minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[CI] Reorganize stage-b 1-GPU tests for 5090 compatibility (#16826)

Browse Source
This commit is contained in:
Alison Shao
2026-01-15 15:23:35 -08:00
committed by GitHub
parent 8b22deef5b
commit 146b5fcc84
136 changed files with 277 additions and 404 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/registered/models/test_compressed_tensors_models.py
+4 -4
View File
@@ -1,13 +1,10 @@
from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
# Model tests for compressed tensors (FP8)
register_cuda_ci(est_time=42, suite="stage-b-test-small-1-gpu")
register_amd_ci(est_time=42, suite="stage-b-test-small-1-gpu-amd")
import unittest
from types import SimpleNamespace
from sglang.srt.utils import kill_process_tree
from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
from sglang.test.few_shot_gsm8k import run_eval
from sglang.test.test_utils import (
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -16,6 +13,9 @@ from sglang.test.test_utils import (
popen_launch_server,
)
register_cuda_ci(est_time=42, suite="stage-b-test-large-1-gpu")
register_amd_ci(est_time=42, suite="stage-b-test-small-1-gpu-amd")
class TestCompressedTensorsLlama3FP8(CustomTestCase):
@classmethod
test/registered/models/test_cross_encoder_models.py
+7 -7
View File
@@ -1,19 +1,19 @@
from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
# Cross encoder model tests
register_cuda_ci(est_time=100, suite="stage-b-test-small-1-gpu")
register_cuda_ci(est_time=105, suite="stage-b-test-small-1-gpu-5090")
register_amd_ci(est_time=150, suite="stage-b-test-small-1-gpu-amd")
import multiprocessing as mp
import random
import unittest
import torch
from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
from sglang.test.runners import TEST_RERANK_QUERY_DOCS, HFRunner, SRTRunner
from sglang.test.test_utils import CustomTestCase, is_in_ci
# Cross encoder model tests
register_cuda_ci(est_time=100, suite="stage-b-test-small-1-gpu")
register_amd_ci(est_time=150, suite="stage-b-test-small-1-gpu-amd")
MODELS = [
("cross-encoder/ms-marco-MiniLM-L6-v2", 1, 1e-2),
("BAAI/bge-reranker-v2-m3", 1, 1e-2),
test/registered/models/test_embedding_models.py
+9 -11
View File
@@ -1,14 +1,3 @@
from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
# Embedding model tests
register_cuda_ci(est_time=73, suite="stage-b-test-small-1-gpu")
register_cuda_ci(est_time=58, suite="stage-b-test-small-1-gpu-5090")
register_amd_ci(
est_time=73,
suite="stage-b-test-small-1-gpu-amd",
disabled="see https://github.com/sgl-project/sglang/issues/11127",
)
# Copyright 2023-2024 SGLang Team
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
@@ -31,6 +20,7 @@ from typing import Optional
import torch
from transformers import AutoConfig, AutoTokenizer
from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
from sglang.test.runners import DEFAULT_PROMPTS, HFRunner, SRTRunner
from sglang.test.test_utils import (
CustomTestCase,
@@ -39,6 +29,14 @@ from sglang.test.test_utils import (
is_in_ci,
)
# Embedding model tests
register_amd_ci(
est_time=73,
suite="stage-b-test-small-1-gpu-amd",
disabled="see https://github.com/sgl-project/sglang/issues/11127",
)
register_cuda_ci(est_time=73, suite="stage-b-test-small-1-gpu")
MODEL_TO_CONFIG = {
"Alibaba-NLP/gte-Qwen2-1.5B-instruct": (1, 1e-5),
"intfloat/e5-mistral-7b-instruct": (1, 1e-5),
test/registered/models/test_encoder_embedding_models.py
+11 -11
View File
@@ -1,8 +1,16 @@
