alibaba / alibaba/TinyNeuralNetwork
[converter] map gather ops with contiguous indices to slice ops
Open
enhancement
- Dominant language
- Python
- Stars
- 879
- Forks
- 134
- PR merge metrics
- No merged PRs in 30d
Description
Benefits:
1. Slices might be slightly faster than indexing operations.
2. `STRIDED_SLICE`s have wider support among various chips
Contributor guide
Research direction
Start by locating the converter's handling of gather operations and the existing STRIDED_SLICE support. Check how contiguous indices are represented, then validate that the conversion preserves behavior and improves support for the referenced chip targets.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python, pytorch
- Domain
- machine-learning, tooling
- Issue type
- Feature
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 35/100