openrunner-sdk 2.37.34__tar.gz → 2.37.35__tar.gz

{openrunner_sdk-2.37.34 → openrunner_sdk-2.37.35}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: openrunner-sdk
-Version: 2.37.34
+Version: 2.37.35
 Summary: OpenRunner SDK - W&B-compatible ML experiment tracking client
 Project-URL: Homepage, https://github.com/jqueguiner/openrunner
 Project-URL: Repository, https://github.com/jqueguiner/openrunner
@@ -32,10 +32,14 @@ Requires-Dist: anthropic<1.0,>=0.20; extra == 'anthropic'
 Provides-Extra: catboost
 Requires-Dist: catboost<2.0,>=1.2; extra == 'catboost'
 Provides-Extra: dev
+Requires-Dist: matplotlib<5.0,>=3.7; extra == 'dev'
+Requires-Dist: num2words2<2.0,>=0.1; extra == 'dev'
 Requires-Dist: numpy<3.0,>=1.24; extra == 'dev'
+Requires-Dist: pillow<13.0,>=10.0; extra == 'dev'
 Requires-Dist: pytest-asyncio<1.0,>=0.23; extra == 'dev'
 Requires-Dist: pytest-cov<6.0,>=5.0; extra == 'dev'
 Requires-Dist: pytest<9.0,>=8.0; extra == 'dev'
+Requires-Dist: scikit-learn<2.0,>=1.3; extra == 'dev'
 Provides-Extra: diffusers
 Requires-Dist: diffusers<1.0,>=0.25; extra == 'diffusers'
 Provides-Extra: fastai

{openrunner_sdk-2.37.34 → openrunner_sdk-2.37.35}/openrunner/cli.py

@@ -4110,10 +4110,37 @@ def gpu_credentials() -> None:
 
 @gpu_credentials.command("add")
 @click.option("--org", "org_id", required=True, help="Organization ID (UUID)")
-@click.option("--provider", required=True, help="Provider, e.g. runpod, vastai, ovhcloud, lambda")
-@click.option("--api-key", "provider_key", required=True, help="Provider API key. For ovhcloud pass a JSON blob: '{\"application_key\":..,\"application_secret\":..,\"consumer_key\":..,\"service_name\":..,\"endpoint\":\"ovh-eu\",\"region\":\"GRA11\"}'")
-def gpu_credentials_add(org_id: str, provider: str, provider_key: str) -> None:
-    """Store a provider API key on the platform (Fernet-encrypted, never returned)."""
+@click.option("--provider", required=True, help="Provider, e.g. runpod, vastai, ovhcloud, lambda, modal")
+@click.option("--api-key", "provider_key", default=None, help="Provider API key. For ovhcloud/lambda pass a JSON blob. For modal use --token-id/--token-secret instead.")
+@click.option("--token-id", default=None, help="Modal token id (from `modal token new`). Use with --token-secret.")
+@click.option("--token-secret", default=None, help="Modal token secret. Use with --token-id.")
+@click.option("--profile", default=None, help="Modal profile / workspace (org), e.g. gladia. Optional.")
+def gpu_credentials_add(org_id: str, provider: str, provider_key: str | None,
+                        token_id: str | None, token_secret: str | None,
+                        profile: str | None) -> None:
+    """Store a provider credential on the platform (Fernet-encrypted, never returned).
+
+    Modal authenticates with a token PAIR, not a single key, plus an optional
+    profile (workspace/org):
+
+        openrunner gpu credentials add --org ORG --provider modal \\
+          --token-id ak-xxxx --token-secret as-xxxx --profile gladia
+    """
+    import json as _json
+
+    # Modal: pack the token pair (+ optional profile) into the JSON blob.
+    if token_id or token_secret:
+        if not (token_id and token_secret):
+            click.echo("Error: modal needs BOTH --token-id and --token-secret.", err=True)
+            raise SystemExit(1)
+        blob = {"token_id": token_id, "token_secret": token_secret}
+        if profile:
+            blob["profile"] = profile
+        provider_key = _json.dumps(blob)
+    if not provider_key:
+        click.echo("Error: pass --api-key (or --token-id/--token-secret for modal).", err=True)
+        raise SystemExit(1)
+
     from openrunner.gpu.api_integration import GpuAPIClient
 
     settings = _load_settings()

