_file: str = None,
@@ -295,7 +295,7 @@ class Server(ManagementTool):
 
         Args:
             port: Port number for the server
-            truncate_inputs: Whether to truncate inputs if they exceed max length
+            truncate_inputs: Truncate messages to this length
             log_level: Logging level to configure
             threaded_mode: Whether this is being set up for threaded execution
         """
@@ -372,7 +372,7 @@ class Server(ManagementTool):
         _=None,
         port: int = DEFAULT_PORT,
         log_level: str = DEFAULT_LOG_LEVEL,
-        truncate_inputs: bool = False,
+        truncate_inputs: Optional[int] = None,
         tray: bool = False,
         log_file: str = None,
     ):
@@ -393,7 +393,7 @@ class Server(ManagementTool):
         port: int = DEFAULT_PORT,
         host: str = "localhost",
         log_level: str = "warning",
-        truncate_inputs: bool = False,
+        truncate_inputs: Optional[int] = None,
     ):
         """
         Set up the server for running in a thread.
@@ -487,6 +487,9 @@ class Server(ManagementTool):
         # Load the model if it's different from the currently loaded one
         await self.load_llm(lc)
 
+        if self.llm_loaded.recipe == "llamacpp":
+            return llamacpp.completion(completion_request, self.llama_telemetry)
+
         # Check if the model supports reasoning
         reasoning_first_token = self.llm_loaded.reasoning
 
@@ -1096,29 +1099,20 @@ class Server(ManagementTool):
             )
             self.input_tokens = len(input_ids[0])
 
-        if (
-            self.llm_loaded.max_prompt_length
-            and self.input_tokens > self.llm_loaded.max_prompt_length
-        ):
-            if self.truncate_inputs:
-                # Truncate input ids
-                truncate_amount = self.input_tokens - self.llm_loaded.max_prompt_length
-                input_ids = input_ids[: self.llm_loaded.max_prompt_length]
-
-                # Update token count
-                self.input_tokens = len(input_ids)
-
-                # Show warning message
-                truncation_message = (
-                    f"Input exceeded {self.llm_loaded.max_prompt_length} tokens. "
-                    f"Truncated {truncate_amount} tokens."
-                )
-                logging.warning(truncation_message)
-            else:
-                raise RuntimeError(
-                    f"Prompt tokens ({self.input_tokens}) cannot be greater "
-                    f"than the model's max prompt length ({self.llm_loaded.max_prompt_length})"
-                )
+        if self.truncate_inputs and self.truncate_inputs > self.input_tokens:
+            # Truncate input ids
+            truncate_amount = self.input_tokens - self.truncate_inputs
+            input_ids = input_ids[: self.truncate_inputs]
+
+            # Update token count
+            self.input_tokens = len(input_ids)
+
+            # Show warning message
+            truncation_message = (
+                f"Input exceeded {self.truncate_inputs} tokens. "
+                f"Truncated {truncate_amount} tokens."
+            )
+            logging.warning(truncation_message)
 
         # Log the input tokens early to avoid this not showing due to potential crashes
         logging.debug(f"Input Tokens: {self.input_tokens}")
@@ -1314,7 +1308,7 @@ class Server(ManagementTool):
         self.tokenizer = None
         self.model = None
 
-        default_message = f"model {model_reference} not found"
+        default_message = "see stack trace and error message below"
         if message:
             detail = message
         else:

lemonade/tools/server/static/styles.css

@@ -1,33 +1,92 @@
+/* === CSS Variables === */
+:root {
+  /* Colors */
+  --primary-yellow: #ffe066;
+  --primary-yellow-dark: #ffd43b;
+  --accent-gold: #e6b800;
+  --accent-gold-dark: #bfa100;
+  
+  --text-primary: #222;
+  --text-secondary: #555;
+  --text-muted: #666;
+  
+  --bg-primary: #fffbe9;
+  --bg-secondary: #fff8dd;
+  --bg-tertiary: #fff5d1;
+  
+  /* Transitions */
+  --transition-fast: 0.2s ease;
+  --transition-medium: 0.3s ease;
+}
+
 body {
   margin: 0;
   font-family: 'Segoe UI', 'Arial', sans-serif;
-  background: #fffbe9;
-  color: #222;
+  background: linear-gradient(135deg, var(--bg-primary) 0%, var(--bg-secondary) 50%, var(--bg-tertiary) 100%);
+  color: var(--text-primary);
   min-height: 100vh;
   display: flex;
   flex-direction: column;
-  padding-bottom: 5rem;
+}
+
+body::before {
+  content: '';
+  position: fixed;
+  top: 0;
+  left: 0;
+  width: 100%;
+  height: 100%;
+  background: 
+    radial-gradient(circle at 20% 20%, rgba(255, 224, 102, 0.1) 0%, transparent 50%),
+    radial-gradient(circle at 80% 80%, rgba(255, 212, 59, 0.1) 0%, transparent 50%);
+  pointer-events: none;
+  z-index: -1;
 }
 
 .navbar {
   display: flex;
-  justify-content: center;
-  gap: 2.5rem;
-  padding: 2rem 0 1.5rem 0;
+  justify-content: space-between;
+  align-items: center;
+  padding: 1rem 3rem 0.5rem 1rem;
   font-size: 1.25rem;
   font-weight: 500;
   background: transparent;
   letter-spacing: 0.02em;
+  position: relative;
+  transition: var(--transition-medium);
 }
 
