MCPcopy Create free account

hub / github.com/LMCache/LMCache / functions

Functions13,371 in github.com/LMCache/LMCache

↓ 9 callersFunctioncreateTestNamespace
createTestNamespace creates a fresh namespace whose name embeds the running spec's text to make failing-spec dumps easier to attribute. The deferred A
operator/test/e2e/smoke_helpers_test.go:62
↓ 9 callersFunctioncreate_engine_group_infos_from_vllm
Build the LMCache engine group infos from vLLM metadata and registered tensors. This is the single entry point for the vLLM -> LMCache conversion
lmcache/integration/vllm/kv_cache_groups.py:90
↓ 9 callersFunctioncreate_memory_obj
(size: int = 256, fill_value: float = 1.0)
tests/v1/distributed/test_mooncake_store_l2_adapter.py:76
↓ 9 callersFunctioncreate_test_config
(gds_path: str, gds_path_sharding: str = "by_gpu")
tests/v1/storage_backend/test_gds_backend.py:42
↓ 9 callersFunctioncreate_test_key
Create a deterministic cache key for unit tests.
tests/v1/storage_backend/test_hfbucket_connector.py:156
↓ 9 callersMethoddelete
(self, keys: list[ObjectKey])
tests/v1/multiprocess/http_apis/test_l2_api.py:46
↓ 9 callersFunctionderefBool
derefBool returns the value pointed to by p, or def if p is nil.
operator/internal/resources/helpers.go:36
↓ 9 callersFunctionemit_salt_counts
Add to *counter* once per entry in *salt_counts*. *salt_counts* maps ``cache_salt`` values to key counts. Empty salt produces a dimensionless
lmcache/v1/mp_observability/subscribers/metrics/utils.py:20
↓ 9 callersFunctionenvDefault
envDefault returns os.Getenv(key) or def if the env var is unset/empty. Pulled out as a helper so package-level var initialisers stay readable.
operator/test/e2e/e2e_suite_test.go:65
↓ 9 callersMethodexecute
Execute the mock command. Args: args: Parsed CLI arguments containing ``name``, ``num_items``, and optionally ``o
lmcache/cli/commands/mock.py:54
↓ 9 callersFunctionfold_unfold
:func:`fold_unfold_ranked` for the single-rank (group-major) layout. Args: found: group-major presence bitmap of length ``len
lmcache/v1/distributed/bitmap_ops/fold.py:194
↓ 9 callersMethodgenerate_test_data
Generate test data based on configuration Each process gets a unique range of instance IDs to avoid conflicts. Format: instance_p{pro
lmcache/tools/controller_benchmark/benchmark.py:165
↓ 9 callersMethodget
(self, key: CacheEngineKey)
lmcache/v1/storage_backend/connector/redis_connector.py:91
↓ 9 callersMethodget
Single get method - NOT SUPPORTED. Use batched_get instead for all operations.
lmcache/v1/storage_backend/connector/mooncakestore_connector.py:634
↓ 9 callersFunctionget_device
Return the device of the KV cache tensors. Descends into any list nesting until a tensor is found; assumes all tensors in *kv_caches* live on
lmcache/v1/gpu_connector/utils.py:391
↓ 9 callersFunctionget_gds_context
Return the process-global :class:`GDSContext` singleton (created empty on first access). Consult :attr:`GDSContext.initialized` to tell whether GD
lmcache/v1/gpu_connector/gds_context.py:392
↓ 9 callersFunctionget_group_data_ptrs
Return device pointers for a group of layers in kernel-expected order. See :meth:`KVFormatSpec.data_ptrs` for the per-format pointer-array sh
lmcache/v1/gpu_connector/utils.py:323
↓ 9 callersMethodget_monitored_count
Get the number of currently monitored pinned objects.
lmcache/v1/pin_monitor.py:247
↓ 9 callersMethodget_next_sequence_number
Get monotonically increasing sequence number for specific instance-worker-location
lmcache/tools/controller_benchmark/benchmark.py:182
↓ 9 callersMethodget_shape
(self, num_tokens: int)
lmcache/v1/gpu_connector/gpu_connectors.py:1623
↓ 9 callersMethodget_total_kv_count
Get total count of KV chunks across all workers.
lmcache/v1/cache_controller/utils.py:628
↓ 9 callersMethodget_transfer_channel_address
(self, lmcache_addresses)
tests/v1/distributed/transfer_channel/test_factory.py:51
↓ 9 callersMethodget_workers
Get worker ids given an instance id.
