Add Function Vectors reproduction example notebook

[Devansh-Ruia]

Adds examples/function_vectors_reproduction.ipynb, which reproduces the function vector extraction from Todd et al. (ICLR 2024) using nnsight's tracing API.

What this does

• Loads [model name you used], extracts a function vector for the [task name] task using N in-context demonstrations
• Injects the FV into a zero-shot context and measures the task-completion delta
• Documents one place where the nnsight port deviates from the original PyTorch hook implementation

Why

The repo has no published example linking a paper's specific result to nnsight idioms. This notebook closes that gap for Function Vectors and gives new users a reference for the trace-and-intervene pattern on real interpretability work.

How it was tested

• Runs end-to-end on a clean kernel restart
• All cells produce output
• Tested on Collab A100

Happy to revise based on review feedback. First contribution to the repo.