EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

[model] Add mamba2 and Falcon-H1 support. (#10988)

Browse Source 
Co-authored-by: Younes Belkada <younes.belkada@tii.ae>
Co-authored-by: Younes B <49240599+younesbelkada@users.noreply.github.com>
This commit is contained in:
ilyasch2
2025-10-02 15:15:36 +04:00
committed by GitHub
parent b658be6f6a
commit 083629c235
18 changed files with 4533 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

147
test/srt/models/test_falcon_h1_models.py Normal file
View File

@@ -0,0 +1,147 @@
import unittest
from types import SimpleNamespace
from sglang.srt.utils import kill_process_tree
from sglang.test.few_shot_gsm8k import run_eval
from sglang.test.test_utils import (
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
DEFAULT_URL_FOR_TEST,
CustomTestCase,
popen_launch_server,
)
class TestFalconH1(CustomTestCase):
@classmethod
def setUpClass(cls):
cls.model = "tiiuae/Falcon-H1-0.5B-Instruct"
cls.base_url = DEFAULT_URL_FOR_TEST
cls.process = popen_launch_server(
cls.model,
cls.base_url,
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
other_args=[
"--tensor-parallel-size",
"1",
],
)
@classmethod
def tearDownClass(cls):
kill_process_tree(cls.process.pid)
def test_gsm8k(self):
args = SimpleNamespace(
num_shots=5,
data_path=None,
num_questions=200,
max_new_tokens=512,
parallel=128,
host="http://127.0.0.1",
port=int(self.base_url.split(":")[-1]),
)
metrics = run_eval(args)
print(f"{metrics=}")
self.assertGreater(metrics["accuracy"], 0.74)
class TestFalconH1TP4(CustomTestCase):
@classmethod
def setUpClass(cls):
cls.model = "tiiuae/Falcon-H1-0.5B-Instruct"
cls.base_url = DEFAULT_URL_FOR_TEST
cls.process = popen_launch_server(
cls.model,
cls.base_url,
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
other_args=[
"--tensor-parallel-size",
"4",
],
)
@classmethod
def tearDownClass(cls):
kill_process_tree(cls.process.pid)
def test_gsm8k(self):
args = SimpleNamespace(
num_shots=5,
data_path=None,
num_questions=200,
max_new_tokens=512,
parallel=128,
host="http://127.0.0.1",
port=int(self.base_url.split(":")[-1]),
)
metrics = run_eval(args)
print(f"{metrics=}")
self.assertGreater(metrics["accuracy"], 0.74)
class TestFalconH1NoGatedRMS(CustomTestCase):
@classmethod
def setUpClass(cls):
cls.model = "tiiuae/Falcon-H1-1.5B-Instruct"
cls.base_url = DEFAULT_URL_FOR_TEST
cls.process = popen_launch_server(
cls.model,
cls.base_url,
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
other_args=[
"--tensor-parallel-size",
"1",
],
)
@classmethod
def tearDownClass(cls):
kill_process_tree(cls.process.pid)
def test_gsm8k(self):
args = SimpleNamespace(
num_shots=5,
data_path=None,
num_questions=200,
max_new_tokens=512,
parallel=128,
host="http://127.0.0.1",
port=int(self.base_url.split(":")[-1]),
)
metrics = run_eval(args)
print(f"{metrics=}")
self.assertGreater(metrics["accuracy"], 0.74)
class TestFalconH1NoGatedTP4(CustomTestCase):
@classmethod
def setUpClass(cls):
cls.model = "tiiuae/Falcon-H1-1.5B-Instruct"
cls.base_url = DEFAULT_URL_FOR_TEST
cls.process = popen_launch_server(
cls.model,
cls.base_url,
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
other_args=[
"--tensor-parallel-size",
"4",
],
)
@classmethod
def tearDownClass(cls):
kill_process_tree(cls.process.pid)
def test_gsm8k(self):
args = SimpleNamespace(
num_shots=5,
data_path=None,
num_questions=200,
max_new_tokens=512,
parallel=128,
host="http://127.0.0.1",
port=int(self.base_url.split(":")[-1]),
)
metrics = run_eval(args)
print(f"{metrics=}")
self.assertGreater(metrics["accuracy"], 0.74)

1
test/srt/run_suite.py
View File

@@ -140,6 +140,7 @@ suites = {
TestFile("test_local_attn.py", 250),
TestFile("test_pp_single_node.py", 372),
TestFile("models/test_qwen3_next_models.py", 200),
TestFile("models/test_falcon_h1_models.py", 200),
TestFile("test_multi_instance_release_memory_occupation.py", 64),
],
"per-commit-8-gpu": [