lmcache/v1/cache_controller/controllers/registration_controller.py:84
↓ 9 callersMethodhandle_batched_kv_operations
Handle batched KV operations by forwarding to WorkerNode. Args: msg: The batched KV operation message. is_fu
lmcache/v1/cache_controller/utils.py:554
↓ 9 callersMethodl2_adapters
(self)
tests/v1/multiprocess/http_apis/test_l2_api.py:75
↓ 9 callersFunctionmake_object_key
Create a test ObjectKey with the given chunk ID.
tests/v1/distributed/serde/test_serde_e2e.py:55
↓ 9 callersFunctionmake_raw_block_file
Create a fixed-size file for raw block backend tests. Args: tmp_path: Pytest temporary directory used to place the backing file.
tests/v1/storage_backend/raw_block_test_utils.py:36
↓ 9 callersFunctionmustCreateNS
mustCreateNS creates a namespace and registers cleanup. Returns the name so callers can put resources in it. We don't rely on namespace teardown to cl
operator/internal/controller/reconcile_helpers_test.go:57
↓ 9 callersMethodname
Return the name of this health check
lmcache/v1/health_monitor/base.py:76
↓ 9 callersMethodnotify
Signal the notifier to wake any blocked poll() waiter.
lmcache/v1/distributed/storage_controllers/store_controller.py:95
↓ 9 callersFunctionparse_args_to_config
Convert parsed command line arguments to a StorageManagerConfig. Args: args: Parsed arguments from the argument parser. Returns
lmcache/v1/distributed/config.py:538
↓ 9 callersFunctionparse_remote_url
Parses the remote URL into its constituent parts with support for: - Multiple hosts (comma-separated) - Path and query parameters in each
lmcache/v1/storage_backend/connector/__init__.py:40
↓ 9 callersMethodread_prefetched_results
(self, keys: list[ObjectKey])
tests/cli/commands/trace/test_dispatch.py:51
↓ 9 callersMethodregister_gpu_buffer
Register a staging buffer (and its CUDA stream) with cuFile. Registered as contiguous <=16 MiB regions (the cuFileBufRegister cap); :
lmcache/v1/gpu_connector/gds_context.py:126
↓ 9 callersMethodreset
(self)
tests/v1/distributed/test_s3_l2_adapter.py:56
↓ 9 callersMethodset_device
Select the given device. Args: device: The device index or identifier to select. Ignored in the stub.
lmcache/v1/platform/cpu/stub_cpu_device.py:306
↓ 9 callersMethodset_used_size
(self, n: int)
tests/v1/distributed/serde/test_async_processor.py:174
↓ 9 callersMethodshutdown
Shut down the pool. Sends a shutdown sentinel to every worker. If *wait* is true, blocks until all workers have exited.
lmcache/v1/multiprocess/affinity_pool.py:92
↓ 9 callersFunctionsubmit_callback_to_stream
Schedule a stream-ordered host callback for *kind* with *payload*. Runs on the dispatcher worker registered for *kind*; never acquires the GIL
lmcache/v1/multiprocess/native_completion.py:144
↓ 9 callersMethodsubmit_store_request
Submit a KV cache store request to LMCache Args: request_id: The ID of the request op: The LoadStoreOp descr
lmcache/integration/vllm/vllm_multi_process_adapter.py:1381
↓ 9 callersMethodsubscribe
Register a callback for a specific event type (thread-safe).
lmcache/v1/mp_observability/event_bus.py:145
↓ 9 callersMethodtransfer_async
DMA a chunk between ``gpu_buffer`` and its slab region. ``READ`` pulls slab -> ``gpu_buffer``; ``WRITE`` pushes the reverse. Split at
lmcache/v1/gpu_connector/gds_context.py:180
↓ 9 callersFunctionunfold_range
Chunk range ``[lo, hi)`` one object group needs to serve a prefix. Full attention needs ``[0, prefix_len)``; a window of ``w`` chunks needs `
lmcache/v1/distributed/bitmap_ops/fold.py:46
↓ 9 callersFunctionuniqueNS
uniqueNS returns a fresh namespace name unique within this test process. The prefix makes it easy to spot the source of a stray object if a test crash
operator/internal/controller/reconcile_helpers_test.go:48
↓ 9 callersFunctionunique_hits_direct_id_numba
Perform direct-address lookup with deduplication of results. This function looks up each hash in a direct-address table using the lower
lmcache/v1/multiprocess/token_hasher.py:392
↓ 9 callersFunctionupdate_table_id_numba
Update the direct-address table with new ID values for given hashes. For each hash in `hashes_u64`, compute the index as: idx = has
lmcache/v1/multiprocess/token_hasher.py:355
↓ 8 callersFunction_bf16_tensor
(*shape: int, seed: int)
tests/v1/distributed/serde/test_asym_k16_v8_v_only.py:56
↓ 8 callersFunction_bf16_tensor_obj
(*shape: int, seed: int = 0)
tests/v1/distributed/serde/test_multi.py:69
↓ 8 callersFunction_byte_buffer
(num_bytes: int)
tests/v1/distributed/serde/test_multi.py:65
↓ 8 callersMethod_check_and_evict
(self, state: L2AdapterEvictionState)
lmcache/v1/distributed/storage_controllers/eviction_controller.py:296
↓ 8 callersFunction_client
(handler)
tests/v1/mp_coordinator/test_registrar.py:22
↓ 8 callersMethod_condition_met
Check whether an item's condition is satisfied.
