laya-cli 0.2.1__tar.gz → 0.2.2__tar.gz

{laya_cli-0.2.1 → laya_cli-0.2.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: laya-cli
-Version: 0.2.1
+Version: 0.2.2
 Summary: Ergonomic CLI for Laya — typed decisions for humans and AI agents (predict, batch, presets, shortlist, router)
 Project-URL: Homepage, https://github.com/MIt9/laya-cli
 Project-URL: Repository, https://github.com/MIt9/laya-cli.git

{laya_cli-0.2.1 → laya_cli-0.2.2}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "laya-cli"
-version = "0.2.1"
+version = "0.2.2"
 description = "Ergonomic CLI for Laya — typed decisions for humans and AI agents (predict, batch, presets, shortlist, router)"
 readme = "README.md"
 requires-python = ">=3.10"

{laya_cli-0.2.1 → laya_cli-0.2.2}/src/laya_cli/__init__.py

@@ -1,3 +1,3 @@
 """laya-cli — streaming JSONL classifier over Laya."""
 
-__version__ = "0.2.1"
+__version__ = "0.2.2"

{laya_cli-0.2.1 → laya_cli-0.2.2}/src/laya_cli/daemon.py

@@ -25,8 +25,58 @@ from typing import Any
 CACHE_DIR = Path.home() / ".cache" / "laya-cli" / "daemons"
 
 
+def _resolve_device(device: str | None) -> str:
+    """Resolve effective device: auto (cuda>mps>cpu) or explicit with fallback to cpu if unavailable."""
+    # explicit
+    if device is not None and str(device).strip() != "":
+        dev = str(device).strip().lower()
+        # normalize aliases
+        if dev in ("cuda", "cuda:0", "gpu"):
+            dev = "cuda"
+        elif dev == "mps":
+            dev = "mps"
+        elif dev == "cpu":
+            dev = "cpu"
+        # check availability, fallback to cpu
+        try:
+            import torch
+
+            if dev == "cuda" and not torch.cuda.is_available():
+                return "cpu"
+            if dev == "mps" and not (hasattr(torch.backends, "mps") and torch.backends.mps.is_available()):
+                return "cpu"
+        except Exception:
+            pass
+        return dev
+    # auto
+    try:
+        import torch
+
+        if torch.cuda.is_available():
+            return "cuda"
+        if hasattr(torch.backends, "mps") and torch.backends.mps.is_available():
+            return "mps"
+        return "cpu"
+    except Exception:
+        return "cpu"
+
+
+def _normalize_config(
+    model: str, subfolder: str | None, device: str | None, router: bool, lang: str | None
+) -> tuple[str, str, str, str, str]:
+    """Single place for config normalization before hashing — used by both serve and client."""
+    norm_model = (model or "convaiinnovations/laya").strip()
+    norm_subfolder = (subfolder or "").strip()
+    # treat "" and None as same (bundle root)
+    norm_device = _resolve_device(device)
+    norm_router = "1" if router else "0"
+    norm_lang = (lang or "").strip().lower()
+    return (norm_model, norm_subfolder, norm_device, norm_router, norm_lang)
+
+
 def config_hash(model: str, subfolder: str | None, device: str | None, router: bool, lang: str | None) -> str:
-    raw = "|".join([model or "", subfolder or "", device or "", "1" if router else "0", lang or ""])
+    norm = _normalize_config(model, subfolder, device, router, lang)
+    raw = "|".join(norm)
     return hashlib.sha256(raw.encode()).hexdigest()[:12]
 
 

{laya_cli-0.2.1 → laya_cli-0.2.2}/tests/test_daemon.py

@@ -241,3 +241,89 @@ def test_daemon_only_localhost(fake_laya):
     time.sleep(0.3)
     if daemon_file.exists():
         daemon_file.unlink()
+
+
+def test_daemon_config_hash_device_normalization(fake_laya):
+    """Regression for TASK2 BUG: hash must be after device resolve, not raw --device.
+
+    On MPS Mac, `serve` without --device (auto=mps) and `serve --device mps` must share file,
+    and `predict` without --device must find daemon started with --device mps.
+    """
+    from laya_cli.daemon import daemon_file_for, daemon_port_for_args
+
+    # Same effective device -> same file (auto resolves to mps on this Mac, so None and "mps" coincide)
+    f_auto = daemon_file_for("convaiinnovations/laya", None, None, False, None)
+    f_mps = daemon_file_for("convaiinnovations/laya", None, "mps", False, None)
+    assert f_auto == f_mps, "hash after resolve: auto=None and explicit mps must be same file on MPS host"
+
+    # Start daemon with explicit --device mps, then client without --device should find it
+    model = "convaiinnovations/laya-test-hash-mps"
+    daemon_file_mps = daemon_file_for(model, None, "mps", False, None)
+    daemon_file_auto = daemon_file_for(model, None, None, False, None)
+    # They should be same file
+    assert daemon_file_mps == daemon_file_auto
+
+    # Also subfolder "" vs None and lang case
+    assert daemon_file_for("m", None, None, False, None) == daemon_file_for("m", "", None, False, None)
+    assert daemon_file_for("m", None, None, False, "EN") == daemon_file_for("m", None, None, False, "en")
+
+    # Now test live daemon discovery across device flag mismatch
+    if daemon_file_mps.exists():
+        daemon_file_mps.unlink()
+    t = threading.Thread(
+        target=run_server,
+        kwargs=dict(
+            model=model,
+            subfolder=None,
+            device="mps",
+            router=False,
+            lang=None,
+            port=0,
+            idle_timeout=10,
+            daemon_file=daemon_file_mps,
+        ),
+        daemon=True,
+    )
+    t.start()
+    for _ in range(30):
+        time.sleep(0.2)
+        alive, _, _ = is_daemon_alive(daemon_file_mps)
+        if alive:
+            break
+    assert alive, "daemon with --device mps did not start"
+
+    # Client without --device (auto) should find it via daemon_port_for_args
+    import types
+
+    fake_args = types.SimpleNamespace(
+        model=model, subfolder=None, device=None, router=False, lang=None, no_daemon=False
+    )
+    port = daemon_port_for_args(fake_args)  # type: ignore
+    assert port is not None, (
+        "predict without --device should find daemon started with --device mps (hash after resolve)"
+    )
+
+    # Also via CLI: predict without --device should auto-use daemon
+    q = {"q1": {"type": "noul", "instructions": "Is it good?"}}
+    out, err = _run_cli(
+        ["predict", "--text", "forest good", "--questions-inline", json.dumps(q), "--model", model, "--format", "json"]
+    )
+    assert "using daemon" in err, "CLI predict without --device should have used daemon started with --device mps"
+
+    # Cleanup
+    from laya_cli.daemon import _http_post_shutdown
+
+    alive, status, port = is_daemon_alive(daemon_file_mps)
+    if alive and port:
+        _http_post_shutdown(port)
+        time.sleep(0.3)
+    if daemon_file_mps.exists():
+        try:
+            daemon_file_mps.unlink()
+        except Exception:
+            pass
+    if daemon_file_auto.exists() and daemon_file_auto != daemon_file_mps:
+        try:
+            daemon_file_auto.unlink()
+        except Exception:
+            pass

{laya_cli-0.2.1 → laya_cli-0.2.2}/uv.lock

@@ -499,7 +499,7 @@ wheels = [
 
 [[package]]
 name = "laya-cli"
-version = "0.2.1"
+version = "0.2.2"
 source = { editable = "." }
 dependencies = [
     { name = "huggingface-hub" },