| LOW | setup.py | 42 | def get_nvshmem_host_lib_name(base_dir): |
| LOW | deep_ep/buffers/legacy.py | 176 | def get_low_latency_rdma_size_hint(num_max_dispatch_tokens_per_rank: int, hidden: int, num_ranks: int, num_experts: |
| LOW | deep_ep/buffers/legacy.py | 672 | def low_latency_update_mask_buffer(self, rank_to_mask: int, mask: bool = False): |
| LOW | deep_ep/buffers/legacy.py | 683 | def low_latency_query_mask_buffer(self, mask_status: torch.Tensor): |
| LOW | deep_ep/buffers/legacy.py | 693 | def low_latency_clean_mask_buffer(self): |
| LOW | deep_ep/buffers/legacy.py | 700 | def get_next_low_latency_combine_buffer(self, handle: object): |
| LOW | deep_ep/buffers/elastic.py | 281 | def get_engram_storage_size_hint(num_entries: int, hidden: int, |
| LOW | deep_ep/buffers/elastic.py | 330 | def get_agrs_num_max_session_bytes(group: dist.ProcessGroup, |
| LOW | deep_ep/buffers/elastic.py | 351 | def get_agrs_buffer_size_hint(group: dist.ProcessGroup, |
| LOW | deep_ep/utils/envs.py | 183 | def check_torch_deterministic() -> None: |
| LOW | deep_ep/utils/envs.py | 223 | def check_fast_rdma_atomic_support(nic_name: str = _DEFAULT_NIC_NAME) -> bool: |
| LOW | deep_ep/utils/comm.py | 74 | def destroy_all_managed_nccl_comm() -> None: |
| LOW | deep_ep/utils/gate.py | 116 | def get_precise_unbalanced_scores(num_tokens: int, num_experts: int, num_ranks: int, num_topk: int, ratio: float): |
| LOW | deep_ep/utils/gate.py | 148 | def map_unbalanced_ratio_to_factor(num_tokens: int, num_experts: int, num_ranks: int, num_topk: int, ratio: float) -> fl |
| LOW | deep_ep/utils/gate.py | 167 | def get_random_unbalanced_scores(num_tokens: int, num_experts: int, num_ranks: int, num_topk: int, ratio: float): |
| LOW | deep_ep/utils/refs.py | 126 | def generate_pre_combine_data(src_token_global_idx: torch.Tensor, |
| LOW | tests/elastic/test_ep.py | 279 | def get_unique_and_valid_dst_count(dst_idx: torch.Tensor, |
| LOW | tests/legacy/test_low_latency.py | 14 | def simulate_failure_and_skip(rank: int, api: Literal["dispatch", "combine", "clean"], expected_masked_ranks: Set[int]): |
| LOW | tests/legacy/test_low_latency.py | 33 | def query_mask_buffer_and_check(api: Literal["dispatch", "combine", "clean"], buffer: deep_ep.Buffer, mask_status: torch |