technode-cli 0.1.0__py3-none-any.whl → 0.2.0__py3-none-any.whl

technode/__init__.py

@@ -1,3 +1,3 @@
 """TechNode CLI — run inference on TechNode's compressed open-model GPU grid."""
 
-__version__ = "0.1.0"
+__version__ = "0.2.0"

technode/cli.py

@@ -403,6 +403,10 @@ def build_parser() -> argparse.ArgumentParser:
 
     gpu.set_defaults(func=lambda a: (gpu.print_help() or 0))
 
+    # provider: run a GPU as a serving node (separate concern from consuming)
+    from . import provider as _provider
+    _provider.add_parser(sub)
+
     return p
 
 

technode/provider.py

@@ -0,0 +1,248 @@
+"""TechNode provider commands — turn a Linux box into a grid serving node.
+
+`technode provider serve` runs the stdlib-only serving daemon (pc_serve.py) in
+*pull mode*: it polls the broker over outbound HTTPS for jobs it can serve, runs
+them on the local GPU, and posts results back. No inbound port, no Tailscale —
+works behind any NAT. This is the cross-platform / marketplace path.
+
+    technode provider register --gpu "RTX 4090" --vram 24
+    technode provider serve --llama-server /opt/llama.cpp/llama-server
+    technode provider status
+
+Provider ops talk to the broker directly (long-poll doesn't fit a serverless
+proxy's time limit). Override with TN_BROKER.
+"""
+
+import json
+import os
+import shutil
+import subprocess
+import sys
+import urllib.error
+import urllib.request
+
+BROKER_URL = (os.environ.get("TN_BROKER", "").strip()
+              or "https://broker.technode.network").rstrip("/")
+PC_SERVE_URL = "https://technode.network/agent/pc_serve.py"
+TN_DIR = os.path.join(os.path.expanduser("~"), ".technode")
+PROVIDER_CFG = os.path.join(TN_DIR, "provider.json")
+PC_SERVE_PATH = os.path.join(TN_DIR, "pc_serve.py")
+MODELS_DIR = os.path.join(TN_DIR, "models")
+LLAMA_DIR = os.path.join(TN_DIR, "llama")
+UA = "technode-cli-provider"
+
+
+def _die(msg, code=1):
+    print("technode: " + msg, file=sys.stderr)
+    raise SystemExit(code)
+
+
+def _req(method, url, body=None, timeout=30):
+    data = json.dumps(body).encode() if body is not None else None
+    headers = {"User-Agent": UA, "Accept": "application/json"}
+    if data is not None:
+        headers["Content-Type"] = "application/json"
+    req = urllib.request.Request(url, data=data, method=method, headers=headers)
+    try:
+        with urllib.request.urlopen(req, timeout=timeout) as r:
+            return json.loads(r.read().decode() or "{}")
+    except urllib.error.HTTPError as e:
+        raw = e.read().decode("utf-8", "replace")
+        try:
+            payload = json.loads(raw)
+        except ValueError:
+            payload = {"error": raw[:300] or e.reason}
+        payload["_status"] = e.code
+        return payload
+    except urllib.error.URLError as e:
+        _die(f"cannot reach broker {BROKER_URL} — {e.reason}")
+
+
+def _load_cfg():
+    try:
+        with open(PROVIDER_CFG, encoding="utf-8") as fh:
+            return json.load(fh)
+    except (FileNotFoundError, ValueError):
+        return {}
+
+
+def _save_cfg(cfg):
+    os.makedirs(TN_DIR, exist_ok=True)
+    tmp = PROVIDER_CFG + ".tmp"
+    with open(tmp, "w", encoding="utf-8") as fh:
+        json.dump(cfg, fh, indent=2)
+    os.replace(tmp, PROVIDER_CFG)
+    try:
+        os.chmod(PROVIDER_CFG, 0o600)
+    except OSError:
+        pass
+
+
+# --------------------------------------------------------------------------- #
+def cmd_register(args):
+    import socket
+    body = {
