local-llm-server 0.1.0__py3-none-any.whl

local_llm_server/__init__.py

@@ -0,0 +1,22 @@
+"""local-llm-server — ローカルLLMを OpenAI 互換 API として起動・管理する。
+
+mlx / mlx-vlm / llama.cpp、および画像とテキストを自動振り分けする router を
+サブプロセスとして起動・監視する。依存は標準ライブラリのみで、任意の
+OpenAI 互換クライアントから利用できる。
+
+公開 API:
+  - サーバー管理: LocalServer / ServerConfig / ServerPool / build_command など
+  - 起動状態の確認: is_ready / list_models / models_match / parse_host_port
+  - 既定値・定数: DEFAULT_MODEL / DEFAULT_VISION_MODEL / DEFAULT_BACKEND / BACKENDS
+  - router: RouterServer
+"""
+from __future__ import annotations
+
+from .constants import (  # noqa: F401
+    BACKENDS,
+    DEFAULT_MODEL,
+    DEFAULT_VISION_MODEL,
+    _env_bool,
+)
+from .server import *  # noqa: F401,F403
+from .router import *  # noqa: F401,F403

local_llm_server/cli.py

@@ -0,0 +1,282 @@
+from __future__ import annotations
+
+import argparse
+import os
+import sys
+import threading
+
+from .constants import DEFAULT_MODEL, DEFAULT_VISION_MODEL, _env_bool
+from .router import RouterServer
+from .server import (
+    BACKENDS,
+    DEFAULT_BACKEND,
+    ServerConfig,
+    ServerPool,
+    build_command,
+    build_pool_configs,
+    find_pids_on_port,
+    install_shutdown_handlers,
+    parallel_supported,
+    stop_pid,
+)
+
+# CLI で選べるバックエンド。router は内部でテキスト LLM と vision VLM を
+# 同時起動し、リクエスト内容で自動振り分けする特別モード。
+ROUTER = "router"
+CLI_BACKENDS = (*BACKENDS, ROUTER)
+
+
+def _run_router(args, extra: list[str], enable_thinking: bool) -> int:
+    """router モード: テキスト LLM と vision VLM を同時起動し、プロキシで振り分ける。
+
+    公開ポート(args.port)にルーターを立て、テキストを port+1、vision を port+2 で
+    起動する。クライアントは公開ポートの 1 つだけを base_url に指定すればよい。
+    vision バックエンドは mlx-vlm（Apple Silicon 前提）、テキストは mlx を使う。
+    """
+    text_port = args.port + 1
+    vision_port = args.port + 2
+    text_cfg = ServerConfig(
+        backend="mlx",
+        model=args.model,
+        host=args.host,
+        port=text_port,
+        disable_thinking=not enable_thinking,
+        extra_args=extra,
+    )
+    vision_cfg = ServerConfig(
+        backend="mlx-vlm",
+        model=args.vision_model,
+        host=args.host,
+        port=vision_port,
+    )
+    pool = ServerPool([text_cfg, vision_cfg])
+    print(f"Starting (text):   {' '.join(build_command(text_cfg))}", file=sys.stderr)
+    print(f"Starting (vision): {' '.join(build_command(vision_cfg))}", file=sys.stderr)
+    try:
+        pool.start()
+        pool.wait_until_ready()
+    except (RuntimeError, TimeoutError) as exc:
+        print(f"Error: {exc}", file=sys.stderr)
+        pool.stop()
+        return 1
+
+    router = RouterServer(
+        (args.host, args.port),
+        (args.host, text_port),
+        (args.host, vision_port),
+    )
+    public_url = f"http://{args.host}:{args.port}/v1"
+    print("Ready (automatic routing):", file=sys.stderr)
+    print(f"  - public: {public_url}", file=sys.stderr)
+    print(f"      text -> http://{args.host}:{text_port}/v1", file=sys.stderr)
+    print(f"      image/media -> http://{args.host}:{vision_port}/v1", file=sys.stderr)
+    print(
+        f'In agent.toml, set only the single base_url = "{public_url}". '
+        "Text-only requests are routed to the LLM, requests with images to the VLM, automatically.",
+        file=sys.stderr,
+    )
+    print(
+        "Note: if the VLM does not support function calling (tools), "
+        'set tool_mode = "prompt" when using images.',
+        file=sys.stderr,
+    )
+
+    thread = threading.Thread(target=router.serve_forever, daemon=True)
+    thread.start()
+    try:
+        pool.wait()  # 子サーバーが終了するまでブロック
+    except KeyboardInterrupt:
+        pass
+    finally:
+        router.shutdown()
+        router.server_close()
+        pool.stop()
+    return 0
+
+
+def _stop_servers(ports: list[int]) -> int:
+    """指定ポートで動いているローカルサーバーを探して停止する（--stop 用）。
+
+    各ポートを LISTEN しているプロセスを lsof で特定し、プロセスグループごと
+    SIGTERM→SIGKILL で止める。1つでも止めれば 0、見つからなければ 1 を返す。
+    """
+    if os.name != "posix":
+        print(
+            "--stop is supported on macOS / Linux only. On Windows, stop the server "
+            "from its own window (Ctrl+C) or via Task Manager.",
+            file=sys.stderr,
+        )
+        return 1
+    stopped = False
+    for port in ports:
+        pids = find_pids_on_port(port)
+        for pid in pids:
+            print(f"Stopping the server on port {port} (pid {pid})...", file=sys.stderr)
+            if stop_pid(pid):
+                stopped = True
+    if not stopped:
+        ports_str = ", ".join(str(p) for p in ports)
+        print(f"No running server found on port(s): {ports_str}.", file=sys.stderr)
+        return 1
+    print("Stopped.", file=sys.stderr)
+    return 0
+
+
+def main(argv: list[str] | None = None) -> int:
+    parser = argparse.ArgumentParser(
+        prog="local-llm-server",
+        description=(
+            "Start a local LLM server (mlx / mlx-vlm / llama.cpp / router). "
+            "Use the vision-capable mlx-vlm for image input (--backend mlx-vlm). "
+            "With --backend router, start a text LLM and a VLM together and route requests automatically based on their content."
+        ),
+        epilog="Backend-specific extra arguments can be passed after --.",
+    )
+    parser.add_argument(
+        "--backend",
+        choices=CLI_BACKENDS,
