MCPcopy Create free account

hub / github.com/LMCache/LMCache / functions

Functions13,371 in github.com/LMCache/LMCache

↓ 2 callersFunction_chunk_slice
(t: torch.Tensor, start: int, end: int)
lmcache/v1/multiprocess/http_apis/cache_api.py:227
↓ 2 callersMethod_cleanup_memory_objects
Safely cleanup memory objects by decrementing reference counts
lmcache/v1/storage_backend/p2p_backend.py:703
↓ 2 callersMethod_clear
( self, tokens: Optional[Union[torch.Tensor, List[int]]] = None, locations: Optional[L
lmcache/v1/cache_engine.py:1575
↓ 2 callersMethod_close_request_span
End the root span and clean up all per-session state. Args: session_id: The request session identifier. end_ts: Wall-
lmcache/v1/mp_observability/subscribers/tracing/mp_server.py:330
↓ 2 callersMethod_close_request_span
End the ``"cb.request"`` root span and clean up per-session state. Args: session_id: The request session identifier.
lmcache/v1/mp_observability/subscribers/tracing/cb_server.py:413
↓ 2 callersMethod_coalesce
Coalesces the current block with the previous and/or successor block. This assumes the curr_block is NOT in self._explicit_list
lmcache/v1/memory_management.py:1321
↓ 2 callersMethod_compute_chunk_hashes
Compute prefix hashes for all chunks using ChunkedTokenDatabase. Args: token_ids: List of token IDs to process Returns:
lmcache/v1/lookup_client/record_strategies/base.py:37
↓ 2 callersFunction_compute_client_checksums
Hash a paged block range from client-side KV tensors. For each chunk (``chunk_size // block_size`` consecutive blocks), feed every layer's by
lmcache/cli/commands/bench/server_bench/helpers.py:617
↓ 2 callersFunction_compute_kv_shape
Compute a kv_shape that yields the given obj_size. The returned shape is in vllm format: ``(num_layers, 2, num_tokens, num_heads, head_size)`
lmcache/v1/check/utils.py:28
↓ 2 callersFunction_compute_tensor_checksum
Compute MD5 checksum of a tensor.
lmcache/v1/internal_api_server/vllm/cache_api.py:133
↓ 2 callersFunction_config
Minimal CPU-only config for HS-store tests.
tests/v1/test_hidden_state_store.py:40
↓ 2 callersMethod_copy_into_buf
Copy bytes into buffer; buf may be non-byte or multi-dimensional (e.g. from bfloat16 tensor).
tests/conftest.py:291
↓ 2 callersMethod_copy_reservation_to_target_locked
( self, source: DaxDeviceEntry, reservation: DaxReadReservation[ObjectKey], )
lmcache/v1/distributed/l2_adapters/dax_l2_adapter.py:1383
↓ 2 callersFunction_create_complex_memory_obj
( size: int = 1024, fill_value: complex = 0j, )
tests/v1/distributed/test_raw_block_l2_adapter.py:104
↓ 2 callersMethod_create_health_monitor
Create, configure, and start the health monitor. Shared implementation used by subclass maybe_create_health_monitor.
lmcache/integration/base_service_factory.py:97
↓ 2 callersMethod_create_heartbeat_socket
(self)
lmcache/v1/cache_controller/worker.py:308
↓ 2 callersMethod_create_key
Convert token IDs to an IPC cache engine key. Args: token_ids: The token IDs. start: Start token index. e
lmcache/integration/vllm/vllm_multi_process_adapter.py:1749
↓ 2 callersMethod_create_key
( self, token_ids: List[int], start: int, end: int, request_id: int,
lmcache/integration/tensorrt_llm/tensorrt_mp_adapter.py:114
↓ 2 callersMethod_create_key
( self, token_ids: List[int], request_id: int, )
lmcache/integration/tensorrt_llm/tensorrt_mp_adapter.py:306
↓ 2 callersFunction_create_l1_manager
Create an L1 memory manager sized to ``size_bytes`` for this benchmark.
lmcache/tools/transfer_channel_benchmark/benchmark.py:73
↓ 2 callersFunction_create_local_cpu_backend
Helper to create a local CPU backend for testing.
tests/v1/storage_backend/test_valkey_connector.py:113
↓ 2 callersFunction_create_local_cpu_backend
Helper to create a local CPU backend for testing.
tests/v1/storage_backend/test_resp_connector.py:64
↓ 2 callersFunction_create_memory_obj
Create a test TensorMemoryObj.
