Operator Test Vectors
Small, deterministic inputs/outputs to validate implementations and demo runs for Operator Set v1.
See also: Operator Set v1
Conventions
- DType defaults: float32 unless specified
- Tolerances: float32 (atol≈1e-5, rtol≈1e-4); float16 (atol≈1e-3, rtol≈1e-2)
- Shapes are explicit; tensors serialized row-major
matmul
Inputs:
- A (2×3): [[1,2,3],[4,5,6]]
- B (3×2): [[7,8],[9,10],[11,12]]
Expected C (2×2):
- [[58,64],[139,154]]
Freivalds sample: n=2, rounds=2 (example r=[1,0])
ewise_add
Inputs:
- X (3): [1,2,3]
- Y (3): [4,5,6]
Expected Z (3): [5,7,9]
reduce_sum (axis=0)
Input:
- X (2×3): [[1,2,3],[4,5,6]]
Expected (3): [5,7,9]
unary_relu
Input:
- X (5): [-2,-0.5,0,0.5,2]
Expected (5): [0,0,0,0.5,2]
transpose
Input:
- X (2×3): [[1,2,3],[4,5,6]]
Permute axes: (1,0)
Expected (3×2): [[1,4],[2,5],[3,6]]
Notes
- Extend with additional ops as Operator Set evolves
- Include binary fixtures when repositories are wired up