{openrunner_sdk-2.37.34 → openrunner_sdk-2.37.35}/openrunner/datasets.py

@@ -330,8 +330,17 @@ class DatasetsClient:
             try:
                 pr = self._http.post(f"/datasets/{ds_id}/files/presign", json=body)
                 _check(pr)
-                url = pr.json()["url"]
-                put = httpx.put(url, content=data, timeout=300.0)
+                pj = pr.json()
+                url = pj["url"]
+                # If the presign signed a storage class (SignedHeaders include
+                # x-amz-storage-class), the PUT MUST echo that exact header value
+                # or S3 returns SignatureDoesNotMatch (403). OVH maps non-json
+                # content types to EXPRESS_ONEZONE, so a plain PUT fails for .jsonl.
+                put_headers = {}
+                sc = pj.get("storage_class")
+                if sc:
+                    put_headers["x-amz-storage-class"] = sc
+                put = httpx.put(url, content=data, headers=put_headers, timeout=300.0)
                 put.raise_for_status()
                 return {"path": rel, "size_bytes": len(data), "sha256": sha, "content_type": ctype}
             except Exception as e:  # noqa: BLE001
@@ -765,9 +774,28 @@ class DatasetsClient:
         return DatasetInfo._from_json(r.json())
 
     def add_derivation(self, repo_id: str, parent_repo_id: str) -> DatasetInfo:
+        """Record that ``repo_id`` is derived from ``parent_repo_id``.
+
+        The parent may be another OpenRunner dataset (``org/name``) OR an external
+        HuggingFace dataset. Prefix the ref with ``hf://`` (or pass a HF repo that
+        isn't an internal dataset) to record external Hub provenance, e.g.::
+
+            add_derivation("gladia/common-voice-16-1", "hf://fsicoli/common_voice_16_1")
+            add_derivation("gladia/common-voice-16-1", "fsicoli/common_voice_16_1")
+        """
         ds = self._resolve(repo_id)
-        parent = self._resolve(parent_repo_id)
-        r = self._http.post(f"/datasets/{ds.id}/derived-from", json={"parent_dataset_id": parent.id})
+        explicit_hf = parent_repo_id.startswith(("hf://", "hf:"))
+        if not explicit_hf:
+            try:
+                parent = self._resolve(parent_repo_id)
+                r = self._http.post(f"/datasets/{ds.id}/derived-from",
+                                    json={"parent_dataset_id": str(parent.id)})
+                _check(r)
+                return DatasetInfo._from_json(r.json())
+            except Exception:
+                pass  # not an internal dataset -> treat as external HF ref
+        r = self._http.post(f"/datasets/{ds.id}/derived-from",
+                            json={"parent_hf_repo": parent_repo_id})
         _check(r)
         return DatasetInfo._from_json(r.json())
 

{openrunner_sdk-2.37.34 → openrunner_sdk-2.37.35}/openrunner/gpu/__init__.py

@@ -142,6 +142,20 @@ def _get_provider_instance(
     Raises:
         ValueError: If provider is unknown or no API key found.
     """
+    # Modal authenticates with a token PAIR (token_id + token_secret), not a
+    # single key. Resolve both and pass them through to ModalProvider.
+    if provider == "modal":
+        token_id, token_secret, profile = _resolve_modal_tokens(api_key)
+        if not (token_id and token_secret):
+            raise ValueError(
+                "No Modal credentials found. Set MODAL_TOKEN_ID + MODAL_TOKEN_SECRET "
+                "(from `modal token new`), pass api_key='{\"token_id\":..,\"token_secret\":..}', "
+                "or store them on the platform: "
+                "`openrunner gpu credentials add --provider modal --token-id .. --token-secret ..`."
+            )
+        cls = get_provider("modal")
+        return cls(api_key=token_id, api_secret=token_secret, profile=profile)
+
     resolved_key = _resolve_api_key(provider, api_key, api_client)
     if not resolved_key:
         raise ValueError(
@@ -154,6 +168,46 @@ def _get_provider_instance(
     return cls(api_key=resolved_key)
 