lmcache/v1/check/check_mode_test_l2_adapter.py:45
↓ 2 callersFunction_create_metadata
(use_mla, kv_caches, engine_kv_format)
tests/v1/test_xpu_connector.py:814
↓ 2 callersFunction_create_metadata
(use_mla, kv_caches, engine_kv_format)
tests/v1/test_gpu_connector.py:939
↓ 2 callersFunction_create_object_key
Create a test ObjectKey.
lmcache/v1/check/check_mode_test_l2_adapter.py:36
↓ 2 callersMethod_create_req_socket
(self)
lmcache/v1/cache_controller/worker.py:297
↓ 2 callersMethod_create_socket
Create and configure a REQ socket.
lmcache/v1/rpc/zmq_transport.py:61
↓ 2 callersMethod_create_test_key
Create a test CacheEngineKey.
tests/v1/internal_api_server/test_load_fs_chunks.py:123
↓ 2 callersMethod_create_user_session
(self)
benchmarks/multi_round_qa/multi-round-qa.py:412
↓ 2 callersMethod_create_zmq_client_transport
Create a ZMQ REQ-REP client transport.
lmcache/v1/lookup_client/factory.py:201
↓ 2 callersMethod_create_zmq_server_transport
Create a ZMQ ROUTER server transport.
lmcache/v1/lookup_client/factory.py:235
↓ 2 callersFunction_decode_single_channel
( cdf_layer_c, bs_np, start_off, end_off, n_tokens, out_layer_c, )
lmcache/python_ops_fallback.py:2254
↓ 2 callersFunction_delete
(salt: str, **kw)
tests/v1/mp_coordinator/test_l2_api.py:34
↓ 2 callersMethod_delete_source_keys_after_migration_locked
( self, source: DaxDeviceEntry, keys: list[ObjectKey], force: bool, )
lmcache/v1/distributed/l2_adapters/dax_l2_adapter.py:1430
↓ 2 callersMethod_deregister_file
Deregister a file from nixl.
lmcache/v1/distributed/l2_adapters/nixl_store_dynamic_l2_adapter.py:175
↓ 2 callersFunction_discover_backends_once
Populate :data:`_BACKENDS` on first use. Walks ``lmcache.v1.platform`` two levels deep (``platform`` -> ``<backend>`` -> ``cache_context``) a
lmcache/v1/platform/cache_context.py:61
↓ 2 callersMethod_do_write
(self, offset: int, memory_obj: MemoryObj, size: int)
lmcache/v1/storage_backend/plugins/dax_backend.py:666
↓ 2 callersMethod_drain_once
(self)
lmcache/v1/multiprocess/native_completion.py:111
↓ 2 callersMethod_draw
Render the real-time stats display.
lmcache/cli/commands/bench/engine_bench/progress.py:100
↓ 2 callersFunction_dtype_to_name
Return a stable torch dtype name without module prefix.
lmcache/v1/multiprocess/modules/server_transfer.py:32
↓ 2 callersFunction_engine_kv_format_name
Return a stable enum member name for Python and pybind enum values.
lmcache/v1/platform/musa/native_kv_transfer.py:318
↓ 2 callersFunction_engine_metric_map
Shape returned by :meth:`lmcache.cli.request.Request.send_request`.
tests/cli/commands/test_query.py:15
↓ 2 callersMethod_ensure_cache_query_connection
Set up ZMQ socket for querying decoder's cache.
lmcache/v1/storage_backend/pd_backend.py:708
↓ 2 callersMethod_ensure_capacity_and_layout
Open the device if needed and compute metadata/data layout.
lmcache/v1/storage_backend/raw_block/core.py:1438
↓ 2 callersFunction_ensure_config_loaded
Trigger lazy import for *name* if it is not yet in the config registry. Raises: ImportError: If the adapter module cannot be
lmcache/v1/distributed/l2_adapters/config.py:75
↓ 2 callersFunction_ensure_driver_open
Idempotently open the cuFile driver and declare async signatures.
lmcache/v1/gpu_connector/_cufile_async.py:92
↓ 2 callersMethod_ensure_engine
(self)
lmcache/integration/vllm/vllm_service_factory.py:239
↓ 2 callersMethod_ensure_heartbeat_started
Lazily start the heartbeat thread on first store/retrieve. The heartbeat starts healthy (the event was set at construction). A live w
lmcache/integration/vllm/vllm_multi_process_adapter.py:1308
↓ 2 callersMethod_ensure_multiprocess_dir
()
lmcache/observability.py:910
↓ 2 callersMethod_ensure_streams
Lazily create MUSA streams on first transfer.
