MCPcopy Create free account

hub / github.com/LMCache/LMCache / functions

Functions13,371 in github.com/LMCache/LMCache

↓ 6 callersFunctionhighest_set_bit
Model-wide prefix hit length: the highest set bit of a fold output. Args: servable: A servable-lengths bitmap from :func:`fold`. Ret
lmcache/v1/distributed/bitmap_ops/fold.py:109
↓ 6 callersMethodlaunch_host_func
Run ``callback(arg)`` synchronously. ``cupy.cuda.Stream.launch_host_func`` schedules the callback on the GPU stream's host-side compl
lmcache/v1/platform/cpu/stub_cpu_device.py:130
↓ 6 callersMethodlaunch_plugins
Launch all configured plugins
lmcache/v1/plugin/runtime_plugin_launcher.py:32
↓ 6 callersMethodlist
(self)
tests/v1/test_remote_mla_worker_id_as0.py:41
↓ 6 callersMethodlist_backends
List all active storage backends. Returns: Dict mapping backend name to its class name.
lmcache/v1/storage_backend/storage_manager.py:1208
↓ 6 callersMethodload_modes
Dynamically load all check mode modules
lmcache/v1/check/__init__.py:29
↓ 6 callersFunctionmake_empty_memory_obj
Create a zero-filled binary tensor memory object. Args: size_bytes: Number of bytes to allocate. Returns: Tensor memory obje
tests/v1/storage_backend/raw_block_test_utils.py:127
↓ 6 callersFunctionmake_object_key
Create a test ObjectKey with the given chunk ID.
tests/v1/distributed/test_store_policy.py:23
↓ 6 callersFunctionmake_temp_key
Create a unique temporary key derived from the original. Appends 16 random bytes (128 bits of entropy) to the original chunk hash. Birthday-c
lmcache/v1/distributed/serde/utils.py:34
↓ 6 callersMethodmark_init_failed
Mark the engine as having failed initialization. This is called by LMCacheManager when an irrecoverable error occurs during
lmcache/v1/cache_engine.py:281
↓ 6 callersFunctionmatch_strided
Aligned-only probe — same hash function as V3 but only probes positions ``{start_pos, start_pos + stride, start_pos + 2*stride, ...}``. Equiv
tests/v1/multiprocess/test_optimized_lookup_v3.py:190
↓ 6 callersMethodmemcheck
(self)
lmcache/v1/memory_management.py:2203
↓ 6 callersFunctionnormalize_url
Ensure *url* has an ``http://`` or ``https://`` scheme.
lmcache/cli/commands/quota/_helpers.py:21
↓ 6 callersFunctionobject_key_to_string
Serialize an ObjectKey using raw-block's reversible key shape. Args: key: Object key supplied by the MP storage layer. Returns:
lmcache/v1/storage_backend/raw_block/key_codec.py:30
↓ 6 callersFunctionon_done
(fut)
tests/v1/multiprocess/test_affinity_pool.py:107
↓ 6 callersMethodon_retrieve_request
Returns the internal "request id" that will be used in on_retrieve_finished
lmcache/observability.py:377
↓ 6 callersMethodopen
(self)
lmcache/v1/storage_backend/connector/hf3fs_connector.py:557
↓ 6 callersMethodp2p_lookup_and_lock
Submit a lookup and lock. After L2 prefetch is enabled, the found chunks will be feteched from L2 to L1. Args: k
lmcache/v1/multiprocess/modules/p2p_controller.py:222
↓ 6 callersMethodp2p_query_lookup_results
Query the results of the lookup request specified by the task ID. Returning a list of TransferChannelAddress objects indicates when the
lmcache/v1/multiprocess/modules/p2p_controller.py:262
↓ 6 callersFunctionpack_metadata
(tensor, fmt: MemoryFormat, **extra_metadata)
lmcache/v1/storage_backend/gds_backend.py:98
↓ 6 callersMethodpop_completed_store_tasks
( self, )
examples/lmc_external_l2_adapter/src/lmc_external_l2_adapter/adapter.py:173
↓ 6 callersMethodposition
LRU rank of *key*: 0 = most-recently used, ``len-1`` = least-recently used. O(log n).
lmcache/tools/cache_simulator/lru_cache.py:92
↓ 6 callersMethodprepare_store
(self, *_args: Any, **_kwargs: Any)
tests/v1/multiprocess/test_engine_driven_transfer.py:450
↓ 6 callersMethodput_many
Store multiple memory objects in DAX slots. Existing committed or in-flight keys are treated as no-op successes. Unsupported multi-te
lmcache/v1/storage_backend/dax/core.py:208
↓ 6 callersMethodquery_prefetch_lookup_hits
Query the number of hits for a prefetch request before it's finished. Args: request_id: The external request ID passed in the loo
lmcache/v1/multiprocess/modules/lookup.py:293
↓ 6 callersMethodquery_read_status
Query the status of a previously submitted read. Args: task_id: The task id returned by ``submit_read``.
