minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 14 Packages Projects Releases Wiki Activity

Support spec v2 for Frozen-KV MTP; remove v1 worker (#27607)

Browse Source 
Co-authored-by: Khoa Pham <khoa.pham@radixark.ai>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This commit is contained in:
Liangsheng Yin
2026-06-09 15:30:20 -07:00
committed by GitHub
co-authored by Khoa Pham Claude Opus 4.8
parent 7f730edfdc
commit decb88e0e3
13 changed files with 750 additions and 1074 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/registered/spec/test_frozen_kv_mtp.py
-8
View File
@@ -1,4 +1,3 @@
import os
import unittest
from types import SimpleNamespace
from typing import Optional
@@ -43,12 +42,6 @@ def get_avg_spec_accept_length(base_url: str) -> Optional[float]:
class TestFrozenKVMTP(CustomTestCase):
base_url = DEFAULT_URL_FOR_TEST
@classmethod
def _server_env(cls) -> dict[str, str]:
env = dict(os.environ)
env["SGLANG_ENABLE_SPEC_V2"] = "0"
return env
@classmethod
def _common_server_args(cls) -> list[str]:
args = [
@@ -110,7 +103,6 @@ class TestFrozenKVMTP(CustomTestCase):
"google/gemma-4-E4B-it",
self.base_url,
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH * 3,
env=self._server_env(),
other_args=self._server_args(topk),
)
requests.get(self.base_url + "/flush_cache", timeout=30)
test/registered/spec/test_gemma4_mtp_26b_a4b_extra.py
-8
View File
@@ -1,4 +1,3 @@
import os
import unittest
from types import SimpleNamespace
from typing import Optional
@@ -68,12 +67,6 @@ def get_avg_spec_accept_length(base_url: str) -> Optional[float]:
class TestGemma4MTP26BA4B(CustomTestCase):
base_url = DEFAULT_URL_FOR_TEST
@classmethod
def _server_env(cls) -> dict[str, str]:
env = dict(os.environ)
env["SGLANG_ENABLE_SPEC_V2"] = "0"
return env
@classmethod
def _common_server_args(cls) -> list[str]:
args = [
@@ -140,7 +133,6 @@ class TestGemma4MTP26BA4B(CustomTestCase):
TARGET_PATH,
self.base_url,
timeout=SERVER_LAUNCH_TIMEOUT,
env=self._server_env(),
other_args=self._server_args(topk),
)
requests.get(self.base_url + "/flush_cache", timeout=30)
test/registered/spec/test_gemma4_mtp_31b_extra.py
-8
View File
@@ -1,4 +1,3 @@
import os
import unittest
from types import SimpleNamespace
from typing import Optional
@@ -58,12 +57,6 @@ def get_avg_spec_accept_length(base_url: str) -> Optional[float]:
class TestGemma4MTP31B(CustomTestCase):
base_url = DEFAULT_URL_FOR_TEST
@classmethod
def _server_env(cls) -> dict[str, str]:
env = dict(os.environ)
env["SGLANG_ENABLE_SPEC_V2"] = "0"
return env
@classmethod
def _common_server_args(cls) -> list[str]:
args = [
@@ -127,7 +120,6 @@ class TestGemma4MTP31B(CustomTestCase):
TARGET_PATH,
self.base_url,
timeout=SERVER_LAUNCH_TIMEOUT,
env=self._server_env(),
other_args=self._server_args(topk),
)
requests.get(self.base_url + "/flush_cache", timeout=30)