+        default=os.environ.get("CODER_BACKEND", DEFAULT_BACKEND),
+        help="Backend to use (env: CODER_BACKEND, default: %(default)s)",
+    )
+    parser.add_argument(
+        "--model",
+        default=os.environ.get("CODER_MODEL", DEFAULT_MODEL),
+        help="Model (mlx: name/path, llama-cpp: .gguf path; used as the text model in router mode). env: CODER_MODEL, default: %(default)s",
+    )
+    parser.add_argument(
+        "--vision-model",
+        default=os.environ.get("CODER_VISION_MODEL", DEFAULT_VISION_MODEL),
+        help="Vision model that handles images/media in router mode. env: CODER_VISION_MODEL, default: %(default)s",
+    )
+    parser.add_argument("--host", default="127.0.0.1", help="Host to bind to")
+    parser.add_argument("--port", type=int, default=8080, help="Port to bind to (the first port when using a pool)")
+    parser.add_argument(
+        "--stop",
+        action="store_true",
+        help="Stop a server already running on --port instead of starting one "
+        "(use the same --port / --instances / --backend router as when you started it). macOS / Linux only.",
+    )
+    parser.add_argument(
+        "--parallel",
+        type=int,
+        help="Number of concurrent processing slots (llama.cpp only; mlx processes sequentially)",
+    )
+    parser.add_argument(
+        "--instances",
+        type=int,
+        default=1,
+        help="Number of servers to start on consecutive ports (to gain parallelism with mlx; each loads its own model)",
+    )
+    parser.add_argument(
+        "--thinking",
+        action=argparse.BooleanOptionalAction,
+        default=None,
+        help="Thinking mode. Use --no-thinking to disable (env: CODER_ENABLE_THINKING)",
+    )
+    parser.add_argument(
+        "--draft-model",
+        default=os.environ.get("CODER_DRAFT_MODEL"),
+        help="MTP drafter for Gemma 4 (HF id/path, e.g. "
+        "mlx-community/gemma-4-E4B-it-qat-assistant-bf16). Use 'auto' to select automatically from the main model name. "
+        "Speeds up inference without changing the main model's output. mlx-vlm backend only. env: CODER_DRAFT_MODEL",
+    )
+    args, extra = parser.parse_known_args(argv)
+
+    if args.instances < 1:
+        parser.error("--instances must be 1 or greater")
+
+    # --stop: 起動の代わりに、起動時と同じポート割り当てを再現して停止する。
+    # router は public(port) / text(port+1) / vision(port+2)、通常は連番ポート。
+    if args.stop:
+        if args.backend == ROUTER:
+            ports = [args.port, args.port + 1, args.port + 2]
+        else:
+            ports = [args.port + i for i in range(args.instances)]
+        return _stop_servers(ports)
+
+    # kill / ターミナルを閉じる（SIGTERM・SIGHUP）でも下流の finally（pool.stop）を
+    # 必ず通し、起動したバックエンドを孫プロセスとして残さない。
+    install_shutdown_handlers()
+
+    if args.parallel is not None:
+        if args.parallel < 1:
+            parser.error("--parallel must be 1 or greater")
+        if not parallel_supported(args.backend):
+            print(
+                f"Warning: {args.backend} does not support parallel slots "
+                "(sequential processing). --parallel is ignored.",
+                file=sys.stderr,
+            )
+
+    if args.backend == "mlx-vlm":
+        print(
+            "Info: starting with the vision backend (mlx-vlm). It supports image input (image_url), "
+            "but some models do not support function calling (tools)/streaming. "
+            'In that case, set tool_mode = "prompt" in agent.toml.',
+            file=sys.stderr,
+        )
+
+    enable_thinking = (
+        args.thinking if args.thinking is not None
+        else _env_bool("CODER_ENABLE_THINKING", False)
+    )
+
+    # argparse が拾った先頭の "--" を除去してバックエンドへ素通し
+    if extra and extra[0] == "--":
+        extra = extra[1:]
+
+    if args.backend == ROUTER:
+        if args.instances != 1:
+            parser.error("--instances cannot be used with --backend router")
+        return _run_router(args, extra, enable_thinking)
+
+    base = ServerConfig(
+        backend=args.backend,
+        model=args.model,
+        host=args.host,
+        port=args.port,
+        parallel=args.parallel,
+        disable_thinking=not enable_thinking,
+        draft_model=args.draft_model,
+        extra_args=extra,
+    )
+    configs = build_pool_configs(base, args.instances)
+    pool = ServerPool(configs)
+
+    for config in configs:
+        print(f"Starting: {' '.join(build_command(config))}", file=sys.stderr)
+    try:
+        pool.start()
+        pool.wait_until_ready()
+    except (RuntimeError, TimeoutError) as exc:
+        print(f"Error: {exc}", file=sys.stderr)
+        pool.stop()
+        return 1
+
+    print("Ready:", file=sys.stderr)
+    for url in pool.base_urls:
+        print(f"  - {url}", file=sys.stderr)
+    if args.instances > 1:
+        print(
+            "Point each OpenAI-compatible client at one of the base URLs above. "
+            f'Example: base_url = "{pool.base_urls[0]}"',
+            file=sys.stderr,
+        )
+    else:
+        print(
+            "Point your OpenAI-compatible client at "
+            f'base_url = "{pool.base_urls[0]}".',
+            file=sys.stderr,
+        )
+
+    try:
+        pool.wait()
+    except KeyboardInterrupt:
+        pass
+    finally:
+        pool.stop()
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