lmcache/v1/distributed/transfer_channel/abstract.py:79
↓ 6 callersMethodread_lock
Context manager for read lock with timeout. Args: timeout: Timeout in seconds. None means wait forever.
lmcache/v1/cache_controller/locks.py:85
↓ 6 callersMethodrecord_evicted
Drop ``key`` from the ledger and return the bytes freed (``0`` if ``key`` was unknown).
lmcache/v1/mp_coordinator/l2/usage_manager.py:69
↓ 6 callersMethodregister
Register a client's DEALER socket with the shared poller. Blocks until the loop thread has completed the registration. Args:
lmcache/v1/multiprocess/mq.py:182
↓ 6 callersMethodrelease_handle
(self, handle)
lmcache/v1/storage_backend/nixl_storage_backend.py:544
↓ 6 callersMethodremove
Remove ``key`` from the backend if present. If the key is in-flight, it is marked canceled. If it is committed, its slot is scheduled
lmcache/v1/storage_backend/plugins/dax_backend.py:302
↓ 6 callersMethodrequest_finished
( self, request: "Request", block_ids: list[int], )
lmcache/integration/vllm/vllm_v1_adapter.py:1833
↓ 6 callersMethodrequest_remove_adapter
Non-blocking function to request the removal of a L2 adapter specified by the adapter ID. New store tasks stop routing to the adapter
lmcache/v1/distributed/storage_controllers/store_controller.py:370
↓ 6 callersMethodresolve
Create a config with ``num_documents`` computed from KV cache budget. Args: kv_cache_volume_gb: Target active KV cache volume in
lmcache/cli/commands/bench/engine_bench/workloads/long_doc_qa.py:56
↓ 6 callersFunctionround_up
Round a value up to the nearest alignment boundary. Args: x: Value to align. align: Positive alignment in bytes. Returns:
lmcache/v1/storage_backend/raw_block/core.py:48
↓ 6 callersFunctionrun
(config: LMCacheEngineConfig, shape, dtype)
tests/v1/test_nixl_storage.py:91
↓ 6 callersFunctionscatter_cpu_to_paged_kv
Scatter CPU chunk tensors back into paged KV tensors. Args: kv_caches: Per-layer KV tensor mapping to write into. block_ids: Flat
lmcache/v1/multiprocess/transfer_context/base.py:541
↓ 6 callersMethodsend_warmup_request
Send a warmup request (``max_tokens=1`` by default).
lmcache/cli/commands/bench/engine_bench/request_sender.py:199
↓ 6 callersMethodserialize
(self)
lmcache/v1/protocol.py:179
↓ 6 callersMethodserialize
(self, src: MemoryObjGroup, dst: MemoryObj)
lmcache/v1/distributed/serde/asym_k16_v8.py:333
↓ 6 callersMethodset_responses
Set predefined responses.
tests/v1/cache_controller/conftest.py:107
↓ 6 callersMethodstart
Start the display daemon thread.
lmcache/cli/commands/bench/engine_bench/progress.py:54
↓ 6 callersMethodstart_full_sync
Start the full sync process. This method orchestrates the full sync by delegating to helper methods: 1. _initialize_sync: St
lmcache/v1/cache_controller/full_sync_sender.py:409
↓ 6 callersMethodstart_statistics
(self)
lmcache/v1/lookup_client/chunk_statistics_lookup_client.py:70
↓ 6 callersFunctionstore_keys
Store keys one at a time using the single-key API.
tests/v1/multiprocess/test_cache_server.py:184
↓ 6 callersFunctionstrip_shape_padding
Strip trailing-zero padding that was added by :func:`pad_shape_to_4d`. Trailing zeros are removed so that the original dimensionality is
lmcache/v1/protocol.py:127
↓ 6 callersMethodsubmit_exists
Submit an EXISTS check.
lmcache/v1/storage_backend/connector/valkey_connector.py:262
↓ 6 callersMethodsubmit_match
Submit a match query for a request once (idempotent per ``rid``). Args: rid: Request id, the poll key. model_scope: S
lmcache/v1/mp_coordinator/blend_client.py:163
↓ 6 callersMethodsubmit_prefetch_task
( self, key: CacheEngineKey, )
lmcache/v1/storage_backend/gds_backend.py:722
↓ 6 callersMethodsubmit_serialize
Submit a batch serialize task to the thread pool.
