fix(py): fix with gemini

Author: MengqinShen-GL

genkit-tools/common/src/types/action.ts

@@ -53,6 +53,10 @@ export const ActionSchema = z
       .record(z.string(), CustomAnySchema)
       .describe('Metadata about the action (e.g. supported model features).')
       .nullish(),
+    sampleInput: z
+      .unknown()
+      .describe('Sample input for the action to use as default in Dev UI.')
+      .nullish(),
   })
   .openapi('Action');
 

py/packages/genkit/src/genkit/ai/_aio.py

@@ -619,7 +619,7 @@ async def embed(
         {version, ...config, ...options} (options take precedence).
 
         Args:
-            embedder: Embedder name (e.g., 'googleai/gemini-embedding-001') or
+            embedder: Embedder name (e.g., 'googleai/text-embedding-004') or
                 an EmbedderRef with configuration.
             content: A single string, Document, or DocumentData to embed.
             metadata: Optional metadata to apply to the document. Only used
@@ -634,25 +634,25 @@ async def embed(
             ValueError: If content is not specified.
 
         Example - Basic string embedding:
-            >>> embeddings = await ai.embed(embedder='googleai/gemini-embedding-001', content='Hello, world!')
+            >>> embeddings = await ai.embed(embedder='googleai/text-embedding-004', content='Hello, world!')
             >>> print(len(embeddings[0].embedding))  # Vector dimensions
 
         Example - With metadata:
             >>> embeddings = await ai.embed(
-            ...     embedder='googleai/gemini-embedding-001',
+            ...     embedder='googleai/text-embedding-004',
             ...     content='Product description',
             ...     metadata={'category': 'electronics'},
             ... )
 
         Example - With embedder options:
             >>> embeddings = await ai.embed(
-            ...     embedder='googleai/gemini-embedding-001',
+            ...     embedder='googleai/text-embedding-004',
             ...     content='Search query',
             ...     options={'task_type': 'RETRIEVAL_QUERY'},
             ... )
 
         Example - Using EmbedderRef:
-            >>> ref = create_embedder_ref('googleai/gemini-embedding-001', config={'task_type': 'CLUSTERING'})
+            >>> ref = create_embedder_ref('googleai/text-embedding-004', config={'task_type': 'CLUSTERING'})
             >>> embeddings = await ai.embed(embedder=ref, content='Text')
         """
         embedder_name = self._resolve_embedder_name(embedder)
@@ -706,7 +706,7 @@ async def embed_many(
         and passes options directly. This matches the JS canonical behavior.
 
         Args:
-            embedder: Embedder name (e.g., 'googleai/gemini-embedding-001') or
+            embedder: Embedder name (e.g., 'googleai/text-embedding-004') or
                 an EmbedderRef.
             content: List of strings, Documents, or DocumentData to embed.
             metadata: Optional metadata to apply to all items. Only used when
@@ -722,22 +722,22 @@ async def embed_many(
 
         Example - Basic batch embedding:
             >>> embeddings = await ai.embed_many(
-            ...     embedder='googleai/gemini-embedding-001',
+            ...     embedder='googleai/text-embedding-004',
             ...     content=['Doc 1', 'Doc 2', 'Doc 3'],
             ... )
             >>> for i, emb in enumerate(embeddings):
             ...     print(f'Doc {i}: {len(emb.embedding)} dims')
 
         Example - With shared metadata:
             >>> embeddings = await ai.embed_many(
-            ...     embedder='googleai/gemini-embedding-001',
+            ...     embedder='googleai/text-embedding-004',
             ...     content=['text1', 'text2'],
             ...     metadata={'batch_id': 'batch-001'},
             ... )
 
         Example - With options (EmbedderRef config is NOT extracted):
             >>> embeddings = await ai.embed_many(
-            ...     embedder='googleai/gemini-embedding-001',
+            ...     embedder='googleai/text-embedding-004',
             ...     content=documents,
             ...     options={'task_type': 'RETRIEVAL_DOCUMENT'},
             ... )

py/packages/genkit/src/genkit/blocks/embedding.py

@@ -31,7 +31,7 @@
     |                   | vectors. Contains 'embedding' field and metadata.    |
     +-------------------+------------------------------------------------------+
     | Embedder          | A model/service that converts text to embeddings.    |
-    |                   | Examples: 'googleai/gemini-embedding-001'. Registered  |
+    |                   | Examples: 'googleai/text-embedding-004'. Registered  |
     |                   | as actions, invoked via embed() and embed_many().    |
     +-------------------+------------------------------------------------------+
     | EmbedderRef       | Reference bundling embedder name with optional       |
@@ -68,14 +68,14 @@
     - ai.embed_many(): Embed multiple pieces of content in batch
 
