[Radix Cache] Add Rust TreeCore backend with shared parity tests (#32710)
Co-authored-by: alphabetc1 <2508695655@qq.com> Co-authored-by: ispobock <ispobaoke@gmail.com>
This commit is contained in:
co-authored by
alphabetc1
ispobock
parent
52e1c24744
commit
9cf157c252
@@ -61,7 +61,9 @@ from sglang.srt.arg_groups.serving_hook import (
|
||||
ssl_verify_of,
|
||||
)
|
||||
from sglang.srt.arg_groups.speculative_hook import handle_speculative_decoding
|
||||
from sglang.srt.arg_groups.validation_hook import check_two_batch_overlap
|
||||
from sglang.srt.arg_groups.validation_hook import (
|
||||
check_two_batch_overlap,
|
||||
)
|
||||
from sglang.srt.entrypoints.sidecar import (
|
||||
SGLANG_GRPC_ENDPOINT_ENV,
|
||||
Sidecar,
|
||||
@@ -789,6 +791,19 @@ class TestLoadBalanceMethod(unittest.TestCase):
|
||||
"mooncake",
|
||||
)
|
||||
|
||||
def test_pd_decode_hicache_allows_rust_tree_core(self):
|
||||
server_args = ServerArgs(
|
||||
model_path="dummy",
|
||||
disaggregation_mode="decode",
|
||||
disaggregation_decode_enable_radix_cache=True,
|
||||
disaggregation_transfer_backend="nixl",
|
||||
enable_hierarchical_cache=True,
|
||||
)
|
||||
with envs.SGLANG_UNIFIED_RADIX_TREE_CORE_BACKEND.override("rust"):
|
||||
handle_pd_disaggregation(server_args)
|
||||
|
||||
self.assertFalse(resolution_result(server_args, "disable_radix_cache"))
|
||||
|
||||
|
||||
class TestSkipTokenizerInit(unittest.TestCase):
|
||||
def test_skip_tokenizer_worker_counts(self):
|
||||
@@ -1475,6 +1490,16 @@ class TestHiCacheArgs(unittest.TestCase):
|
||||
expected_decode_backend,
|
||||
)
|
||||
|
||||
def test_buffer_only_accepts_both_tree_cores(self):
|
||||
for backend in ("python", "rust"):
|
||||
args = self._make_args(
|
||||
enable_hierarchical_cache=True,
|
||||
hicache_host_memory_mode="buffer_only",
|
||||
hicache_storage_backend="file",
|
||||
)
|
||||
with envs.SGLANG_UNIFIED_RADIX_TREE_CORE_BACKEND.override(backend):
|
||||
handle_hicache(args)
|
||||
|
||||
def test_hicache_io_backend_and_mem_layout_compatibility(self):
|
||||
cases = [
|
||||
{
|
||||
|
||||
Reference in New Issue
Block a user