lmcache/cli/commands/bench/engine_bench/interactive/state.py:323
↓ 8 callersFunction_config
(size_bytes: int)
tests/v1/distributed/memory_manager/test_gds_l1_memory_manager.py:22
↓ 8 callersFunction_create_json_response
(data: dict, status_code: int = 200)
lmcache/v1/internal_api_server/vllm/chunk_statistics_api.py:21
↓ 8 callersFunction_create_random_tensor
( shape: list, dtype: torch.dtype, device: torch.device )
tests/v1/test_mp_mem_kernels.py:29
↓ 8 callersFunction_create_zero_tensor
( shape: list, dtype: torch.dtype, device: torch.device )
tests/v1/test_mp_mem_kernels.py:37
↓ 8 callersFunction_enumerate_permutations
Enumerate up to ``num_permutations`` distinct permutations. Returns all N! permutations when num_permutations >= N!. For large N uses random
lmcache/cli/commands/bench/engine_bench/workloads/long_doc_permutator.py:194
↓ 8 callersFunction_filename_to_object_key
Reverse ``_object_key_to_filename``. Accepts both the 4-field unsalted shape and the 5-field salted shape (trailing ``cache_salt``). Returns
lmcache/v1/distributed/l2_adapters/fs_l2_adapter.py:123
↓ 8 callersMethod_get_file_path
Get file path for the given key
lmcache/v1/storage_backend/connector/fs_connector.py:148
↓ 8 callersMethod_get_lookup_type_str
Get type string for lookup client/server, including wrapper info. Returns format: "OuterWrapper(InnerWrapper(CoreType))" or "None"
lmcache/v1/manager.py:469
↓ 8 callersMethod_get_table
Lazy initialization and caching of TableAsync instance.
lmcache/v1/storage_backend/connector/bigtable_connector.py:158
↓ 8 callersFunction_grp
Cast a tuple of test fakes to the production MemoryObjGroup type.
tests/v1/distributed/serde/test_asym_k16_v8_multi.py:52
↓ 8 callersFunction_l1_write
(keys: list)
tests/v1/mp_observability/subscribers/metrics/test_sm_lifecycle.py:66
↓ 8 callersFunction_make_key
Create an ObjectKey compatible with both old and new API signatures.
examples/lmc_external_l2_adapter/tests/test_plugin.py:61
↓ 8 callersFunction_make_stub_backend
Return a NixlDynamicStorageBackend instance with __init__ bypassed.
tests/v1/test_device_id_race.py:26
↓ 8 callersMethod_on_store_end
(self, event: Event)
lmcache/v1/mp_observability/subscribers/logging/mp_server.py:43
↓ 8 callersFunction_read_counters_by_attrs
Snapshot counter values keyed by (metric_name, frozenset(attrs)).
tests/v1/mp_observability/subscribers/metrics/test_l1.py:32
↓ 8 callersFunction_same_shape_caches
(names: list[str])
tests/v1/test_vllm_kv_cache_groups.py:63
↓ 8 callersMethod_set_cached_object_size
Store an object size in the TTL cache.
lmcache/v1/storage_backend/connector/hfbucket_connector.py:688
↓ 8 callersFunction_store_and_wait
Helper: store one key and wait for the store event fd to fire.
tests/v1/distributed/test_mock_l2_adapter.py:661
↓ 8 callersMethod_table
(self, size: int = 8)
tests/v1/multiprocess/test_token_hasher.py:308
↓ 8 callersMethodallocate
Reserve ``count`` slab regions for the given layout. All-or-nothing: on the first slab OOM, frees what was reserved and returns ``(L1
lmcache/v1/distributed/memory_manager/gds_l1_memory_manager.py:54
↓ 8 callersMethodallocate
Allocate a memory object of shape and dtype evict if necessary. Storage manager should always call local_cpu_backend.allocate
lmcache/v1/storage_backend/local_cpu_backend.py:614
↓ 8 callersFunctionattempt_permute_to_contiguous_view
Return a contiguous view of *kv_caches*, metadata-only (no copy). For a tensor leaf: reorders the dims by stride magnitude so shape lines up
lmcache/v1/gpu_connector/kv_format/contiguity.py:20
↓ 8 callersFunctionbenchmark
()
tests/conftest.py:36
↓ 8 callersMethodclose
(self)
tests/v1/distributed/serde/test_factory.py:102
↓ 8 callersMethodclose
Close the batched message sender and flush remaining messages.