lmcache/v1/gpu_connector/musa_connectors.py:817
↓ 2 callersMethod_evict_keys_locked
( self, source: DaxDeviceEntry, keys: list[ObjectKey], *, force: bool,
lmcache/v1/distributed/l2_adapters/dax_l2_adapter.py:1450
↓ 2 callersMethod_evict_one_lru
(self)
lmcache/v1/hidden_state_store.py:347
↓ 2 callersFunction_execute_cache_operation
Execute a cache operation and return standardized response. Args: operation_name: Name of the operation for error messages operat
lmcache/v1/internal_api_server/vllm/cache_api.py:405
↓ 2 callersMethod_execute_eviction_action
( self, adapter: L2AdapterInterface, action: EvictionAction )
lmcache/v1/distributed/storage_controllers/eviction_controller.py:389
↓ 2 callersMethod_execute_write_phase
Execute the write phase and return elapsed time in seconds. This is shared between single-phase (write-only) and two-phase (write+rea
benchmarks/storage_backend_io/storage_backend_io_benchmark.py:451
↓ 2 callersMethod_exists_in_put_tasks_or_cache
Check whether key exists in put tasks or presence cache. This method only checks the local data structures and does not call the expe
lmcache/v1/storage_backend/nixl_storage_backend.py:1871
↓ 2 callersFunction_extra_cfg
Return ``plugin.frontend.*`` extra config dict. Reads ``runtime_plugin_extra_config`` from the top-level of the aggregated ``LMCACHE_RUNTIME_
lmcache/lmcache_frontend/lmcache_mp_plugin/lmcache_mp_frontend_plugin.py:87
↓ 2 callersMethod_fail_all
Fail all pending futures with the given exception.
lmcache/v1/storage_backend/resp_client.py:115
↓ 2 callersMethod_fail_all
(self, exc: Exception)
lmcache/v1/storage_backend/native_clients/connector_client_base.py:69
↓ 2 callersFunction_fake_obj_key
(chunk_hash: bytes, worker_id: int)
tests/v1/multiprocess/test_blend_v3_load_store_opts.py:134
↓ 2 callersMethod_fetch_object_sizes_sync
(self, object_paths: list[str])
lmcache/v1/distributed/l2_adapters/hfbucket_l2_adapter.py:690
↓ 2 callersMethod_finalize_put
(self, key: KeyT, *, write_failed: bool)
lmcache/v1/storage_backend/dax/core.py:940
↓ 2 callersFunction_flashinfer_available
Check if flashinfer is importable.
lmcache/v1/compute/attention/utils.py:26
↓ 2 callersFunction_flat_to_nested_sglang_mha
(kvcaches_flat, num_layers: int)
tests/v1/test_xpu_sglang_connector.py:80
↓ 2 callersMethod_fmt
(key: CacheEngineKey)
tests/v1/test_nixl_doca_memos.py:83
↓ 2 callersFunction_fold_python
Reference (pure-Python) implementation of :func:`fold`, for testing only. Kept as the oracle the native :func:`fold` is validated against in the
lmcache/v1/distributed/bitmap_ops/fold.py:214
↓ 2 callersFunction_fp8_max
FP8 representable absolute max via torch.finfo.
lmcache/v1/kv_codec/asym_k16_v8.py:60
↓ 2 callersFunction_free_port
()
tests/v1/mp_coordinator/test_integration.py:22
↓ 2 callersMethod_free_slot_locked
(self, slot_id: int)
lmcache/v1/storage_backend/dax/core.py:1007
↓ 2 callersFunction_generate_kv
(num_tokens, device)
tests/benchmarks/test_cachegen.py:32
↓ 2 callersMethod_generate_random_endpoint
Generate random IP and port for test messages
lmcache/tools/controller_benchmark/handlers/register.py:52
↓ 2 callersMethod_generate_random_endpoint
Generate random IP and port for test messages
lmcache/tools/controller_benchmark/handlers/heartbeat.py:52
↓ 2 callersMethod_generate_sync_id
Generate a unique sync session ID
lmcache/v1/cache_controller/full_sync_sender.py:116
↓ 2 callersMethod_get
(self, key: CacheEngineKey)
lmcache/v1/storage_backend/connector/eic_connector.py:416
↓ 2 callersMethod_get_all_hot_cache_keys
Get all chunk hashes from the hot cache
lmcache/v1/cache_controller/full_sync_sender.py:120
↓ 2 callersMethod_get_base_logger
Return an existing logger to reuse registered Prometheus collectors.
