Methodforward Forward pass for Sparse Attention. Args: query: Query tensor [sq, b, np, hn]. key: Key tensor [skv, b, np, h
Megatron-LM/megatron/core/transformer/experimental_attention_variant/dsa.py:732
MethodforwardInput: (B, F, 3, 336, 336) if video, else (B, 3, 336, 336) or (num_frames, 3, 336, 336).
Megatron-LM/examples/mimo/model_providers/hf_clip_encoder.py:30
Methodforward_by_lease(
self,
global_lease: str,
path: str,
payload: dict[str, Any],
timeout
terminal-rl/router_server.py:154