local_llm_server/constants.py

@@ -0,0 +1,26 @@
+"""エージェント側（agent/）とサーバー側（server/）で共有する中立な定数・ヘルパー。
+
+どちらのサブパッケージにも依存しないため、ここを基点に server/ と agent/ を
+完全に分離できる（server/ が agent/ を import する逆向き依存を作らないための層）。
+"""
+from __future__ import annotations
+
+import os
+
+# サーバー・モデルが明示されないときに使う既定モデル
+DEFAULT_MODEL = "mlx-community/Qwen3.6-27B-4bit"
+
+# 画像・メディアを処理する vision モデルの既定。
+# 既定モデル(Qwen3.6)はマルチモーダルなので、テキストと共通のものを使う。
+DEFAULT_VISION_MODEL = DEFAULT_MODEL
+
+# 起動可能なローカルLLMサーバーのバックエンド一覧。
+# server/ が実装し、agent/ は agent.toml の backend 検証に使う（共有値）。
+BACKENDS = ("mlx", "mlx-vlm", "llama-cpp")
+
+
+def _env_bool(name: str, default: bool) -> bool:
+    value = os.environ.get(name)
+    if value is None:
+        return default
+    return value.strip().lower() in ("1", "true", "yes", "on")

local_llm_server/router.py

@@ -0,0 +1,136 @@
+"""テキスト LLM と vision VLM を内部で振り分けるルーティングプロキシ。
+
+`local-llm-server --backend router` で起動する。1 つの OpenAI 互換
+エンドポイント（例 http://127.0.0.1:8080/v1）を公開し、受信した
+`/v1/chat/completions` の `messages` を検査して:
+
+  - 画像など非テキストのコンテンツパートを含む → vision バックエンド(VLM)
+  - テキストのみ                              → テキスト LLM
+
+へ転送する（ストリーミングはそのまま中継）。クライアントの base_url は
+1 つのままで、内部の 2 プロセスを意識しなくてよい。
+"""
+from __future__ import annotations
+
+import http.client
+import json
+from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
+from typing import Any
+
+
+def needs_vision(messages: Any) -> bool:
+    """messages に非テキストのコンテンツパート（画像など）が含まれるか。
+
+    OpenAI 互換の content は文字列か、{"type": "text"|"image_url"|...} の配列。
+    type が text 以外のパートが 1 つでもあれば vision 扱いにする。
+    """
+    if not isinstance(messages, list):
+        return False
+    for message in messages:
+        if not isinstance(message, dict):
+            continue
+        content = message.get("content")
+        if not isinstance(content, list):
+            continue
+        for part in content:
+            if isinstance(part, dict) and part.get("type") not in (None, "text"):
+                return True
+    return False
+
+
+class _RouterHandler(BaseHTTPRequestHandler):
+    server_version = "local-llm-router"
+
+    # HTTP/1.0: 応答ボディは接続クローズ区切り。ストリーミング/非ストリーミングを
+    # 一律に「上流を読み切るまで中継して閉じる」形で扱え、Content-Length 計算が不要。
+    protocol_version = "HTTP/1.0"
+
+    def log_message(self, *_args) -> None:  # アクセスログは出さない
+        pass
+
+    # --- ルーティング ---------------------------------------------------
+    def do_GET(self) -> None:
+        # is_ready 用の /v1/models など。読み取り系はテキスト上流へ。
+        self._proxy(self.server.text_addr, body=b"")  # type: ignore[attr-defined]
+
+    def do_POST(self) -> None:
+        length = int(self.headers.get("Content-Length") or 0)
+        body = self.rfile.read(length) if length else b""
+        target = self._select_target(body)
+        self._proxy(target, body)
+
+    def _select_target(self, body: bytes) -> tuple[str, int]:
+        srv = self.server  # type: ignore[assignment]
+        # chat/completions 以外（埋め込み等）はテキスト上流へ素通し。
+        if not self.path.rstrip("/").endswith("/chat/completions"):
+            return srv.text_addr
+        try:
+            payload = json.loads(body or b"{}")
+        except (json.JSONDecodeError, ValueError):
+            return srv.text_addr
+        if needs_vision(payload.get("messages")):
+            return srv.vision_addr
+        return srv.text_addr
+
+    # --- 転送 -----------------------------------------------------------
+    def _proxy(self, addr: tuple[str, int], body: bytes) -> None:
+        host, port = addr
+        headers = {"Content-Type": self.headers.get("Content-Type", "application/json")}
+        auth = self.headers.get("Authorization")
+        if auth:
+            headers["Authorization"] = auth
+        if self.command == "POST":
+            headers["Content-Length"] = str(len(body))
+        try:
+            conn = http.client.HTTPConnection(host, port, timeout=self.server.timeout_s)  # type: ignore[attr-defined]
+            conn.request(self.command, self.path, body=body or None, headers=headers)
+            resp = conn.getresponse()
+        except OSError as exc:
+            self._error(502, f"upstream {host}:{port} unreachable: {exc}")
+            return
+        try:
+            self.send_response(resp.status)
+            ctype = resp.getheader("Content-Type")
+            if ctype:
+                self.send_header("Content-Type", ctype)
+            self.end_headers()
+            while True:
+                chunk = resp.read(8192)
+                if not chunk:
+                    break
+                self.wfile.write(chunk)
+                self.wfile.flush()
+        except (BrokenPipeError, ConnectionResetError):
+            pass  # クライアント切断
+        finally:
+            conn.close()
+
+    def _error(self, status: int, message: str) -> None:
+        payload = json.dumps({"error": message}).encode("utf-8")
+        try:
+            self.send_response(status)
+            self.send_header("Content-Type", "application/json")
+            self.end_headers()
+            self.wfile.write(payload)
+        except (BrokenPipeError, ConnectionResetError):
+            pass
+
+
+class RouterServer(ThreadingHTTPServer):
+    """テキスト/vision 上流へ振り分けるプロキシ HTTP サーバー。"""
+
+    daemon_threads = True
+    allow_reuse_address = True
+
+    def __init__(
+        self,
+        addr: tuple[str, int],
+        text_addr: tuple[str, int],
+        vision_addr: tuple[str, int],
+        timeout_s: float | None = None,
+    ) -> None:
+        super().__init__(addr, _RouterHandler)
+        self.text_addr = text_addr
+        self.vision_addr = vision_addr
+        # None なら無制限（長時間生成に備える）。
+        self.timeout_s = timeout_s

