Yvonne Yu

This commit introduces a **new feature** to the **Vertex AI Live API** within the `google.genai` library, enabling users to specify `language_codes` for audio transcription configurations. It **enhances the `AudioTranscriptionConfig`** by adding a new `language_codes` field in `google/genai/types.py` and its corresponding dictionary type. To support this, new converter functions, `_AudioTranscriptionConfig_to_mldev`, were implemented and integrated into the `_LiveClientSetup_to_mldev` and `_LiveConnectConfig_to_mldev` conversion flows across `_live_converters.py` and `_tokens_converters.py`. This allows for more precise control over audio transcription, improving accuracy by explicitly defining the expected language(s) for the **Vertex AI Live API**.

This commit **updates the sample code** within the **Google GenAI public samples** to reflect newer model versions. Specifically, the `test_gemini_text_only.py` test now utilizes `gemini-2.5-flash` instead of `gemini-1.5-flash-002` for its `generate_content` calls. This is a **maintenance update** to ensure the provided examples are current and guide users towards the latest available **Gemini models**. The change improves the accuracy and relevance of the **documentation**, helping users avoid outdated model identifiers when interacting with the **Gemini API**.

This commit **updates the chat tests** within the `google.genai` library, specifically in `google/genai/tests/chats/test_send_message.py`, to ensure compatibility with **Gemini 2 or above models**. As a **test improvement**, it modifies model creation to use a new `MODEL_NAME` constant and adjusts assertions to correctly validate **function call parts** returned by these newer models. Additionally, the commit updates expected **chunk counts in streaming tests**, reflecting potential changes in how newer models stream responses. This ensures the library's chat functionality is accurately tested against the latest model capabilities, preventing regressions or false positives.

This commit **updates the test suite** for the **Generative AI library**, specifically targeting the **chat history functionality**. It modifies `google/genai/tests/chats/test_get_history.py` to use the `gemini-2.5-flash` model instead of `gemini-1.5-flash`. This **test update** ensures that the **chat history retrieval** is validated against a more recent and capable model. The change improves the relevance and robustness of the **testing infrastructure**, ensuring continued compatibility and performance validation for the `genai` library's chat features.

This commit performs a **documentation update** by removing outdated references and updating example **GCS URIs for tuning data**. Specifically, it modifies the `README.md`, the reStructuredText source `docs/_sources/index.rst.txt`, and the generated `docs/index.html` to reflect current best practices or available models. This **maintenance** ensures that users consulting the **project documentation** are provided with accurate and up-to-date examples for configuring their tuning data. The change primarily impacts users who rely on the provided examples for setting up their data pipelines.

This commit performs a **test infrastructure update** by upgrading the model used in **byte-related tests** within the `google.genai` library. Specifically, it modifies `generate_content` calls in `test_bytes_internal.py` to utilize `gemini-2.5-flash-001` instead of the older `gemini-1.5-flash-001`. This ensures that the library's **byte handling capabilities** are validated against a more recent and capable Gemini model, thereby improving the relevance and robustness of the test suite.

This commit **improves the logging behavior** within the **`google.genai.types` module** by ensuring that warning messages related to **JSON schema conversions** are logged only once. Specifically, it introduces internal flags to prevent redundant warnings from the `json_schema` property and the `from_json_schema` method. This **maintenance change** reduces log spam, providing a cleaner and more focused output for developers using the library's schema functionalities. New tests were added to confirm this single-logging behavior for both `json_schema` and `from_json_schema`.

This commit **updates the testing suite** for the `google.genai` library. Specifically, it modifies the **`compute_tokens` and `count_tokens` tests** by changing the target model from `gemini-1.5-flash` to `gemini-2.5-flash` in their parameters and calls. This **maintenance update** ensures that the tokenization logic is validated against a more recent and relevant Gemini model version. The change is confined to the **testing infrastructure**, with no direct impact on the core library's runtime behavior or public API.

This commit **updates the test suite** for the `google.genai` library, specifically targeting the **local tokenizer tests**. It modifies the setup in `test_local_tokenizer.py` to initialize the tokenizer with a newer Gemini model, `gemini-2.5-flash`, instead of an older version. This **maintenance** change ensures that the **local tokenizer tests** are run against a more current and relevant model, improving the accuracy and reliability of test results for the `google.genai` library's tokenizer functionality. The update reflects a move to use Gemini 2 or above for these specific tests.

docs: Regenerate docs for 1.49.0

This commit **updates the testing suite** for the `google.genai` library, specifically modernizing tests related to **tool integration** and **model generation**. It modifies test cases in `test_generate_content_tools.py` to utilize the `gemini-2.5-flash` model, ensuring compatibility with newer Gemini versions. Furthermore, this **maintenance update** adapts tests to reflect the updated tool name `google_search` from `google_search_retrieval` and refines exception handling for Vertex AI interactions. This ensures the reliability and accuracy of tool-related tests, validating proper functionality with the latest model and API definitions.

