#343: Support collections of tensors in args/kwargs for compile

[akhilg-nv]

No description provided.

[github-actions]

⚠️ Performance Alert ⚠️

Possible performance regression was detected for benchmark.
Benchmark result of this commit is worse than the previous benchmark result exceeding threshold 1.10.

Benchmark suite	Current: 87ecff6	Previous: 48b5d7d	Ratio
tests/performance/test_perf.py::test_perf_regression[sdpa-float16]	2983.0722283127584 iter/sec (stddev: 0.0002213579824486067)	3302.897668253486 iter/sec (stddev: 0.00007090329654256612)	1.11

This comment was automatically generated by workflow using github-action-benchmark.

[Copilot]

Pull Request Overview

This PR adds support for collections of tensors (nested dictionaries and lists) in the args and kwargs parameters of the compile function. Previously, only individual InputInfo or DimensionInputInfo objects were supported as runtime inputs, but now these can be organized within container structures.

Key Changes:

• Recursive processing of nested data structures (dicts, lists, tuples) containing InputInfo objects during compilation
• Flattening of nested input structures at runtime to match the compilation structure
• New validation for input extraction ensuring all compiled inputs are provided at runtime

Reviewed Changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.

File	Description
tripy/nvtripy/backend/api/compile.py	Added recursive process_arg_and_flag function to handle nested containers during compilation and maintain flattened trace inputs
tripy/nvtripy/backend/api/executable.py	Added extract_inputs function to recursively flatten runtime tensor containers and validate against expected compiled inputs
tripy/tests/backend/api/test_compile.py	Added three comprehensive test cases covering nested dicts, nested sequences, and mixed container scenarios

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[akhilg-nv]

I see that some additional docs tests are failing due to the new required argument (which caches the input structure) that I added for creating an executable. Initially I implemented it as an optional argument and then re-created it in the executable if it wasn't provided, but this didn't seem to make much sense to me and I was hoping to have the relevant logic only in one place. I could do this entirely in the __init__ of executable if preferred.