local_llm_server/server.py

@@ -0,0 +1,495 @@
+from __future__ import annotations
+
+import json
+import os
+import platform
+import signal
+import subprocess
+import sys
+import tempfile
+import time
+import urllib.error
+import urllib.parse
+import urllib.request
+from dataclasses import dataclass, field, replace
+
+# 起動可能なバックエンド一覧は同梱の constants から取得（OpenAI互換APIの公開値）。
+from .constants import BACKENDS  # noqa: F401
+
+
+def default_backend() -> str:
+    """OS に応じた既定バックエンド。
+
+    Apple Silicon の macOS なら mlx-vlm（vision 対応）を既定にする。既定モデルの
+    Qwen3.6 はマルチモーダルなので、1プロセスでテキストも画像も扱え、画像・動画
+    入力がそのまま動く。テキスト専用で軽くしたい場合は backend="mlx" を選ぶ。
+    それ以外（Linux / Windows / Intel Mac）は llama.cpp を既定にする。
+    """
+    if sys.platform == "darwin" and platform.machine() == "arm64":
+        return "mlx-vlm"
+    return "llama-cpp"
+
+
+# サーバー未起動・バックエンド未指定のときに使う既定バックエンド（OSで自動判定）
+DEFAULT_BACKEND = default_backend()
+
+# POSIX（macOS / Linux）か。プロセスグループ操作（killpg / setsid）の可否に使う。
+_POSIX = os.name == "posix"
+
+
+def install_shutdown_handlers() -> None:
+    """SIGTERM / SIGHUP を Ctrl+C と同じ KeyboardInterrupt に変換する。
+
+    既定では Python は SIGTERM を受け取ると finally を実行せずに即終了するため、
+    `kill <pid>` やターミナルを閉じた（SIGHUP）ときに、自動起動した LLM サーバーが
+    孫プロセスとして置き去りになる。これらのシグナルを KeyboardInterrupt として
+    送出することで、各エントリポイントの既存 try/finally（= server.stop()）を必ず通す。
+    シグナルハンドラはメインスレッドからのみ登録できる（それ以外では黙って無視）。
+    """
+    def _raise_keyboard_interrupt(signum, frame):  # noqa: ANN001
+        raise KeyboardInterrupt
+
+    for name in ("SIGTERM", "SIGHUP"):
+        sig = getattr(signal, name, None)  # SIGHUP は Windows に無い
+        if sig is None:
+            continue
+        try:
+            signal.signal(sig, _raise_keyboard_interrupt)
+        except (ValueError, OSError):
+            pass  # メインスレッド以外などでは登録できない
+
+
+def _signal_process_tree(proc: subprocess.Popen, *, kill: bool) -> None:
+    """proc とその子孫（プロセスグループ）へ終了シグナルを送る。
+
+    POSIX では start() が start_new_session=True で子を独立したプロセスグループに
+    しているため、killpg でグループ全体へ一括送信できる。これによりバックエンドが
+    内部で起こすワーカー等の孫プロセスも取りこぼさない。Windows には killpg が無い
+    ので proc 自身を terminate / kill する。
+    """
+    if proc.poll() is not None:
+        return  # 既に終了している
+    if _POSIX:
+        sig = signal.SIGKILL if kill else signal.SIGTERM
+        try:
+            os.killpg(os.getpgid(proc.pid), sig)
+            return
+        except (ProcessLookupError, PermissionError, OSError):
+            pass  # グループ送信に失敗したら単体送信にフォールバック
+    if kill:
+        proc.kill()
+    else:
+        proc.terminate()
+
+
+def find_pids_on_port(port: int) -> list[int]:
+    """指定ポートを LISTEN しているプロセスの PID 一覧を返す（POSIX のみ、lsof を利用）。
+
+    lsof が無い / Windows などでは空リストを返す（呼び出し側で案内する）。
+    """
+    if not _POSIX:
+        return []
+    try:
+        # プロトコルとポートは 1 つの -i セレクタにまとめる。-iTCP と -i:PORT を
+        # 分けると lsof は両者を OR 解釈し、全 TCP プロセスにマッチしてしまう。
+        result = subprocess.run(
+            ["lsof", "-t", f"-iTCP:{port}", "-sTCP:LISTEN"],
+            capture_output=True, text=True, timeout=5,
+        )
+    except (FileNotFoundError, subprocess.SubprocessError):
+        return []
+    pids: list[int] = []
+    for token in result.stdout.split():
+        try:
+            pids.append(int(token))
+        except ValueError:
+            pass
+    return pids
+
+
+def stop_pid(pid: int, timeout: float = 10.0) -> bool:
+    """PID とそのプロセスグループへ SIGTERM→（猶予後）SIGKILL を送って停止する。
+
+    停止を試みたら True、対象が既にいなければ False（POSIX 以外も False）。
+    """
+    if not _POSIX:
+        return False
+    try:
+        pgid = os.getpgid(pid)
+    except ProcessLookupError:
+        return False
+
+    def _alive() -> bool:
+        try:
+            os.kill(pid, 0)
+            return True
+        except ProcessLookupError:
+            return False
+        except PermissionError:
+            return True
+
+    try:
+        os.killpg(pgid, signal.SIGTERM)
+    except ProcessLookupError:
+        return False
+    except PermissionError:
+        return False  # 権限が無い（他ユーザーの）プロセスには手を出さない
+    deadline = time.monotonic() + timeout
+    while time.monotonic() < deadline:
+        if not _alive():
