Nc/1oct/py prepare multipart

python/langsmith/schemas.py

@@ -12,6 +12,7 @@
     List,
     Optional,
     Protocol,
+    Tuple,
     Union,
     runtime_checkable,
 )
@@ -43,6 +44,9 @@
 SCORE_TYPE = Union[StrictBool, StrictInt, StrictFloat, None]
 VALUE_TYPE = Union[Dict, str, None]
 
+Attachments = Dict[str, Tuple[str, bytes]]
+"""Attachments associated with the run. Each entry is a tuple of (mime_type, bytes)."""
+
 
 class ExampleBase(BaseModel):
     """Example base model."""
@@ -318,6 +322,9 @@ class Run(RunBase):
     """  # noqa: E501
     in_dataset: Optional[bool] = None
     """Whether this run is in a dataset."""
+    attachments: Attachments = Field(default_factory=dict)
+    """Attachments associated with the run.
+    Each entry is a tuple of (mime_type, bytes)."""
     _host_url: Optional[str] = PrivateAttr(default=None)
 
     def __init__(self, _host_url: Optional[str] = None, **kwargs: Any) -> None:
@@ -376,6 +383,7 @@ class RunLikeDict(TypedDict, total=False):
     output_attachments: Optional[dict]
     trace_id: UUID
     dotted_order: str
+    attachments: Attachments
 
 
 class RunWithAnnotationQueueInfo(RunBase):
@@ -637,6 +645,8 @@ class AnnotationQueue(BaseModel):
 class BatchIngestConfig(TypedDict, total=False):
     """Configuration for batch ingestion."""
 
+    use_multipart_endpoint: bool
+    """Whether to use the multipart endpoint for batch ingestion."""
     scale_up_qsize_trigger: int
     """The queue size threshold that triggers scaling up."""
     scale_up_nthreads_limit: int

python/langsmith/utils.py

@@ -146,9 +146,6 @@ def raise_for_status_with_text(
     except requests.HTTPError as e:
         raise requests.HTTPError(str(e), response.text) from e  # type: ignore[call-arg]
 
-    except httpx.HTTPError as e:
-        raise httpx.HTTPError(str(e), response.text) from e  # type: ignore[call-arg]
-
 
 def get_enum_value(enu: Union[enum.Enum, str]) -> str:
     """Get the value of a string enum."""

python/pyproject.toml

@@ -33,6 +33,7 @@ pydantic = [
 requests = "^2"
 orjson = "^3.9.14"
 httpx = ">=0.23.0,<1"
+requests-toolbelt = "^1.0.0"
 
 [tool.poetry.group.dev.dependencies]
 pytest = "^7.3.1"
@@ -59,6 +60,7 @@ pytest-rerunfailures = "^14.0"
 pytest-socket = "^0.7.0"
 pyperf = "^2.7.0"
 py-spy = "^0.3.14"
+multipart = "^1.0.0"
 
 [tool.poetry.group.lint.dependencies]
 openai = "^1.10"

python/tests/integration_tests/test_client.py

@@ -615,7 +615,10 @@ def test_create_chat_example(
     langchain_client.delete_dataset(dataset_id=dataset.id)
 
 
-def test_batch_ingest_runs(langchain_client: Client) -> None:
+@pytest.mark.parametrize("use_multipart_endpoint", [True, False])
+def test_batch_ingest_runs(
+    langchain_client: Client, use_multipart_endpoint: bool
+) -> None:
     _session = "__test_batch_ingest_runs"
     trace_id = uuid4()
     trace_id_2 = uuid4()
@@ -669,7 +672,12 @@ def test_batch_ingest_runs(langchain_client: Client) -> None:
             "outputs": {"output1": 4, "output2": 5},
         },
     ]
-    langchain_client.batch_ingest_runs(create=runs_to_create, update=runs_to_update)
+    if use_multipart_endpoint:
+        langchain_client.multipart_ingest_runs(
+            create=runs_to_create, update=runs_to_update
+        )
+    else:
+        langchain_client.batch_ingest_runs(create=runs_to_create, update=runs_to_update)
     runs = []
     wait = 4
     for _ in range(15):