+        "name": args.name or socket.gethostname(),
+        "hostname": socket.gethostname(),
+        "gpu_name": args.gpu or "",
+        "vram_gb": int(args.vram or 0),
+        "owner_email": args.email or "",
+    }
+    existing = _load_cfg()
+    if existing.get("register_token"):
+        # Re-register: prove ownership so the broker refreshes rather than rejects.
+        body["provider_id"] = existing.get("provider_id", "")
+        body["register_token"] = existing["register_token"]
+    res = _req("POST", BROKER_URL + "/provider/register", body)
+    if res.get("error"):
+        _die(f"register failed — {res.get('error')}")
+    cfg = {
+        "provider_id": res.get("provider_id"),
+        "register_token": res.get("register_token"),
+        "dashboard_token": res.get("dashboard_token"),
+        "broker": BROKER_URL,
+    }
+    _save_cfg(cfg)
+    print("Registered ✓")
+    print(f"  provider_id: {cfg['provider_id']}")
+    print(f"  dashboard:   {res.get('dashboard_url', '')}")
+    print(f"  creds saved: {PROVIDER_CFG} (chmod 600)")
+    print("\nNext:  technode provider serve   (needs operator approval to receive jobs)")
+    return 0
+
+
+def cmd_status(args):
+    cfg = _load_cfg()
+    if not cfg.get("provider_id"):
+        print("Not registered. Run `technode provider register`.")
+        return 1
+    print(f"provider_id: {cfg['provider_id']}")
+    print(f"broker:      {cfg.get('broker', BROKER_URL)}")
+    me = _req("GET", BROKER_URL + f"/provider/me?t={cfg.get('dashboard_token','')}", timeout=15)
+    if me.get("error"):
+        print(f"approval:    unknown ({me.get('error')})")
+    else:
+        print(f"approved:    {me.get('approved')}")
+        if me.get("serving_models"):
+            print(f"models:      {', '.join(me['serving_models'])}")
+    llama = _find_llama(args.llama_server if hasattr(args, "llama_server") else None)
+    print(f"llama-server: {llama or 'NOT FOUND (set --llama-server or install llama.cpp)'}")
+    return 0
+
+
+def _find_llama(explicit=None):
+    for cand in (explicit, os.environ.get("TN_LLAMA_BIN"),
+                 shutil.which("llama-server"),
+                 os.path.join(LLAMA_DIR, "llama-server"),
+                 os.path.join(LLAMA_DIR, "build", "bin", "llama-server")):
+        if cand and os.path.isfile(cand) and os.access(cand, os.X_OK):
+            return cand
+    return None
+
+
+def _ensure_pc_serve():
+    if os.path.isfile(PC_SERVE_PATH):
+        return PC_SERVE_PATH
+    os.makedirs(TN_DIR, exist_ok=True)
+    print(f"downloading serving daemon → {PC_SERVE_PATH}")
+    req = urllib.request.Request(PC_SERVE_URL, headers={"User-Agent": UA})
+    try:
+        with urllib.request.urlopen(req, timeout=30) as r:
+            data = r.read()
+    except Exception as e:
+        _die(f"could not download pc_serve.py from {PC_SERVE_URL} — {e}")
+    with open(PC_SERVE_PATH, "wb") as fh:
+        fh.write(data)
+    return PC_SERVE_PATH
+
+
+def cmd_serve(args):
+    cfg = _load_cfg()
+    if not cfg.get("register_token"):
+        _die("not registered. Run `technode provider register` first.")
+    llama = _find_llama(args.llama_server)
+    if not llama:
+        _die("llama-server not found.\n"
+             "        Point to it:  technode provider serve --llama-server /path/to/llama-server\n"
+             "        or set TN_LLAMA_BIN, or put it in ~/.technode/llama/.\n"
+             "        NVIDIA build: https://github.com/ggml-org/llama.cpp/releases "