 
+def _resolve_modal_tokens(
+    api_key: str | None,
+) -> tuple[str | None, str | None, str | None]:
+    """Resolve Modal's (token_id, token_secret, profile) triple.
+
+    ``profile`` is Modal's workspace/organization (e.g. ``gladia``) and is
+    OPTIONAL. Priority: explicit ``api_key`` (a JSON blob
+    ``{"token_id","token_secret","profile"?}``) > MODAL_TOKEN_ID +
+    MODAL_TOKEN_SECRET (+ MODAL_PROFILE) env > a JSON blob in MODAL_API_KEY.
+    Returns (None, None, None) if the tokens aren't resolvable.
+    """
+    import json as _json
+
+    def _from_blob(raw: str) -> tuple[str | None, str | None, str | None]:
+        try:
+            c = _json.loads(raw)
+        except Exception:
+            return None, None, None
+        if not isinstance(c, dict):
+            return None, None, None
+        return (
+            c.get("token_id") or c.get("MODAL_TOKEN_ID"),
+            c.get("token_secret") or c.get("MODAL_TOKEN_SECRET"),
+            c.get("profile") or c.get("MODAL_PROFILE"),
+        )
+
+    if api_key:
+        tid, tsec, prof = _from_blob(api_key)
+        if tid and tsec:
+            return tid, tsec, prof
+    tid = os.environ.get("MODAL_TOKEN_ID")
+    tsec = os.environ.get("MODAL_TOKEN_SECRET")
+    if tid and tsec:
+        return tid, tsec, os.environ.get("MODAL_PROFILE")
+    blob = os.environ.get("MODAL_API_KEY")
+    if blob:
+        return _from_blob(blob)
+    return None, None, None
+
+
 def _cost_guard_loop(
     provider: GpuProvider,
     instance: GpuInstance,

{openrunner_sdk-2.37.34 → openrunner_sdk-2.37.35}/openrunner/gpu/providers/modal_provider.py

@@ -86,7 +86,7 @@ class ModalProvider(GpuProvider):
     it runs Python functions on-demand on GPU containers.
     """
 
-    def __init__(self, api_key: str, api_secret: str = "") -> None:
+    def __init__(self, api_key: str, api_secret: str = "", profile: str | None = None) -> None:
         if not HAS_MODAL:
             raise ProviderError(
                 provider="modal",
@@ -97,6 +97,12 @@ class ModalProvider(GpuProvider):
             )
         self._api_key = api_key
         self._api_secret = api_secret
+        # Modal's workspace/organization (e.g. "gladia"); optional. Surfaced to
+        # the Modal client via the MODAL_PROFILE env var so deploys target it.
+        self._profile = profile or None
+        if self._profile:
+            import os
+            os.environ.setdefault("MODAL_PROFILE", self._profile)
         # Track deployed apps in memory (Modal doesn't have a list-instances API)
         self._apps: dict[str, dict[str, Any]] = {}
 

{openrunner_sdk-2.37.34 → openrunner_sdk-2.37.35}/openrunner/mcp_server.py

@@ -3232,7 +3232,10 @@ async def _tool_upload_artifact(args: dict) -> str:
     # 3. Create a lightweight run for the upload
     ts = datetime.now().strftime("%Y%m%d-%H%M%S")
     run_data = await _api_post("/runs", data={
-        "project": project,
+        # project_id already resolved above — pass it directly so run creation
+        # never re-resolves the "org/project" string (which needs a separate org
+        # field server-side and 400s without it).
+        "project_id": project_id,
         "display_name": f"artifact-upload-{art_name}-{ts}",
         "tags": ["artifact-upload"] + ([art_type] if art_type else []),
         "config": {"artifact_name": art_name, "artifact_type": art_type, "file_count": len(file_entries)},
@@ -3276,7 +3279,7 @@ async def _tool_upload_artifact(args: dict) -> str:
     # 5. Upload files to presigned URLs
     uploaded = 0
     skipped = 0
-    url_map = {u["name"]: u["url"] for u in upload_urls}
+    url_map = {u["name"]: (u.get("url") or u.get("presigned_url")) for u in upload_urls}
 
     async with httpx.AsyncClient(timeout=300.0) as client:
         for entry in file_entries:

{openrunner_sdk-2.37.34 → openrunner_sdk-2.37.35}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "openrunner-sdk"
-version = "2.37.34"
+version = "2.37.35"
 description = "OpenRunner SDK - W&B-compatible ML experiment tracking client"
 readme = "README.md"
 license = {text = "MIT"}
@@ -65,7 +65,7 @@ voice-agent = []
 forced-alignment = ["matplotlib>=3.7,<4.0", "numpy>=1.24,<3.0"]
 mcp = ["mcp>=1.0,<2.0"]
 patent = ["httpx>=0.27,<1.0"]
-dev = ["pytest>=8.0,<9.0", "pytest-asyncio>=0.23,<1.0", "pytest-cov>=5.0,<6.0", "numpy>=1.24,<3.0"]
+dev = ["pytest>=8.0,<9.0", "pytest-asyncio>=0.23,<1.0", "pytest-cov>=5.0,<6.0", "numpy>=1.24,<3.0", "Pillow>=10.0,<13.0", "scikit-learn>=1.3,<2.0", "matplotlib>=3.7,<5.0", "num2words2>=0.1,<2.0"]
 
 [tool.hatch.build.targets.wheel]
 packages = ["openrunner"]

{openrunner_sdk-2.37.34 → openrunner_sdk-2.37.35}/tests/conftest.py

@@ -13,11 +13,21 @@ import pytest
 
 
 @pytest.fixture
-def clean_env(monkeypatch: pytest.MonkeyPatch) -> pytest.MonkeyPatch:
-    """Clear all OPENRUNNER_*, OPENRUN_*, and WANDB_* env vars for a clean test environment."""
+def clean_env(monkeypatch: pytest.MonkeyPatch, tmp_path: Path) -> pytest.MonkeyPatch:
+    """Clear all OPENRUNNER_*, OPENRUN_*, and WANDB_* env vars for a clean test environment.
+
+    Also redirects HOME (and USERPROFILE) to a throwaway dir so tests never pick up
+    the developer's real ``~/.openrunner/session_config.json`` creds -- otherwise a
+    machine that has run ``openrunner login`` fails "no api key" / default-URL tests
+    that pass on a clean CI box.
+    """
     for var in list(os.environ):
         if var.startswith(("OPENRUNNER_", "OPENRUN_", "WANDB_")):
             monkeypatch.delenv(var, raising=False)
+    fake_home = tmp_path / "home"
+    fake_home.mkdir(exist_ok=True)
+    monkeypatch.setenv("HOME", str(fake_home))
+    monkeypatch.setenv("USERPROFILE", str(fake_home))
     return monkeypatch
 
 

{openrunner_sdk-2.37.34 → openrunner_sdk-2.37.35}/tests/integration/conftest.py

@@ -23,5 +23,11 @@ def pytest_collection_modifyitems(config: pytest.Config, items: list[pytest.Item
     skip = pytest.mark.skip(
         reason=f"live integration: set {', '.join(missing)} to enable"
     )
+    # NOTE: a subdirectory conftest's modifyitems hook receives the WHOLE session's
+    # items, not just this directory's. Only skip the live integration tests that
+    # actually live under tests/integration/ -- otherwise the entire unit suite is
+    # skipped whenever these env vars are unset.
+    integration_dir = os.path.dirname(__file__)
     for item in items:
-        item.add_marker(skip)
+        if str(item.fspath).startswith(integration_dir):
+            item.add_marker(skip)

{openrunner_sdk-2.37.34 → openrunner_sdk-2.37.35}/tests/test_aliases.py

@@ -187,6 +187,8 @@ def test_run_link_artifact_sets_aliases():
     run._job_type = None
     run._step = 0
     run._state = "running"
+    run._artifacts_logged = 0
+    run._artifacts_used = 0
 
     from openrunner.api_client import APIClient
     from openrunner.buffer import WriteBuffer
@@ -236,6 +238,8 @@ def test_run_use_artifact_colon_syntax():
     run._name = "test-run"
     run._step = 0
     run._state = "running"
+    run._artifacts_logged = 0
+    run._artifacts_used = 0
     run._client = APIClient("http://test", "test-key", transport=transport)
     run._buffer = WriteBuffer()
     run._summary = Summary()