MCPcopy Index your code
hub / github.com/PaddlePaddle/FastDeploy / CacheConfig

Class CacheConfig

fastdeploy/config.py:1380–1554  ·  view source on GitHub ↗

Configuration for the KV cache. Attributes: block_size (int): Size of a cache block in number of tokens. gpu_memory_utilization (float): Fraction of GPU memory to use for model execution. cache_dtype (str): Data type for kv cache storage. Default is 'bfloat16'.

Source from the content-addressed store, hash-verified

1378
1379
1380class CacheConfig:
1381 """
1382 Configuration for the KV cache.
1383
1384 Attributes:
1385 block_size (int): Size of a cache block in number of tokens.
1386 gpu_memory_utilization (float): Fraction of GPU memory to use for model execution.
1387 cache_dtype (str): Data type for kv cache storage. Default is 'bfloat16'.
1388 num_gpu_blocks_override (Optional[int]): Number of GPU blocks to use.
1389 Overrides profiled num_gpu_blocks if provided.
1390 kv_cache_ratio (float): Ratio for calculating the maximum block number.
1391 enc_dec_block_num (int): Number of encoder-decoder blocks.
1392 prealloc_dec_block_slot_num_threshold (int): Number of token slot threadshold to allocate next blocks for decoding.
1393 enable_prefix_caching (bool): Flag to enable prefix caching.
1394 enable_output_caching (bool): Flag to enable kv cache output tokens, only works in V1 scheduler.
1395 """
1396
1397 def __init__(self, args):
1398 """
1399 Initialize the CacheConfig class.
1400
1401 Args:
1402 block_size (int): Size of a cache block in number of tokens.
1403 gpu_memory_utilization (float): Fraction of GPU memory to use.
1404 cache_dtype (str): Data type for cache storage. Default is 'bfloat16'.
1405 num_gpu_blocks_override (Optional[int]): Override for number of GPU blocks.
1406 num_cpu_blocks (Optional[int]): Number of CPU blocks.
1407 kv_cache_ratio (float): Ratio for max block calculation.
1408 enc_dec_block_num (int): Number of encoder-decoder blocks.
1409 prealloc_dec_block_slot_num_threshold (int): Number of token slot threshold to allocate next blocks for decoding, used when ENABLE_V1_KVCACHE_SCHEDULER=1.
1410 enable_prefix_caching (bool): Enable prefix caching.
1411 max_encoder_cache(int): Maximum number of tokens in the encoder cache.
1412 max_processor_cache(int): Maximum number of bytes in the processor cache.
1413 """
1414 self.block_size = 64
1415 self.gpu_memory_utilization = 0.9
1416 self.num_gpu_blocks_override = None
1417 if envs.ENABLE_V1_KVCACHE_SCHEDULER:
1418 self.kv_cache_ratio = 1.0
1419 else:
1420 self.kv_cache_ratio = 0.75
1421 self.enc_dec_block_num = envs.FD_ENC_DEC_BLOCK_NUM
1422 self.prealloc_dec_block_slot_num_threshold = 12
1423 self.cache_dtype = "bfloat16"
1424 self.model_cfg = None
1425 self.enable_chunked_prefill = False
1426 self.rdma_comm_ports = None
1427 self.local_rdma_comm_ports = None
1428 self.cache_transfer_protocol = None
1429 self.pd_comm_port = None
1430 self.local_pd_comm_port = None
1431 self.enable_prefix_caching = False
1432 self.enable_ssd_cache = False
1433 self.cache_queue_port = None
1434 self.local_cache_queue_port = None
1435 self.swap_space = None
1436 self.max_encoder_cache = None
1437 self.max_processor_cache = None

Callers 15

create_engine_configMethod · 0.90
initialize_fd_configFunction · 0.90
test_fdconfig_nnodeMethod · 0.90
test_fdconfig_ipsMethod · 0.90
real_fd_configFunction · 0.90
mock_fd_configFunction · 0.90
mock_fd_config_tp2Function · 0.90

Calls

no outgoing calls

Tested by 15

test_fdconfig_nnodeMethod · 0.72
test_fdconfig_ipsMethod · 0.72
real_fd_configFunction · 0.72
mock_fd_configFunction · 0.72
mock_fd_config_tp2Function · 0.72
mock_fd_config_qwen3Function · 0.72
get_fd_configFunction · 0.72