+            return True
+        time.sleep(0.2)
+    try:
+        os.killpg(pgid, signal.SIGKILL)  # 猶予内に終わらなければ強制終了
+    except (ProcessLookupError, PermissionError):
+        pass
+    return True
+
+
+@dataclass
+class ServerConfig:
+    """起動するローカルLLMサーバーの設定。"""
+
+    backend: str  # "mlx" | "llama-cpp"
+    model: str
+    host: str = "127.0.0.1"
+    port: int = 8080
+    parallel: int | None = None  # 同時処理スロット数（llama.cpp のみ）
+    disable_thinking: bool = False  # Qwen3 系の思考モードを無効化して起動
+    # 投機的デコード（speculative decoding）用ドラフター。今回は公式対応する
+    # Gemma 4 の MTP（Multi-Token Prediction）ドラフターに限定する。
+    # draft_model にドラフターの HF id / パス（例
+    # mlx-community/gemma-4-E4B-it-qat-assistant-bf16）を指定すると、本体の出力を
+    # 変えずに高速化する。"auto" にすると本体名から対応ドラフターを自動選択する
+    # （MTP_DRAFTERS の対応表）。指定時は本体＋ドラフターの2モデルを mlx-vlm が
+    # 初回に自動ダウンロードする。MTP は vision 対応の mlx-vlm バックエンドのみ対応。
+    draft_model: str | None = None
+    extra_args: list[str] = field(default_factory=list)
+
+    @property
+    def base_url(self) -> str:
+        return f"http://{self.host}:{self.port}/v1"
+
+
+def parallel_supported(backend: str) -> bool:
+    """そのバックエンドが並列スロット指定に対応するか。"""
+    return backend == "llama-cpp"
+
+
+# 本体（target）→ 対応する MTP ドラフター（assistant）の内蔵対応表。
+# mlx-community のペアで、いずれも実機で検証済み。draft_model = "auto" のときに
+# 本体名から対応ドラフターを引く（明示指定すればここを介さない）。未収載のモデルを
+# auto にした場合はエラーで明示指定を促す（MTP 自体は非収載でも明示すれば使える）。
+# Gemma 4 が中心だが、Qwen3.6 も MTP 方式で動作確認済み（mlx_vlm --draft-kind mtp）。
+MTP_DRAFTERS = {
+    "mlx-community/gemma-4-E4B-it-qat-4bit":
+        "mlx-community/gemma-4-E4B-it-qat-assistant-bf16",
+    "mlx-community/gemma-4-12B-it-qat-4bit":
+        "mlx-community/gemma-4-12B-it-qat-assistant-4bit",
+    "mlx-community/gemma-4-26B-A4B-it-qat-4bit":
+        "mlx-community/gemma-4-26B-A4B-it-qat-assistant-nvfp4",
+    "mlx-community/gemma-4-31B-it-qat-4bit":
+        "mlx-community/gemma-4-31B-it-qat-assistant-bf16",
+    # 非QAT 8bit（26B-A4B）。ドラフターは非QAT の assistant-bf16。
+    "mlx-community/gemma-4-26b-a4b-it-8bit":
+        "mlx-community/gemma-4-26B-A4B-it-assistant-bf16",
+    # Qwen3.6-27B（既定モデル）の MTP ドラフター。実測 ~2倍速（38→75 tok/s, 採択93%）。
+    "mlx-community/Qwen3.6-27B-4bit":
+        "mlx-community/Qwen3.6-27B-MTP-4bit",
+}
+
+
+def resolve_drafter(model: str, draft_model: str | None) -> str | None:
+    """draft_model を解決する。
+
+    - None / 空 … ドラフター無し（投機的デコードを使わない）。
+    - "auto"   … 本体名 model から対応する MTP ドラフター（Gemma 4 / Qwen3.6）を
+                 内蔵表で引く。未収載なら ValueError（HF id を明示するよう促す）。
+    - それ以外 … その値（ドラフターの HF id / パス）をそのまま使う。
+    """
+    if not draft_model:
+        return None
+    if draft_model != "auto":
+        return draft_model
+    drafter = MTP_DRAFTERS.get(model)
+    if drafter is None:
+        known = ", ".join(sorted(MTP_DRAFTERS))
+        raise ValueError(
+            f'draft_model="auto" に対応するドラフターが見つかりません（model={model!r}）。'
+            f" 自動対応している本体: {known}。"
+            " 他のモデルでは draft_model にドラフターの HF id を明示してください。"
+        )
+    return drafter
+
+
+def build_command(config: ServerConfig) -> list[str]:
+    """バックエンドに応じた起動コマンドを組み立てる。
+
+    いずれも OpenAI 互換サーバーを立ち上げる:
+      - mlx       : mlx_lm.server（テキスト専用。逐次処理。並列スロットの概念なし）
+      - mlx-vlm   : mlx_vlm.server（vision 対応。画像入力 image_url を受けられる）
+      - llama-cpp : llama-server（--parallel で並列スロットを確保）
+    """
+    if config.backend == "mlx":
+        command = [
+            "mlx_lm.server",
+            "--model", config.model,
+            "--host", config.host,
+            "--port", str(config.port),
+        ]
+        if config.disable_thinking:
+            command += ["--chat-template-args", '{"enable_thinking": false}']
+    elif config.backend == "mlx-vlm":
+        # mlx_vlm の OpenAI 互換サーバー（vision 対応）。コンソールスクリプトが
+        # 無い環境もあるため `python -m` で確実に起動する。逐次処理で並列スロットの
+        # 概念はないため --parallel は渡さない。thinking はサーバー既定が OFF
+        # （--enable-thinking を渡さなければ無効）であり、リクエスト毎の明示制御は
+        # クライアントがトップレベル enable_thinking で行う（llm.py 参照）。
+        command = [
+            sys.executable, "-m", "mlx_vlm.server",
+            "--model", config.model,
+            "--host", config.host,
