Fix FactoredMatrix indexing returning empty result for -1 index

[Kymi808]

Description

FactoredMatrix.__getitem__ routes an integer index into the matrix (ldim/rdim) dimensions through _convert_to_slice, which turns an int v into slice(v, v + 1). For v == -1 this becomes slice(-1, 0) — an empty slice — so indexing the last row/column with a negative index silently returns a (0, ...) tensor instead of the requested element.

import torch
from transformer_lens import FactoredMatrix

fm = FactoredMatrix(torch.rand(2, 4, 3), torch.rand(2, 3, 5))  # shape (2, 4, 5)

fm[0, -1].AB.squeeze().shape      # got (0, 5), expected (5,)
fm[0, 1, -1].AB.squeeze().shape   # got (0,),  expected ()

Only -1 is affected; other negative indices (-2, -3, …) work because v + 1 stays negative and produces a valid slice.

Fix

In _convert_to_slice, use None as the slice stop when v == -1, so slice(-1, None) keeps the final element. Positive indices and other negatives are unchanged.

Type of change

• Bug fix (non-breaking change which fixes an issue)

Checklist:

• I have commented my code, particularly in hard-to-understand areas
• My changes generate no new warnings
• I have added tests that prove my fix is effective (negative indices on each matrix dimension in tests/unit/factored_matrix/test_get_item.py; they fail on main and pass with this change)
• New and existing unit tests pass locally with my changes (pytest tests/unit/factored_matrix/ → 147 passed, 1 skipped)
• I have not rewritten tests relating to key interfaces which would affect backward compatibility

[jlarson4]

Excellent work on this @Kymi808! Thanks for your contribution, merging now