lmcache/v1/distributed/serde/async_processor.py:84
↓ 6 callersMethodsubmit_store_task
( self, keys: list[ObjectKey], objects: list[MemoryObj], )
examples/lmc_external_l2_adapter/src/lmc_external_l2_adapter/adapter.py:160
↓ 6 callersMethodsubmit_unlock
Release L2 locks acquired by lookup-and-lock.
lmcache/v1/distributed/l2_adapters/raw_block_l2_adapter.py:486
↓ 6 callersMethodto_dict
(self)
lmcache/v1/memory_management.py:152
↓ 6 callersMethodto_tensor
Return the wrapped local CUDA tensor (test-only).
tests/v1/platform/test_gpu_cache_context.py:185
↓ 6 callersMethodunpin
Unpin a key in storage.
tests/v1/storage_backend/test_storage_plugin.py:103
↓ 6 callersMethodupdate_interval_remote_time_to_put
(self, put_time: float)
lmcache/observability.py:545
↓ 6 callersFunctionwait_for_prefetch_status
Poll query_prefetch_status until it returns a non-None value. Returns the contiguous prefix-hit count (``count_leading_ones``) of the found b
tests/v1/distributed/test_distributed_storage_manager.py:158
↓ 6 callersFunctionwrite_resp
Write text to the specified output file (if any), otherwise to stdout.
benchmarks/long_doc_qa/long_doc_qa.py:203
↓ 5 callersFunctionBuildCBEngineArgs
BuildCBEngineArgs returns the server CLI args for the blend_v3 engine: the proven LMCacheEngine serialization (--host/--port/--l1-size-gb/--chunk-size
operator/internal/resources/cacheblend_engine.go:96
↓ 5 callersFunctionBuildCBEngineDaemonSet
BuildCBEngineDaemonSet constructs the DaemonSet for the blend_v3 engine of the given CacheBlendEngine. It reuses the shared GPU/security pod-template
operator/internal/resources/cacheblend_engine.go:115
↓ 5 callersFunctionCoordinatorSelectorLabels
CoordinatorSelectorLabels returns the immutable label subset used for the coordinator's pod/Service selectors. It carries a distinct app.kubernetes.io
operator/internal/resources/coordinator_deployment.go:61
↓ 5 callersFunctionCoordinatorStandardLabels
CoordinatorStandardLabels returns the full label set for coordinator-owned resources.
operator/internal/resources/coordinator_deployment.go:71
↓ 5 callersMethodCreateNixlStorageBackend
Create a Nixl backend with the given configuration. :param config: The LMCache engine configuration. :param loop: The asynci
lmcache/v1/storage_backend/nixl_storage_backend.py:958
↓ 5 callersFunctionRun
Run executes the provided command within this context
operator/test/utils/utils.go:103
↓ 5 callersMethod__init__
(self, metadata: MemoryObjMetadata)
lmcache/v1/memory_management.py:204
↓ 5 callersMethod__init__
( self, num_ops: int, concurrency: int, remote_url: str, use_odirect:
benchmarks/storage_backend_io/storage_backend_io_benchmark.py:951
↓ 5 callersFunction_append_range_or_elements
Helper to append range or individual elements based on length. Only compresses to [start, end] if there are 3 or more consecutive elements.
lmcache/utils.py:146
↓ 5 callersFunction_bare_gpu_module
A LMCacheDrivenTransferModule with only the liveness state initialized. Bypasses __init__ (which starts a CUDA host-func dispatcher) so the l
tests/v1/multiprocess/test_worker_liveness.py:34
↓ 5 callersFunction_block_mask_to_csr
Convert a 2D boolean block mask to CSR sparse format. Args: block_mask: [num_q_blocks, num_kv_blocks] bool tensor device: target
lmcache/v1/compute/attention/metadata.py:88
↓ 5 callersFunction_build_token_ids
Build token sequence: ``(seq_no, hello, hello, ...)``.
lmcache/cli/commands/bench/server_bench/helpers.py:163
↓ 5 callersFunction_build_transfer_limit_backend
( dev_path: str, max_data_transfer_size: int | None = None, )
tests/v1/storage_backend/test_raw_block_uring_cmd.py:137
↓ 5 callersFunction_byte_buffer
(num_bytes: int)
tests/v1/distributed/serde/test_asym_k16_v8_multi.py:46
↓ 5 callersMethod_cache_add
(self, chunk_hash: int)
lmcache/v1/storage_backend/nixl_storage_backend.py:1457
↓ 5 callersMethod_call
(agent: Mock, reg_list)
tests/v1/test_nixl_batched_contains.py:86
↓ 5 callersMethod_call
(backend: Mock, key: CacheEngineKey, pin: bool = False)
tests/v1/test_nixl_batched_contains.py:127
↓ 5 callersFunction_capture_workload_access_order
Run pass 1 + pass 2 and return the prefix-index access order of each. The mocked *sender* records which prefix index it was asked to send. Th
tests/cli/commands/bench/engine_bench/workloads/test_prefix_suffix_tuner.py:596
↓ 5 callersFunction_check_kv_equal
Dispatch to the correct comparison helper based on MLA mode.