-.navbar a {
+.navbar-brand {
+  display: flex;
+  align-items: center;
+}
+
+.brand-title {
+  font-size: 1.5rem;
+  font-weight: 700;
+  color: var(--text-primary);
+  text-decoration: none;
+  letter-spacing: 0.01em;
+}
+
+.brand-title a {
+  color: inherit;
+  text-decoration: none;
+}
+
+.navbar-links {
+  display: flex;
+  gap: 2.5rem;
+}
+
+.navbar-links a {
   color: #444;
   text-decoration: none;
-  transition: color 0.2s;
+  transition: var(--transition-fast);
 }
 
-.navbar a:hover {
-  color: #e6b800;
+.navbar-links a:hover {
+  color: var(--accent-gold);
 }
 
 .main {
@@ -37,26 +96,14 @@ body {
   align-items: center;
   justify-content: flex-start;
   min-height: 60vh;
-  margin-top: 3rem;
-}
-
-.title {
-  font-size: 3rem;
-  font-weight: 700;
-  margin-bottom: 2.5rem;
-  letter-spacing: 0.01em;
-  text-align: center;
-  color: #222;
+  margin-top: 2rem;
+  padding-top: 1rem;
 }
 
 .site-footer {
-  position: fixed;
-  left: 0;
-  bottom: 0;
-  width: 100%;
-  background-color: #fffbe9;
+  background: transparent;
   padding-top: 0.5rem;
-  z-index: 100;
+  margin-top: auto;
 }
 
 .dad-joke {
@@ -483,6 +530,10 @@ body {
   background-color: #ca4747;
 }
 
+.model-label.coding {
+  background-color: #ff6b35;
+}
+
 .model-labels-container {
   display: flex;
   align-items: center;
@@ -983,6 +1034,50 @@ body {
   }
 }
 
+/* === Responsive Navbar === */
+@media (max-width: 800px) {
+  .navbar {
+    flex-direction: column;
+    gap: 1rem;
+    padding: 1rem 1rem 0.5rem 1rem;
+    align-items: center;
+  }
+  
+  .navbar-brand {
+    margin-bottom: 0.5rem;
+  }
+  
+  .brand-title {
+    font-size: 1.3rem;
+  }
+  
+  .navbar-links {
+    gap: 1.5rem;
+    font-size: 1rem;
+  }
+}
+
+@media (max-width: 600px) {
+  .navbar {
+    padding: 0.5rem 0.5rem 0.25rem 0.5rem;
+  }
+  
+  .brand-title {
+    font-size: 1.2rem;
+  }
+  
+  .navbar-links {
+    gap: 1rem;
+    font-size: 0.9rem;
+    flex-wrap: wrap;
+    justify-content: center;
+  }
+  
+  .main {
+    margin-top: 0.5rem;
+  }
+}
+
 /* Ensure form container allows tooltip overflow */
 .model-mgmt-register-form {
   position: relative;

lemonade/tools/server/static/webapp.html

@@ -12,14 +12,19 @@
     {{SERVER_MODELS_JS}}
 </head>
 <body>
-    <nav class="navbar">
-        <a href="https://github.com/lemonade-sdk/lemonade" target="_blank">GitHub</a>
-        <a href="https://lemonade-server.ai/docs/" target="_blank">Docs</a>
-        <a href="https://lemonade-server.ai/docs/server/server_models/" target="_blank">Models</a>
-        <a href="https://lemonade-server.ai/docs/server/apps/" target="_blank">Featured Apps</a>
+    <nav class="navbar" id="navbar">
+        <div class="navbar-brand">
+            <span class="brand-title"><a href="https://lemonade-server.ai">🍋 Lemonade Server</a></span>
+        </div>
+        <div class="navbar-links">
+            <a href="https://github.com/lemonade-sdk/lemonade" target="_blank">GitHub</a>
+            <a href="https://lemonade-server.ai/docs/" target="_blank">Docs</a>
+            <a href="https://lemonade-server.ai/docs/server/server_models/" target="_blank">Models</a>
+            <a href="https://lemonade-server.ai/docs/server/apps/" target="_blank">Featured Apps</a>
+            <a href="https://lemonade-server.ai/news/" target="_blank">News</a>
+        </div>
     </nav>
     <main class="main">
-        <div class="title">🍋 Lemonade Server</div>
         <div class="tab-container"> 
             <div class="tabs"> 
                 <button class="tab active" id="tab-chat" onclick="showTab('chat')">LLM Chat</button> 
@@ -104,6 +109,7 @@
                         </label>
                         <select id="register-recipe" name="recipe" required>
                             <option value="llamacpp">llamacpp</option>
+                            <option value="oga-npu">oga-npu</option>
                             <option value="oga-hybrid">oga-hybrid</option>
                             <option value="oga-cpu">oga-cpu</option>
                         </select>
@@ -408,6 +414,8 @@
                     labelClass = 'reasoning';
                 } else if (labelLower === 'reranking') {
                     labelClass = 'reranking';
+                } else if (labelLower === 'coding') {
+                    labelClass = 'coding';
                 }
                 labelSpan.className = `model-label ${labelClass}`;
                 labelSpan.textContent = label;

lemonade/tools/server/tray.py

@@ -87,8 +87,15 @@ class LemonadeTray(SystemTray):
         Update the latest version information.
         """
         try:
+            # Prepare headers for GitHub API request
+            headers = {}
+            github_token = os.environ.get("GITHUB_TOKEN")
+            if github_token:
+                headers["Authorization"] = f"token {github_token}"
+
             response = requests.get(
                 "https://api.github.com/repos/lemonade-sdk/lemonade/releases/latest",
+                headers=headers,
                 timeout=10,  # Add timeout to prevent hanging
             )
             response.raise_for_status()

lemonade/version.py

@@ -1 +1 @@
-__version__ = "8.0.5"
+__version__ = "8.1.0"