This commit introduces a **new capability** by adding **detailed completion statistics** to the `BatchJob` object within the `google.genai` library. It defines new `CompletionStats` and `CompletionStatsDict` types in `google/genai/types.py` and integrates the `completion_stats` field into the `BatchJob` and `BatchJobDict` structures. The `_BatchJob_from_vertex` conversion logic in `google/genai/batches.py` is updated to **populate this new `completion_stats` field** when converting from Vertex AI batch job objects. This enhancement provides users with more comprehensive information about their batch job executions, improving observability and analysis of **batch processing results**.

This commit performs a **documentation update** within the **`google.genai.types`** module, specifically refining informational messages. It modifies the `Schema.json_schema` and `Schema.from_json_schema` methods by removing a sentence that previously mentioned future field additions. This change clarifies the current behavior and expectations of these **schema handling** methods, ensuring that any logged or displayed information is accurate and up-to-date. The scope is limited to improving the precision of internal or user-facing messages, with no functional impact on the library's behavior.

This commit **fixes** and **enhances** the `t_part` function within the **`google/genai/_transformers.py`** module by implementing **duck typing**. This update allows the function to support a wider range of input types, including duck-typed `File` objects, dictionaries, and PIL `Image` objects. The change significantly improves the flexibility of **content part handling**, enabling more versatile and direct input of various content types. This makes the **`genai` library's content processing** more robust and adaptable to diverse data structures.

This commit introduces **deprecation warnings and updated documentation** for the `json_schema` and `from_json_schema` methods within the **`google.genai.types`** module. The change advises users about the availability of native JSON Schema support and recommends alternative methods for schema handling. This **documentation and maintenance** update aims to guide API consumers towards preferred patterns, ensuring a smoother transition away from the deprecated methods and improving overall API clarity.

This commit **adds new tests** to the `google.genai` library, specifically targeting the **chat functionality**. It introduces `test_thinking_budget` and `test_thinking_budget_stream` within `google/genai/tests/chats/test_send_message.py` to verify the correct behavior of the chat's internal "thinking budget" and "thought generation" mechanisms. This **test addition** enhances the **reliability and robustness** of the `Chat` API by ensuring its internal reasoning processes are functioning as intended. The new tests help prevent regressions in how the model manages resources and generates intermediate thoughts during message processing.

This commit **refactors** the logging mechanism within the **`async_generate_content_stream`** method in the `google/genai/models.py` module. As a **maintenance** task, it adjusts the logging to be conditional on the iteration count, specifically for asynchronous content generation streams. This change optimizes the verbosity of logs, ensuring relevant information is captured without excessive output during frequent iterations. The improvement enhances the internal efficiency and clarity of the **`google-generative-ai` library's** streaming capabilities.

This commit introduces a **bug fix** and **refactoring** within the **`google.genai._transformers` module** to prevent potential dual-import confusion during type assertions. It modifies various transformer functions, including those handling types like `t_function_response`, `t_blob`, and `t_content`, to utilize the `_is_duck_type_of` utility for more robust and reliable type checking. This change addresses subtle issues that could arise from multiple imports of the same type definition, improving the stability of the **`google.genai` library**. New unit tests have also been added for the `_is_duck_type_of` utility to ensure its correctness.

This commit **refactors** the **content type assertion** logic within the `google/genai/_transformers.py` module. It introduces a new helper function, `_is_duck_type_of`, which is then utilized by the `t_part` and `t_contents` functions to perform type checks. This change acts as a **bug fix** by preventing potential **dual import issues** that could arise during content type assertions. Ultimately, it enhances the **robustness** and **maintainability** of the **content transformation** pipeline within the `genai` library.

This commit performs **documentation cleanup** by removing redundant examples from the docstrings of the **`Models`** and **`AsyncModels`** classes within `google/genai/models.py`. Specifically, it targets and eliminates superfluous examples related to 'meta/llama' that were cluttering the API documentation. This improves the **readability and conciseness of the `genai` library's model API documentation**, ensuring users receive clearer and more focused information without altering any functional behavior.