lmcache/v1/storage_backend/batched_message_sender.py:211
↓ 8 callersFunctionclose_zmq_socket
Close a ZeroMQ socket cleanly. :param socket: The zmq.Socket to be closed. :param linger: LINGER period (in milliseconds). Default i
lmcache/v1/rpc_utils.py:74
↓ 8 callersFunctionconditionBool
(b bool)
operator/internal/controller/reconcile_helpers.go:403
↓ 8 callersFunctioncreate_connector
Create a connector with an in-memory fake bucket client.
tests/v1/storage_backend/test_hfbucket_connector.py:167
↓ 8 callersFunctioncreate_mock_memory_obj
Helper to create a mock MemoryObj with proper structure
tests/v1/storage_backend/test_audit_connector.py:37
↓ 8 callersFunctioncreate_test_config
Create a test configuration for LocalCPUBackend.
tests/v1/storage_backend/test_batched_message_sender.py:42
↓ 8 callersFunctioncreate_test_metadata
Create a test metadata for LMCacheMetadata. Returns: LMCacheMetadata with test configuration
tests/v1/storage_backend/test_hf3fs_connector.py:119
↓ 8 callersFunctioncreate_transfer_context
Create a transfer context from KV cache device type. The device check is intentionally centralized here. Routing can be overridden via the ``
lmcache/v1/multiprocess/transfer_context/worker_transfer.py:503
↓ 8 callersFunctiondecode_tokens
Decode a base64 token string produced by :func:`encode_tokens`. Args: tokens_b64: Base64 of a little-endian ``uint32`` token buffer.
lmcache/v1/mp_coordinator/schemas.py:39
↓ 8 callersMethoddelete
Delete a batch of objects from L2 storage. Args: keys (list[ObjectKey]): The keys of the objects to delete. Not
lmcache/v1/distributed/l2_adapters/base.py:459
↓ 8 callersFunctiondelete_transfer_channel_context
Delete the global transfer channel context, if it exists.
lmcache/v1/distributed/transfer_channel/__init__.py:99
↓ 8 callersMethoddispatch
Invoke the handler for *qualname*. Args: qualname: The recorded qualname. context: Replay context passed through to t
lmcache/cli/commands/trace/_dispatch.py:111
↓ 8 callersFunctionencode_legacy_key
Encode a legacy non-MP cache key for raw-block storage. Args: key: Legacy cache key from the non-MP storage plugin path. Returns:
lmcache/v1/storage_backend/raw_block/key_codec.py:122
↓ 8 callersMethodexecute
Execute this command on the worker. Override in subclasses.
lmcache/v1/cache_controller/commands/base.py:31
↓ 8 callersMethodexists
(self, key)
tests/conftest.py:187
↓ 8 callersMethodfinish_write_and_reserve_read
Atomically finish write and acquire read lock for the given keys. This is used by the prefetch controller after successfully loading
lmcache/v1/distributed/l1_manager.py:589
↓ 8 callersFunctiongather_paged_kv_to_cpu
Gather paged KV blocks into CPU chunk tensors. Args: kv_caches: Per-layer KV tensor mapping. block_ids: Flattened block IDs for a
lmcache/v1/multiprocess/transfer_context/base.py:302
↓ 8 callersMethodget_active_storage_backends
Get the active storage backends based on freeze mode, bypass mode, and filters. :param Optional[str] location: If specified, only yi
lmcache/v1/storage_backend/storage_manager.py:1161
↓ 8 callersMethodget_and_touch_context_entry
Return the entry for ``instance_id``, refreshing its last-seen time. The refresh keeps an actively transferring worker from being reaped
lmcache/v1/multiprocess/modules/lmcache_driven_transfer.py:455
↓ 8 callersMethodget_blocking
Blocking get function.
lmcache/v1/storage_backend/remote_backend.py:341
↓ 8 callersMethodget_eviction_actions
Select victims in LRU order, scoped to a single ``cache_salt``. IsolatedLRU is "isolated only" by contract — callers must pass a conc
lmcache/v1/distributed/eviction_policy/isolated_lru.py:106
↓ 8 callersFunctionget_head_size
Return the head size for a layer (defaults to layer 0).
lmcache/v1/gpu_connector/utils.py:291
↓ 8 callersFunctionget_num_heads
Return the number of heads for a layer (defaults to layer 0).
lmcache/v1/gpu_connector/utils.py:273
↓ 8 callersMethodget_socket
Get the socket for a given instance and worker ID.
lmcache/v1/cache_controller/controllers/registration_controller.py:57
← previousnext →701–800 of 13,371, ranked by callers