↓ 2 callersMethodconcrete_shape_str``describe_shape`` with real dims, e.g. ``80 x [2, 2048, 128, 8, 128]``.
lmcache/v1/gpu_connector/kv_format/specs/base.py:172
↓ 2 callersFunctioncreate_memory_objects(
kv_dim: int,
nl: int,
tokens_per_object: int,
hidden_dim: int,
num_objects: int,
dty
tests/v1/test_mp_mem_kernels.py:145
↓ 2 callersFunctioncreate_vllm_tensors(
engine_kv_format,
nl: int,
nb: int,
bs: int,
nh: int,
hs: int,
dtype: torch.dtyp
tests/v1/test_mp_mem_kernels.py:69
↓ 2 callersFunctioncreate_zero_vllm_tensors(
engine_kv_format,
nl: int,
nb: int,
bs: int,
nh: int,
hs: int,
dtype: torch.dtyp
tests/v1/test_mp_mem_kernels.py:107
↓ 2 callersMethodexport_csvWrite per-qualname stats to a CSV file. Columns: ``qualname, count, errors, mean_ms, p50_ms, p90_ms, p99_ms, min_ms, max_ms``. One r
lmcache/cli/commands/trace/_stats.py:208