lyceum-cli 1.2.2__tar.gz → 1.2.3__tar.gz

{lyceum_cli-1.2.2 → lyceum_cli-1.2.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: lyceum-cli
-Version: 1.2.2
+Version: 1.2.3
 Summary: Command-line interface for Lyceum Cloud
 Author-email: Lyceum <support@lyceum.technology>
 License: MIT

{lyceum_cli-1.2.2 → lyceum_cli-1.2.3}/lyceum/_version.py

@@ -7,4 +7,4 @@ __version__ lives in this module. A future refactor could move these modules
 under lyceum/cli/, claiming one subpackage the way the other two do.
 """
 
-__version__ = "1.2.2"
+__version__ = "1.2.3"

{lyceum_cli-1.2.2 → lyceum_cli-1.2.3}/lyceum/external/compute/inference/claude_code.py

@@ -19,7 +19,7 @@ console = Console()
 # or ~/.claude settings on this machine.
 CLAUDE_PROFILE_DIR = Path.home() / ".lyceum" / "claude"
 
-DEFAULT_MODEL = "moonshotai/kimi-k3"
+DEFAULT_MODEL = "moonshotai/kimi-k3.1-pro"
 DEFAULT_FAST_MODEL = "z-ai/glm-5.3-flash"
 
 # Use explicit Lyceum IDs so the backend never has to guess a substitute.

{lyceum_cli-1.2.2 → lyceum_cli-1.2.3}/lyceum/external/compute/inference/infer.py

@@ -74,7 +74,7 @@ def _is_dedicated_inference_hidden() -> bool:
 
 def _fetch_dedicated_models() -> list[dict]:
     """Serverless models a dedicated deployment may run, on their default GPU shape."""
-    return api_request("GET", "/api/v2/external/inference/models", timeout=20.0).get("models", [])
+    return api_request("GET", "/external/inference/models", timeout=20.0).get("models", [])
 
 
 def _format_gpus(gpu_type: str, gpu_count: int) -> str:
@@ -181,14 +181,27 @@ def deploy_model(
     target_latency: float = typer.Option(None, "--target-latency", hidden=True),
     stabilisation: int = typer.Option(None, "--stabilisation", hidden=True),
     gpu_count: int = typer.Option(
-        None, "--gpu-count", help="GPUs per replica, defaults per model (also sets --tensor-parallel-size)"
+        None,
+        "--gpu-count",
+        help="GPUs per replica, defaults per model (also sets --tensor-parallel-size)",
+    ),
+    offload: bool = typer.Option(
+        None,
+        "--offload/--no-offload",
+        help=(
+            "Serve requests the deployment cannot from Lyceum's serverless capacity (default off). That may include a "
+            "third-party provider under Lyceum's data terms unless your org is restricted to Lyceum's own"
+        ),
     ),
     yes: bool = typer.Option(False, "--yes", "-y", help="Do not ask before creating"),
     wait: bool = typer.Option(False, "--wait", "-w", help="Wait until deployment has healthy replicas"),
 ):
     """Deploy a serverless model on dedicated GPUs."""
     models = _fetch_dedicated_models()
-    model = next((m for m in models if model_id.lower() in (m["serving_name"].lower(), m["hf_model_id"].lower())), None)
+    model = next(
+        (m for m in models if model_id.lower() in (m["serving_name"].lower(), m["hf_model_id"].lower())),
+        None,
+    )
     if model is None:
         console.print(f"[red]{model_id} cannot be deployed on dedicated GPUs.[/red]")
         console.print("[dim]See the models that can: lyceum infer models --dedicated[/dim]")
@@ -222,7 +235,10 @@ def deploy_model(
             raise typer.Exit(0)
 
     if hf_token is None and not yes and not (os.getenv("HF_TOKEN") or os.getenv("HUGGING_FACE_HUB_TOKEN")):
-        if typer.confirm("No HF token detected. Provide one (required for gated models)?", default=False):
+        if typer.confirm(
+            "No HF token detected. Provide one (required for gated models)?",
+            default=False,
+        ):
             hf_token = typer.prompt("HuggingFace Token", hide_input=True)
 