+            "--port", str(config.port),
+        ]
+        # Gemma 4 の MTP ドラフターによる投機的デコード。draft_kind は mtp に固定する
+        # （他種別＝dflash / eagle3 は今回は対象外）。draft_model="auto" は本体名から
+        # 対応ドラフターを自動選択する。指定があれば本体とドラフターの両方を mlx-vlm が
+        # 初回に自動ダウンロードする。
+        drafter = resolve_drafter(config.model, config.draft_model)
+        if drafter:
+            command += ["--draft-model", drafter, "--draft-kind", "mtp"]
+    elif config.backend == "llama-cpp":
+        command = [
+            "llama-server",
+            "-m", config.model,
+            "--host", config.host,
+            "--port", str(config.port),
+        ]
+        if config.parallel is not None:
+            command += ["--parallel", str(config.parallel)]
+        if config.disable_thinking:
+            command += ["--chat-template-kwargs", '{"enable_thinking": false}']
+    else:
+        raise ValueError(
+            f"unknown backend: {config.backend!r} (choose from {BACKENDS})"
+        )
+    return command + config.extra_args
+
+
+def is_ready(base_url: str, timeout: float = 1.0) -> bool:
+    """OpenAI互換サーバーが応答可能かを判定する。"""
+    try:
+        with urllib.request.urlopen(f"{base_url}/models", timeout=timeout) as resp:
+            return resp.status == 200
+    except (urllib.error.URLError, OSError):
+        return False
+
+
+def list_models(base_url: str, timeout: float = 5.0) -> list[str]:
+    """サーバーが公開する全モデル id を /v1/models から返す（取得失敗時は []）。
+
+    単一モデルサーバーはロード済みの1件を返す。一方、複数モデルを束ねるルーター型
+    サーバーはカタログとして多数を並べる（先頭が必ずしもアクティブとは限らない）ので、
+    モデルの提供有無は「リストに含まれるか」で判定する（→ model_available）。
+    """
+    try:
+        with urllib.request.urlopen(f"{base_url}/models", timeout=timeout) as resp:
+            data = json.loads(resp.read().decode("utf-8"))
+    except (urllib.error.URLError, OSError, ValueError):
+        return []
+    items = data.get("data") if isinstance(data, dict) else None
+    if not isinstance(items, list):
+        return []
+    out: list[str] = []
+    for it in items:
+        if isinstance(it, dict):
+            mid = it.get("id")
+            if isinstance(mid, str) and mid:
+                out.append(mid)
+    return out
+
+
+def running_model(base_url: str, timeout: float = 5.0) -> str | None:
+    """起動中サーバーの代表モデル（/v1/models の最初の id）を返す。取得不可は None。
+
+    注意: ルーター型（多モデル）サーバーでは先頭が必ずしもアクティブとは限らない。
+    設定モデルが使えるかの判定には model_available（リスト全体を見る）を使うこと。
+    """
+    models = list_models(base_url, timeout)
+    return models[0] if models else None
+
+
+def model_available(base_url: str, model: str | None, timeout: float = 5.0) -> bool | None:
+    """設定モデル model がサーバーで提供されているかを返す。
+
+    - True  … /v1/models のいずれかと一致（単一モデル一致／ルーターのカタログに存在）
+    - False … モデル一覧は取れたが、その中に一致が無い
+    - None  … 判定不能（model 未指定、または一覧を取得できない）→ 警告しない
+    """
+    if not model:
+        return None
+    models = list_models(base_url, timeout)
+    if not models:
+        return None
+    return any(models_match(m, model) for m in models)
+
+
+def models_match(a: str | None, b: str | None) -> bool:
+    """2つのモデル名が同じものを指すかを大まかに判定する。
+
+    パス指定とリポジトリ名のゆれ（例 /abs/path/Foo と org/Foo）を吸収するため、
+    末尾要素（basename）を小文字で比較する。
+    """
+    if not a or not b:
+        return True  # どちらか不明なら警告しない（誤検知を避ける）
+    if a == b:
+        return True
+    base = lambda s: s.rstrip("/").split("/")[-1].lower()  # noqa: E731
+    return base(a) == base(b)
+
+
+def parse_host_port(base_url: str, default_port: int = 8080) -> tuple[str, int]:
+    """base_url（例 http://127.0.0.1:8080/v1）から host と port を取り出す。"""
+    parsed = urllib.parse.urlparse(base_url)
+    return parsed.hostname or "127.0.0.1", parsed.port or default_port
+
+
+class LocalServer:
+    """ローカルLLMサーバーをサブプロセスとして管理する。"""
+
+    def __init__(self, config: ServerConfig, log_path: str | None = None) -> None:
+        self.config = config
+        self._proc: subprocess.Popen | None = None
+        self._log_file = None
+        # サーバーの大量ログ（INFO/Stream finished 等）で対話画面が乱れないよう、
+        # 標準出力・標準エラーはこのログファイルへ逃がす（端末には流さない）。
+        self.log_path = log_path
+
+    @property
+    def base_url(self) -> str:
+        return self.config.base_url