lmcache/observability.py:1872
↓ 2 callersMethod_get_base_path
Get file base path for the given key
lmcache/v1/storage_backend/connector/fs_connector.py:133
↓ 2 callersMethod_get_buffer_slice
Hook: Get buffer slice. Override for custom buffer access.
lmcache/v1/memory_management.py:1668
↓ 2 callersMethod_get_cached_object_size
Return a live cache entry, pruning expired metadata opportunistically.
lmcache/v1/storage_backend/connector/hfbucket_connector.py:676
↓ 2 callersMethod_get_connector_metadata
Get the connector metadata. This function should only be called inside the connector. Returns: ConnectorMetadata: the co
lmcache/integration/vllm/lmcache_mp_connector_0201.py:544
↓ 2 callersMethod_get_connector_metadata
Get the connector metadata. This function should only be called inside the connector. Returns: ConnectorMetadata: the co
lmcache/integration/vllm/lmcache_mp_connector_0180.py:523
↓ 2 callersFunction_get_copy_lib
Lazily load and cache the CUDA/ROCm runtime library, or None for CPU fallback.
lmcache/python_ops_fallback.py:36
↓ 2 callersMethod_get_device_for_path_locked
( self, device_path: str, )
lmcache/v1/distributed/l2_adapters/dax_l2_adapter.py:1268
↓ 2 callersMethod_get_device_index_from_uuid
Get the physical device ordinal from its UUID.
lmcache/v1/multiprocess/custom_types.py:69
↓ 2 callersFunction_get_enum_members
Extract {name: value} from a Python enum or pybind11 enum.
tests/v1/test_c_ops_fallback_parity.py:69
↓ 2 callersMethod_get_file_and_tmp_path
Get file and tmp path for the given key
lmcache/v1/storage_backend/connector/fs_connector.py:154
↓ 2 callersMethod_get_file_name
(self, key: CacheEngineKey)
lmcache/v1/storage_backend/connector/fs_connector.py:145
↓ 2 callersMethod_get_int
Extract a positive integer from config Args: key: Configuration key default: Default value if key not found
lmcache/v1/storage_backend/connector/hf3fs_adapter.py:132
↓ 2 callersFunction_get_keyspace_hits
Query keyspace_hits from all Valkey/Redis cluster nodes.
examples/kv_cache_reuse/remote_backends/valkey/benchmark_l2.py:155
↓ 2 callersFunction_get_kvcaches_and_device
Get kvcaches and device from engine's gpu_connector. Args: engine: LMCache engine instance Returns: Tuple of (kvcaches, devi
lmcache/v1/internal_api_server/vllm/cache_api.py:105
↓ 2 callersMethod_get_local_cpu_backend
Get the LocalCPUBackend from the storage manager with proper type. Args: storage_manager: The storage manager instance
lmcache/v1/health_monitor/base.py:292
↓ 2 callersFunction_get_lookup_client
( request: Request, )
lmcache/v1/internal_api_server/vllm/chunk_statistics_api.py:29
↓ 2 callersFunction_get_metadata
Helper to create test metadata.
tests/v1/storage_backend/test_valkey_connector.py:98
↓ 2 callersFunction_get_missing_args
Return list of missing required CLI flags.
lmcache/cli/commands/bench/engine_bench/command.py:331
↓ 2 callersFunction_get_object_size
Safely read the ``size`` attribute from a Hugging Face path object.
lmcache/v1/storage_backend/connector/hfbucket_connector.py:930
↓ 2 callersFunction_get_object_type
Safely read the ``type`` attribute from a Hugging Face path object.
lmcache/v1/storage_backend/connector/hfbucket_connector.py:924
↓ 2 callersMethod_get_or_create_request_span
Return the root span and its OTel context, creating them if absent. The span is stored in the registry under ``span_name="request"`` so
lmcache/v1/mp_observability/subscribers/tracing/mp_server.py:300
↓ 2 callersMethod_get_ping_timeout
Get the ping timeout from the backend config.
lmcache/v1/health_monitor/checks/remote_backend_check.py:135
↓ 2 callersFunction_get_python_params
Extract [(name, has_default, default_value)] via inspect.signature.
tests/v1/test_c_ops_fallback_parity.py:191
← previousnext →2,201–2,300 of 13,371, ranked by callers