+             "(or build with -DGGML_CUDA=ON).")
+    pc_serve = _ensure_pc_serve()
+    os.makedirs(MODELS_DIR, exist_ok=True)
+    cmd = [sys.executable, pc_serve, "--pull",
+           "--provider-id", cfg["provider_id"],
+           "--register-token", cfg["register_token"],
+           "--bin", llama, "--models", MODELS_DIR]
+    if args.models:
+        cmd += ["--serve-models", args.models]
+    if args.no_inbound:
+        # pull mode doesn't need the inbound server; bind it to localhost only is
+        # not exposed via a flag, so we just note it. (Harmless when unreachable.)
+        pass
+    print(f"starting pull-mode serving: provider={cfg['provider_id']}  llama={llama}")
+    print(f"  broker={BROKER_URL}  models-cache={MODELS_DIR}")
+    print("  (Ctrl-C to stop)\n")
+    env = dict(os.environ, TN_BROKER=BROKER_URL)
+    try:
+        return subprocess.call(cmd, env=env)
+    except KeyboardInterrupt:
+        return 130
+
+
+def cmd_install(args):
+    """Emit a systemd unit that runs `technode provider serve` on boot."""
+    cfg = _load_cfg()
+    if not cfg.get("register_token"):
+        _die("register first: technode provider register")
+    tn = shutil.which("technode") or os.path.join(os.path.dirname(sys.executable), "technode")
+    llama = _find_llama(args.llama_server) or "/path/to/llama-server"
+    user = os.environ.get("USER", "root")
+    unit = f"""[Unit]
+Description=TechNode provider (pull-mode serving)
+After=network-online.target
+Wants=network-online.target
+
+[Service]
+Type=simple
+User={user}
+ExecStart={tn} provider serve --llama-server {llama}
+Restart=always
+RestartSec=5
+Environment=TN_BROKER={BROKER_URL}
+
+[Install]
+WantedBy=multi-user.target
+"""
+    print("# Save as /etc/systemd/system/technode-provider.service, then:")
+    print("#   sudo systemctl daemon-reload && sudo systemctl enable --now technode-provider")
+    print("# ----------------------------------------------------------------")
+    print(unit)
+    return 0
+
+
+def add_parser(sub):
+    p = sub.add_parser("provider", help="run a GPU as a grid serving node (Linux/cross-platform)")
+    psub = p.add_subparsers(dest="provider_command")
+
+    s = psub.add_parser("register", help="register this machine as a provider")
+    s.add_argument("--name", help="display name (default: hostname)")
+    s.add_argument("--gpu", help="GPU name, e.g. \"RTX 4090\"")
+    s.add_argument("--vram", help="GPU VRAM in GB")
+    s.add_argument("--email", help="owner email (optional)")
+    s.set_defaults(func=cmd_register)
+
+    s = psub.add_parser("serve", help="serve models in pull mode (outbound-only, NAT-friendly)")
+    s.add_argument("--llama-server", help="path to the llama-server binary")
+    s.add_argument("--models", help="comma-separated catalog ids to advertise (default: auto by VRAM)")
+    s.add_argument("--no-inbound", action="store_true", help="pull only (no inbound serving)")
+    s.set_defaults(func=cmd_serve)
+
+    s = psub.add_parser("status", help="show registration + approval + llama-server")
+    s.add_argument("--llama-server", help="path to the llama-server binary")
+    s.set_defaults(func=cmd_status)
+
+    s = psub.add_parser("install", help="print a systemd unit for boot persistence")
+    s.add_argument("--llama-server", help="path to the llama-server binary")
+    s.set_defaults(func=cmd_install)
+
+    p.set_defaults(func=lambda a: (p.print_help() or 0))