+
+    def start(self) -> None:
+        if self._proc is not None:
+            raise RuntimeError("server already started")
+        if self.log_path is None:
+            fd, self.log_path = tempfile.mkstemp(
+                prefix="local-llm-server-", suffix=".log"
+            )
+            os.close(fd)
+        try:
+            self._log_file = open(self.log_path, "a", encoding="utf-8")
+            self._proc = subprocess.Popen(
+                build_command(self.config),
+                stdout=self._log_file,
+                stderr=subprocess.STDOUT,
+                # 子を独立したプロセスグループにして、停止時に孫まで一括終了できるようにする
+                # （POSIX のみ。Windows では無視される）。stop() の killpg と対になる。
+                start_new_session=_POSIX,
+            )
+        except FileNotFoundError as exc:
+            self._close_log()
+            raise RuntimeError(
+                f"バックエンド実行ファイルが見つかりません: {exc.filename}。"
+                " mlx_lm / mlx_vlm / llama.cpp がインストール・PATH 上にあるか確認してください。"
+            ) from exc
+
+    def _close_log(self) -> None:
+        if self._log_file is not None:
+            try:
+                self._log_file.close()
+            except OSError:
+                pass
+            self._log_file = None
+
+    def wait_until_ready(self, timeout: float = 120.0, interval: float = 1.0) -> None:
+        deadline = time.monotonic() + timeout
+        while time.monotonic() < deadline:
+            if self._proc is not None and self._proc.poll() is not None:
+                raise RuntimeError(
+                    f"server exited early (code {self._proc.returncode})"
+                )
+            if is_ready(self.config.base_url):
+                return
+            time.sleep(interval)
+        raise TimeoutError(
+            f"server not ready within {timeout}s at {self.config.base_url}"
+        )
+
+    def wait(self) -> int:
+        """サーバープロセスが終了するまでブロックする。"""
+        if self._proc is None:
+            raise RuntimeError("server not started")
+        return self._proc.wait()
+
+    def stop(self) -> None:
+        if self._proc is None:
+            self._close_log()
+            return
+        proc = self._proc
+        # graceful: プロセスグループ全体へ SIGTERM を送って 10 秒待つ
+        _signal_process_tree(proc, kill=False)
+        try:
+            proc.wait(timeout=10)
+        except subprocess.TimeoutExpired:
+            # 終わらなければグループ全体へ SIGKILL で強制終了
+            _signal_process_tree(proc, kill=True)
+            try:
+                proc.wait(timeout=5)
+            except subprocess.TimeoutExpired:
+                pass
+        self._proc = None
+        self._close_log()
+
+    def __enter__(self) -> "LocalServer":
+        self.start()
+        self.wait_until_ready()
+        return self
+
+    def __exit__(self, *exc: object) -> None:
+        self.stop()
+
+
+def build_pool_configs(base: ServerConfig, instances: int) -> list[ServerConfig]:
+    """base を起点に、連番ポート（port, port+1, ...）の設定を instances 個作る。"""
+    return [replace(base, port=base.port + i) for i in range(instances)]
+
+
+class ServerPool:
+    """複数のローカルLLMサーバーをまとめて起動・管理する。
+
+    mlx のように1プロセスが逐次処理のバックエンドで並列性を得る用途を想定。
+    各インスタンスは別ポートで起動し、それぞれモデルを個別にロードする
+    （= 重みのメモリはインスタンス数分かかる）。
+    """
+
+    def __init__(self, configs: list[ServerConfig]) -> None:
+        self._servers = [LocalServer(c) for c in configs]
+
+    @property
+    def base_urls(self) -> list[str]:
+        return [server.base_url for server in self._servers]
+
+    def start(self) -> None:
+        for server in self._servers:
+            server.start()
+
+    def wait_until_ready(self, timeout: float = 120.0) -> None:
+        for server in self._servers:
+            server.wait_until_ready(timeout=timeout)
+
+    def wait(self) -> None:
+        for server in self._servers:
+            server.wait()
+
+    def stop(self) -> None:
+        for server in self._servers:
+            server.stop()
+
+    def __enter__(self) -> "ServerPool":
+        self.start()
+        self.wait_until_ready()
+        return self
+
+    def __exit__(self, *exc: object) -> None:
+        self.stop()