     payload = {
@@ -233,6 +249,7 @@ def deploy_model(
     overrides = {
         "gpu_count": gpu_count,
         "gpu_type": gpu_type,
+        "offload": offload,
         "stabilisation_window": stabilisation,
         "target_latency_p95_ms": target_latency,
         "target_rps": target_rps,
@@ -241,7 +258,7 @@ def deploy_model(
     if hf_token:
         payload["hf_token"] = hf_token
 
-    data = api_request("POST", "/api/v2/external/inference/create", json=payload)
+    data = api_request("POST", "/external/inference/create", json=payload)
     deployment_id = data.get("deployment_id", "?")
 
     console.print("\n[green]Deployment created![/green]")
@@ -261,6 +278,30 @@ def deploy_model(
         _wait_for_healthy(deployment_id, timeout=model["ready_timeout_s"])
 
 
+def _offload_label(enabled: bool | None) -> str:
+    return "enabled" if enabled else "disabled"
+
+
+@infer_app.command("offload", hidden=_is_dedicated_inference_hidden())
+def set_offload(
+    deployment_id: str = typer.Argument(..., help="Deployment ID"),
+    enabled: bool = typer.Option(..., "--enable/--disable", help="Turn offloading on or off"),
+):
+    """Turn offloading on or off for a deployment.
+
+    Enabled, requests the deployment cannot serve go to Lyceum's serverless capacity,
+    which may include a third-party provider under Lyceum's data terms unless your
+    org is restricted to Lyceum's own.
+    """
+    data = api_request(
+        "POST",
+        "/external/inference/offload",
+        json={"deployment_id": deployment_id, "offload": enabled},
+    )
+    console.print(f"  Deployment ID: [cyan]{data.get('deployment_id')}[/cyan]")
+    console.print(f"  Offload:       [yellow]{_offload_label(data.get('offload'))}[/yellow]")
+
+
 @infer_app.command("stop")
 def stop_model(
     deployment_id: str = typer.Argument(..., help="Deployment ID to stop"),
@@ -268,7 +309,7 @@ def stop_model(
     """Stop a deployed model and all its replicas."""
     console.print(f"[dim]Stopping deployment {deployment_id[:12]}...[/dim]")
 
-    data = api_request("DELETE", "/api/v2/external/inference/stop", json={"deployment_id": deployment_id})
+    data = api_request("DELETE", "/external/inference/stop", json={"deployment_id": deployment_id})
     console.print("[green]Deployment stopped![/green]")
     console.print(f"  Deployment ID: [cyan]{data.get('deployment_id')}[/cyan]")
     console.print(f"  Status:        [yellow]{data.get('status')}[/yellow]")
@@ -283,7 +324,7 @@ def deployment_status(
     """Show detailed status of a deployment."""
     dep = api_request(
         "GET",
-        "/api/v2/external/inference/get",
+        "/external/inference/get",
         params={"deployment_id": deployment_id, "include_terminated": str(include_terminated).lower()},
         timeout=15.0,
     )
@@ -307,6 +348,8 @@ def deployment_status(
         console.print(f"  Target p95:     {dep['target_latency_p95_ms']} ms")
     if dep.get("stabilisation_window") is not None:
         console.print(f"  Stabilisation:  {dep['stabilisation_window']} s")
+    if dep.get("offload") is not None:
+        console.print(f"  Offload:        {_offload_label(dep['offload'])}")
 
