remove fp16 flag (pytorch#581)

Summary:
Pull Request resolved: pytorch#581

fp16 is no longer used in table batched embedding op.

Reviewed By: jspark1105

Differential Revision: D27380547

fbshipit-source-id: 944fa2eb8b492a2e162eb2b19fd6aa5e5fc95e2d

fbgemm_gpu/fbgemm_gpu/split_table_batched_embeddings_ops.py

@@ -165,7 +165,6 @@ def __init__(  # noqa C901
         cache_sets: int = 0,
         cache_reserved_memory: float = 0.0,
         cache_precision: SparseType = SparseType.FP32,
-        fp16: bool = False,
         weights_precision: SparseType = SparseType.FP32,
         enforce_hbm: bool = False,  # place all weights/momentums in HBM when using cache
         optimizer: OptimType = OptimType.EXACT_SGD,