local_llm_server-0.1.0.dist-info/METADATA

@@ -0,0 +1,68 @@
+Metadata-Version: 2.4
+Name: local-llm-server
+Version: 0.1.0
+Summary: ローカルLLM（mlx / mlx-vlm / llama.cpp / router）を OpenAI 互換 API として起動・管理する軽量サーバー
+Author: ToPo-ToPo-ToPo
+License-Expression: Apache-2.0
+Project-URL: Repository, https://github.com/ToPo-ToPo-ToPo/local-llm-server
+Keywords: llm,local-llm,mlx,mlx-vlm,llama.cpp,openai,server,inference
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Developers
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Topic :: Software Development
+Classifier: Operating System :: OS Independent
+Requires-Python: >=3.11
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Provides-Extra: mlx
+Requires-Dist: mlx-lm>=0.31.3; (sys_platform == "darwin" and platform_machine == "arm64") and extra == "mlx"
+Requires-Dist: mlx-vlm>=0.5.0; (sys_platform == "darwin" and platform_machine == "arm64") and extra == "mlx"
+Dynamic: license-file
+
+# local-llm-server
+
+ローカルLLM（**mlx** / **mlx-vlm** / **llama.cpp**）を **OpenAI 互換 API** として
+起動・管理する軽量サーバー。テキストと画像（vision）を自動で振り分ける
+**router** モードを備え、任意の OpenAI 互換クライアントからそのまま利用できる。
+
+- コア機能（プロセスの起動・監視・graceful shutdown・router プロキシ）は
+  **標準ライブラリのみ**で動作。
+- 実際の推論バックエンドは extras で導入（Apple Silicon では `mlx` を自動選択）。
+
+## インストール
+
+```bash
+pip install local-llm-server          # コア（バックエンドは別途用意）
+pip install "local-llm-server[mlx]"   # Apple Silicon 向け mlx / mlx-vlm 同梱
+```
+
+## 使い方
+
+```bash
+# テキストLLMを起動（既定バックエンドは環境に応じて自動選択）
+local-llm-server --backend mlx
+
+# 画像入力対応（vision）
+local-llm-server --backend mlx-vlm
+
+# テキストLLMとVLMを同時起動し、リクエスト内容で自動振り分け
+local-llm-server --backend router
+```
+
+起動後、表示される `base_url`（例 `http://127.0.0.1:8080/v1`）を
+OpenAI 互換クライアントに設定する。
+
+```python
+import local_llm_server as srv
+
+config = srv.ServerConfig(model="mlx-community/Qwen3.6-27B-4bit", backend="mlx")
+with srv.LocalServer(config) as server:
+    server.wait_until_ready()
+    print(srv.list_models(server.base_url))
+```
+
+## ライセンス
+
+Apache-2.0

local_llm_server-0.1.0.dist-info/RECORD

@@ -0,0 +1,11 @@
+local_llm_server/__init__.py,sha256=LJGIg6QjQQgK0eR4hh1X1Wf2RJyj1c7ZHFsJoTMAMcs,887
+local_llm_server/cli.py,sha256=Ss2ZKaPp1Lv1iX2wSxgrphHH0h9Ec-h7eszo8cOisIk,10338
+local_llm_server/constants.py,sha256=u5i6eKTMMgjmzD9uqfA0V7TNb15EDeoWt4VLDgtSKS4,1127
+local_llm_server/router.py,sha256=kRDdSVIO8LJxlIsYhiNSYFTJj_tmNY3X7MqWaM3AZ-8,5485
+local_llm_server/server.py,sha256=f_9IrVdv9HcJX0cak5q9UWwRubK3KMhL-vLgrU_erjY,20795
+local_llm_server-0.1.0.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+local_llm_server-0.1.0.dist-info/METADATA,sha256=AY7Du5iQxBYjcwywAnuCUwOJenfWshUr83AGT6w0Rlw,2592
+local_llm_server-0.1.0.dist-info/WHEEL,sha256=aeYiig01lYGDzBgS8HxWXOg3uV61G9ijOsup-k9o1sk,91
+local_llm_server-0.1.0.dist-info/entry_points.txt,sha256=ucTzyiXpSFkjlzhAMYE2ACKI6Yw69a9ZHQex5_74gYA,63
+local_llm_server-0.1.0.dist-info/top_level.txt,sha256=zKFVIO_D2dl7AzNgAyiF8NSCmjG0jG26HgxX7fPG9Cs,17
+local_llm_server-0.1.0.dist-info/RECORD,,

local_llm_server-0.1.0.dist-info/WHEEL

@@ -0,0 +1,5 @@
+Wheel-Version: 1.0
+Generator: setuptools (82.0.1)
+Root-Is-Purelib: true
+Tag: py3-none-any
+

local_llm_server-0.1.0.dist-info/entry_points.txt

@@ -0,0 +1,2 @@
+[console_scripts]
+local-llm-server = local_llm_server.cli:main

local_llm_server-0.1.0.dist-info/licenses/LICENSE

@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

local_llm_server-0.1.0.dist-info/top_level.txt

@@ -0,0 +1 @@
+local_llm_server