{technode_cli-0.1.0.dist-info → technode_cli-0.2.0.dist-info}/METADATA

@@ -1,7 +1,7 @@
 Metadata-Version: 2.4
 Name: technode-cli
-Version: 0.1.0
-Summary: TechNode CLI — run inference on TechNode's compressed open-model GPU grid (RunPod-compatible).
+Version: 0.2.0
+Summary: TechNode CLI — run inference on compressed open models, rent GPUs, or serve as a provider (RunPod-compatible).
 Author-email: TechNode <smlee3636@gmail.com>
 License: MIT
 Project-URL: Homepage, https://technode.network
@@ -35,6 +35,22 @@ Zero dependencies — pure Python stdlib, runs anywhere Python ≥3.8 does.
 | `technode models [--json]` | List available models (id, quantization, role). |
 | `technode infer PROMPT [-m MODEL] [-n MAX_TOKENS] [-t TEMP] [--json] [-q]` | Text generation. `-` or piped stdin reads the prompt from stdin. |
 | `technode whoami` | Show the active key (masked) + endpoint. |
+| `technode gpu lease/list/status/release` | Rent a whole GPU (Jupyter lab session). |
+
+## Become a provider (share your GPU)
+
+Got an NVIDIA Linux box? Join the grid and serve models — **outbound-only, works
+behind any NAT** (no Tailscale, no inbound ports):
+
+```bash
+technode provider register --gpu "RTX 4090" --vram 24
+technode provider serve --llama-server /path/to/llama-server   # pull-mode worker
+technode provider status
+```
+
+`serve` polls the broker for jobs it can run, executes them on your GPU, and
+returns the results. Needs a llama.cpp `llama-server` binary (CUDA build for
+NVIDIA) and operator approval before it receives live jobs.
 
 ## Configuration
 

technode_cli-0.2.0.dist-info/RECORD

@@ -0,0 +1,10 @@
+technode/__init__.py,sha256=t8NTvxRnioJRH5gE2COIcS3sHfgge-pZ4ObN0WPnETI,106
+technode/__main__.py,sha256=hBG70kNbP-Vixov5nkJkcoh8Aws0E1a3MJjNDacxRP4,136
+technode/cli.py,sha256=JzEqvFDAh5Lv0ewyyv7Po7eAqrTuJlZ87_mOl2SFETQ,15337
+technode/provider.py,sha256=yvFbVTpQ6WERjwHl_xIlfGUd2ShtNlruaPeVJUb5YAo,9595
+technode_cli-0.2.0.dist-info/licenses/LICENSE,sha256=dUZBEi7BA8K033XkmDt5oHuhCaG2QcmbZLBv5jPw9zI,1065
+technode_cli-0.2.0.dist-info/METADATA,sha256=20Sy3-9N1mBoy6zh8K7zQPGU4OOkbg41SdSATrTwiU8,2676
+technode_cli-0.2.0.dist-info/WHEEL,sha256=aeYiig01lYGDzBgS8HxWXOg3uV61G9ijOsup-k9o1sk,91
+technode_cli-0.2.0.dist-info/entry_points.txt,sha256=e2IUIHePhD2EYqlsMua7TWWQCmtTN_etV95FTZhQ0fo,47
+technode_cli-0.2.0.dist-info/top_level.txt,sha256=PqqGyKCWgMCEx78tqZkiXmq4Y6qkkAm1KP7WA1NSfnM,9
+technode_cli-0.2.0.dist-info/RECORD,,

technode_cli-0.1.0.dist-info/RECORD

@@ -1,9 +0,0 @@
-technode/__init__.py,sha256=o5xSTehxhArSIzNV7FrRrir6SY_hl1nLnuU8RLW9veQ,106
-technode/__main__.py,sha256=hBG70kNbP-Vixov5nkJkcoh8Aws0E1a3MJjNDacxRP4,136
-technode/cli.py,sha256=34flilgXxBf_UT67HLuED-1NmYebQRyPAvlfhPVcOHU,15188
-technode_cli-0.1.0.dist-info/licenses/LICENSE,sha256=dUZBEi7BA8K033XkmDt5oHuhCaG2QcmbZLBv5jPw9zI,1065
-technode_cli-0.1.0.dist-info/METADATA,sha256=Tb6pVdlyVgZc1U2yawW_Net3mpos46gDxG_AEYjX1xI,2010
-technode_cli-0.1.0.dist-info/WHEEL,sha256=aeYiig01lYGDzBgS8HxWXOg3uV61G9ijOsup-k9o1sk,91
-technode_cli-0.1.0.dist-info/entry_points.txt,sha256=e2IUIHePhD2EYqlsMua7TWWQCmtTN_etV95FTZhQ0fo,47
-technode_cli-0.1.0.dist-info/top_level.txt,sha256=PqqGyKCWgMCEx78tqZkiXmq4Y6qkkAm1KP7WA1NSfnM,9
-technode_cli-0.1.0.dist-info/RECORD,,