     replicas = dep.get("replicas", [])
     if replicas:
@@ -359,7 +402,7 @@ def list_deployments(
     """List your dedicated deployments."""
     data = api_request(
         "GET",
-        "/api/v2/external/inference/list",
+        "/external/inference/list",
         params={"include_terminated": str(include_terminated).lower()},
         timeout=15.0,
     )
@@ -425,7 +468,7 @@ def replica_logs(
     replica_id: str = typer.Argument(..., help="Replica ID"),
 ):
     """Show last container logs for a failed replica."""
-    data = api_request("GET", f"/api/v2/external/inference/replica/{replica_id}/logs", timeout=15.0)
+    data = api_request("GET", f"/external/inference/replica/{replica_id}/logs", timeout=15.0)
     error_logs = data.get("error_logs")
     if not error_logs:
         console.print(f"[yellow]No logs available for replica {replica_id}[/yellow]")
@@ -445,7 +488,7 @@ def _wait_for_healthy(deployment_id: str, timeout: int = 600, poll_interval: int
         while time.time() < deadline:
             try:
                 response = httpx.get(
-                    f"{config.base_url}/api/v2/external/inference/get",
+                    f"{config.base_url}/external/inference/get",
                     headers=headers,
                     params={"deployment_id": deployment_id},
                     timeout=15.0,

{lyceum_cli-1.2.2 → lyceum_cli-1.2.3}/lyceum_cli.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: lyceum-cli
-Version: 1.2.2
+Version: 1.2.3
 Summary: Command-line interface for Lyceum Cloud
 Author-email: Lyceum <support@lyceum.technology>
 License: MIT

{lyceum_cli-1.2.2 → lyceum_cli-1.2.3}/pyproject.toml

@@ -40,6 +40,11 @@ Dashboard = "https://dashboard.lyceum.technology"
 [project.scripts]
 lyceum = "lyceum.main:app"
 
+[dependency-groups]
+test = [
+    "pytest==9.0.2",
+]
+
 [tool.setuptools.dynamic]
 version = { attr = "lyceum._version.__version__" }
 

{lyceum_cli-1.2.2 → lyceum_cli-1.2.3}/tests/test_dedicated_deploy.py

@@ -29,11 +29,14 @@ def gateway(monkeypatch):
     creates = []
 
     def api_request(method, path, **kw):
-        if path == "/api/v2/external/inference/models":
+        if path == "/external/inference/models":
             return {"models": [GLM52, FLASH]}
-        if path == "/api/v2/external/inference/create":
+        if path == "/external/inference/create":
             creates.append(kw["json"])
             return {"deployment_id": "d1", "status": "pending"}
+        if path == "/external/inference/offload":
+            creates.append(kw["json"])
+            return kw["json"]
         raise AssertionError(path)
 
     monkeypatch.setattr(infer, "api_request", api_request)
@@ -103,3 +106,24 @@ def test_wait_uses_the_model_ready_timeout(gateway, monkeypatch):
     monkeypatch.setattr(infer, "_wait_for_healthy", lambda d, timeout: waited.append((d, timeout)))
     assert _run("deploy", "deepseek/deepseek-v4-flash-0731", "--wait").exit_code == 0
     assert waited == [("d1", 4800)]
+
+
+def test_deploy_offload_is_sent_only_when_given(gateway):
+    assert _run("deploy", "deepseek/deepseek-v4-flash-0731").exit_code == 0
+    assert "offload" not in gateway[0]
+    assert _run("deploy", "deepseek/deepseek-v4-flash-0731", "--offload").exit_code == 0
+    assert gateway[1]["offload"] is True
+    assert _run("deploy", "deepseek/deepseek-v4-flash-0731", "--no-offload").exit_code == 0
+    assert gateway[2]["offload"] is False
+
+
+def test_offload_command_posts_the_switch(gateway):
+    result = _run("offload", "d1", "--enable")
+    assert result.exit_code == 0
+    assert "enabled" in result.output
+    assert _run("offload", "d1", "--disable").exit_code == 0
+    assert gateway == [
+        {"deployment_id": "d1", "offload": True},
+        {"deployment_id": "d1", "offload": False},
+    ]
+    assert _run("offload", "d1").exit_code == 2