 Example - Single embedding:
-    >>> embeddings = await ai.embed(embedder='googleai/gemini-embedding-001', content='Hello, world!')
+    >>> embeddings = await ai.embed(embedder='googleai/text-embedding-004', content='Hello, world!')
     >>> vector = embeddings[0].embedding
 
 Example - Batch embedding:
-    >>> embeddings = await ai.embed_many(embedder='googleai/gemini-embedding-001', content=['Doc 1', 'Doc 2', 'Doc 3'])
+    >>> embeddings = await ai.embed_many(embedder='googleai/text-embedding-004', content=['Doc 1', 'Doc 2', 'Doc 3'])
 
 Example - Using EmbedderRef with configuration:
-    >>> ref = create_embedder_ref('googleai/gemini-embedding-001', config={'task_type': 'CLUSTERING'}, version='v1')
+    >>> ref = create_embedder_ref('googleai/text-embedding-004', config={'task_type': 'CLUSTERING'}, version='v1')
     >>> embeddings = await ai.embed(embedder=ref, content='My text')
 
 Note on embed() vs embed_many():
@@ -195,16 +195,12 @@ def embedder_action_metadata(
 
     embedder_info['customOptions'] = options.config_schema if options.config_schema else None
 
-    # Default sample input for easier testing in Dev UI
-    sample_input: dict[str, object] = {'input': [{'content': [{'text': 'Hello, world!'}]}]}
-
     return ActionMetadata(
         kind=cast(ActionKind, ActionKind.EMBEDDER),
         name=name,
         input_json_schema=to_json_schema(EmbedRequest),
         output_json_schema=to_json_schema(EmbedResponse),
         metadata=embedder_metadata_dict,
-        sample_input=sample_input,
     )
 
 

py/packages/genkit/src/genkit/blocks/prompt.py

@@ -851,11 +851,9 @@ class RecipeInput(BaseModel):
         middleware = effective_opts.get('use') or self._use
         context = effective_opts.get('context')
 
-        rendered_options = await self.render(input=input, opts=effective_opts)
-
         result = await generate_action(
             self._registry,
-            rendered_options,
+            await self.render(input=input, opts=effective_opts),
             on_chunk=on_chunk,
             middleware=middleware,
             context=context if context else ActionRunContext._current_context(),  # pyright: ignore[reportPrivateUsage]

py/packages/genkit/src/genkit/core/action/_action.py

@@ -92,8 +92,7 @@
 from functools import cached_property
 from typing import Any, Generic, Protocol, cast, get_type_hints
 
-from pydantic import BaseModel, ConfigDict, TypeAdapter, ValidationError
-from pydantic.alias_generators import to_camel
+from pydantic import BaseModel, TypeAdapter, ValidationError
 from typing_extensions import Never, TypeVar
 
 from genkit.aio import Channel, ensure_async
@@ -525,8 +524,6 @@ def _initialize_io_schemas(
 class ActionMetadata(BaseModel):
     """Metadata for actions."""
 
-    model_config = ConfigDict(alias_generator=to_camel, populate_by_name=True)
-
     kind: ActionKind
     name: str
     description: str | None = None
@@ -536,7 +533,6 @@ class ActionMetadata(BaseModel):
     output_json_schema: dict[str, object] | None = None
     stream_schema: object | None = None
     metadata: dict[str, object] | None = None
-    sample_input: dict[str, object] | None = None
 
 
 _SyncTracingWrapper = Callable[[object | None, ActionRunContext], ActionResponse[Any]]

py/packages/genkit/src/genkit/core/reflection.py

@@ -122,7 +122,6 @@ def _build_actions_payload(
             'inputSchema': getattr(meta, 'input_json_schema', None),
             'outputSchema': getattr(meta, 'output_json_schema', None),
             'metadata': getattr(meta, 'metadata', None),
-            'sampleInput': getattr(meta, 'sample_input', None),
         }
 
         if key not in actions:

py/samples/framework-evaluator-demo/evaluator_demo/pdf_rag.py

@@ -141,14 +141,16 @@ def chunk_text(text: str, chunk_size: int, overlap: int) -> list[str]:
 # Define a flow to index documents into the "vector store"
 # genkit flow:run indexPdf '"./docs/sfspca-cat-adoption-handbook-2023.pdf"'
 @ai.flow(name='index_pdf')
-async def index_pdf(file_path: str = 'samples/framework-evaluator-demo/docs/cat-wiki.pdf') -> None:
+async def index_pdf(
+    file_path: str = 'samples/framework-evaluator-demo/docs/cat-wiki.pdf',
+) -> None:
     """Index a PDF file.
 
     Args:
         file_path: Path to the PDF.
 
     Example:
-        >>> await index_pdf('doc.pdf')
+        >>> await index_pdf('samples/framework-evaluator-demo/docs/cat-wiki.pdf')
     """
     if not file_path:
         file_path = 'samples/framework-evaluator-demo/docs/cat-wiki.pdf'