tests/v1/test_xpu_sglang_connector.py:46
↓ 5 callersFunction_client
()
tests/v1/mp_coordinator/test_blend_directory_api.py:16
↓ 5 callersFunction_create_obj
Create a test TensorMemoryObj filled with *fill*.
examples/lmc_external_l2_adapter/tests/test_plugin.py:127
↓ 5 callersMethod_drain_all
Pop all queued events and dispatch to subscribers.
lmcache/v1/mp_observability/event_bus.py:312
↓ 5 callersMethod_ensure_peer_connection
( self, receiver_id: str, receiver_host: str, receiver_init_port: int,
lmcache/v1/storage_backend/pd_backend.py:420
↓ 5 callersFunction_entry
( instance_id: str, url: str, ip: str = "10.0.0.9", mq_port: int = 5555, )
tests/v1/multiprocess/test_p2p_controller.py:402
↓ 5 callersFunction_expected_kernel_group_shape
Compute the expected kernel-group buffer shape from the manager's public metadata (kv_size, num_layers, slots, hidden_dim).
tests/v1/platform/test_gpu_cache_context.py:132
↓ 5 callersMethod_finalize_load
(self, wrapped_id: L2TaskId, bitmap: Bitmap)
lmcache/v1/distributed/l2_adapters/serde_wrapper.py:647
↓ 5 callersMethod_finalize_store
( self, wrapped_id: L2TaskId, success: bool, bytes_transferred: int = 0, )
lmcache/v1/distributed/l2_adapters/serde_wrapper.py:631
↓ 5 callersFunction_find_model_meta
Find the GPU metadata entry matching *model_name*. Args: gpu_meta: The ``cache_context_meta`` dict from ``/status``. model_name:
lmcache/cli/commands/bench/engine_bench/config.py:122
↓ 5 callersFunction_flush
Wait for the drain thread to process pending events.
tests/v1/mp_observability/trace/test_recorder.py:61
↓ 5 callersMethod_free_lookup_locks
( self, token_ids: list[int], start: int, end: int, request_id: str,
lmcache/integration/sglang/multi_process_adapter.py:255
↓ 5 callersFunction_gb
Convert bytes to GiB.
lmcache/v1/mp_coordinator/http_apis/l2_api.py:39
↓ 5 callersMethod_get_all_workers_cached
Get all registered workers from the registry tree.
lmcache/v1/cache_controller/controllers/full_sync_tracker.py:76
↓ 5 callersMethod_get_interpreter
Get interpreter from first line comment
lmcache/v1/plugin/runtime_plugin_launcher.py:144
↓ 5 callersMethod_get_positive_int
Extract a positive integer from config with validation. Args: config_dict: Configuration dictionary key: Con
lmcache/v1/storage_backend/connector/sagemaker_hyperpod_adapter.py:186
↓ 5 callersMethod_get_row_filters_module
(self)
lmcache/v1/storage_backend/connector/bigtable_connector.py:174
↓ 5 callersFunction_install_fake_mooncake_extension
Install a fake lmcache_mooncake module for factory unit tests.
tests/v1/distributed/test_mooncake_store_l2_adapter.py:359
↓ 5 callersFunction_instance
(instance_id: str = "mp-1")
tests/v1/mp_coordinator/test_resync_manager.py:26
↓ 5 callersFunction_is_musa_contiguous_tensor
Return whether a tensor can be passed directly to MUSA native kernels.
lmcache/v1/platform/musa/native_kv_transfer.py:66
↓ 5 callersMethod_is_passive
A 'passive' CacheEngine means that the node itself will not store/retrieve the data directly, but from the "active" worker (i.e., ran
lmcache/v1/cache_engine.py:1912
↓ 5 callersFunction_key
(i: int)
tests/cli/commands/trace/test_dispatch.py:67
↓ 5 callersFunction_key
(salt: str, h: str = "aa", model: str = "m", rank: int = 0)
tests/v1/mp_coordinator/test_l2_api.py:17
↓ 5 callersFunction_make_3d_mla_kv
Create MLA-style 3D CPU tensors: ``[NB, BS, HS]`` (block_axis=0).
tests/v1/multiprocess/test_compute_mp_checksums.py:34
← previousnext →1,101–1,200 of 13,371, ranked by callers