#730 update panel flyte 2 app Pull request opened by cosmicBboy This pull request adds several new example applications to the

examples/apps/panel_app

directory, demonstrating Flyte's capabilities for async workflows, distributed training, caching/retries, agent-based orchestration, and integration with external APIs. Additionally, it fixes a serialization bug in the scaling metric logic and updates related unit tests for correct protobuf mapping. New example applications: • Added

sample_distributed_random_forest.py

demonstrating distributed training of a random forest model using async tasks and Flyte's file/directory abstractions. • Added

sample_mnist_training.py

showing async MNIST data loading, model training, and evaluation with PyTorch, including tensor serialization via Flyte's file API. • Added

sample_langgraph_gemini_agent.py

integrating LangGraph and Gemini for weather forecast agent workflows, including tool invocation and external API calls. • Added

sample_pbj_agent.py

implementing a dependency-respecting agent workflow for a "PBJ sandwich" plan, with concurrent async execution of steps. • Added

sample_async.py

,

sample_caching_and_retries.py

, and

sample_hello_world.py

as simple async, caching/retries, and basic workflow examples. [1] [2] [3] Bug fixes and test improvements: • Fixed serialization bug in

_get_scaling_metric

by mapping

val

to

target_value

for both concurrency and request rate metrics. • Updated unit tests in

test_app_serde.py

to verify correct protobuf field mapping for scaling metrics, replacing previous bug documentation with positive assertions. flyteorg/flyte-sdk