Even though litellm type definitions and openai API specifies content as list of dictionaries (with type and relevant attributes potentially to allow multimodal inputs/outputs in addition to text), ollama has been demonstrating marshal errors. As a workaround this change simplifies the content as string when there is no more than one content part.
Fixes#642, #928, #376
PiperOrigin-RevId: 766890141
This is in response to the litellm v1.71.2 + ollama v0.9.0 sending function call chunks with 0 indices across multiple calls and lacking call ids.
Solutions introduced:
1. increment fallback index when accumulated arg becomes json parsable.
2. tolerate finish reason == stop when tool calls are present
3. fallback to index when tool call id is None
Fixes https://github.com/google/adk-python/issues/294
PiperOrigin-RevId: 766258344
This is useful for injecting artifacts and session state variable into instruction template typically in instruction providers.
PiperOrigin-RevId: 761595473
When the user provides instruction provider, we assume that they will inject the session state parameters if needed. This assumption allows users to return code snippets in the instruction provider without any template replacement.
PiperOrigin-RevId: 759705471
Also included a token_usage sample that showcases the token usage of subagents with different models under a parent agent.
PiperOrigin-RevId: 759347015
(before/after) tool callbacks are invoked throughout the provided chain until one callback does not return None. Callbacks can be async and sync.
PiperOrigin-RevId: 756526507
(before/after) agent callbacks are invoked throughout the provided chain until one callback does not return None. Callbacks can be async and sync.
PiperOrigin-RevId: 756359693
(before/after) model callbacks are invoked throughout the provided chain until one callback does not return None. Callbacks can be async and sync.
PiperOrigin-RevId: 755565583