import multiprocessing as mp
import random
import time
import unittest
import torch
from transformers import AutoConfig, AutoTokenizer
from sglang.test.ci.ci_register import register_cuda_ci
from sglang.test.runners import DEFAULT_PROMPTS, HFRunner, SRTRunner
from sglang.test.test_utils import CustomTestCase, get_similarities, is_in_ci
# Encoder embedding model tests (CUDA only)
register_cuda_ci(est_time=270, suite="stage-b-test-small-1-gpu")
register_cuda_ci(est_time=144, suite="stage-b-test-small-1-gpu-5090")
# Copyright 2023-2024 SGLang Team
# Licensed under the Apache License, Version 2.0 (the "License");
@@ -20,16 +28,8 @@ register_cuda_ci(est_time=144, suite="stage-b-test-small-1-gpu-5090")
# python -m unittest test_encoder_embedding_models.TestEncoderEmbeddingModels.test_prefill_logits
import multiprocessing as mp
import random
import time
import unittest
import torch
from transformers import AutoConfig, AutoTokenizer
from sglang.test.runners import DEFAULT_PROMPTS, HFRunner, SRTRunner
from sglang.test.test_utils import CustomTestCase, get_similarities, is_in_ci
register_cuda_ci(est_time=270, suite="stage-b-test-small-1-gpu")
MODELS = [("BAAI/bge-small-en", 1, 1e-5), ("BAAI/bge-m3", 1, 1e-5)]
test/registered/models/test_generation_models.py
+1 -1
View File
@@ -1,7 +1,7 @@
from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
# Generation model tests (CUDA only)
register_cuda_ci(est_time=103, suite="stage-b-test-small-1-gpu")
register_cuda_ci(est_time=103, suite="stage-b-test-large-1-gpu")
register_amd_ci(est_time=106, suite="stage-b-test-small-1-gpu-amd")
# Copyright 2023-2024 SGLang Team
test/registered/models/test_nvidia_nemotron_nano_v2_vl.py
+7 -6
View File
@@ -1,16 +1,17 @@
from sglang.test.ci.ci_register import register_cuda_ci
# NVIDIA Nemotron Nano V2 VL model tests (CUDA only)
# GSM8k + MMMU evaluation
register_cuda_ci(est_time=214, suite="stage-b-test-small-1-gpu")
import unittest
from sglang.test.ci.ci_register import register_cuda_ci
from sglang.test.kits.gsm8k_accuracy_kit import GSM8KMixin
from sglang.test.kits.mmmu_vlm_kit import MMMUMixin
from sglang.test.server_fixtures.default_fixture import DefaultServerBase
from sglang.test.server_fixtures.mmmu_fixture import MMMUServerBase
# NVIDIA Nemotron Nano V2 VL model tests (CUDA only)
# GSM8k + MMMU evaluation
register_cuda_ci(est_time=214, suite="stage-b-test-large-1-gpu")
MODEL = "nvidia/NVIDIA-Nemotron-Nano-12B-v2-VL-BF16"
test/registered/models/test_qwen_models.py
+4 -5
View File
@@ -1,14 +1,10 @@
from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
# Qwen model tests
register_cuda_ci(est_time=90, suite="stage-b-test-small-1-gpu")
register_cuda_ci(est_time=84, suite="stage-b-test-small-1-gpu-5090")
register_amd_ci(est_time=130, suite="stage-b-test-small-1-gpu-amd")
import unittest
from types import SimpleNamespace
from sglang.srt.utils import kill_process_tree
from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
from sglang.test.few_shot_gsm8k import run_eval
from sglang.test.test_utils import (
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -17,6 +13,9 @@ from sglang.test.test_utils import (
popen_launch_server,
)
register_cuda_ci(est_time=90, suite="stage-b-test-small-1-gpu")
register_amd_ci(est_time=130, suite="stage-b-test-small-1-gpu-amd")
class TestQwen2(CustomTestCase):
@classmethod
test/registered/models/test_reward_models.py
+9 -9
View File
@@ -1,9 +1,13 @@
import multiprocessing as mp
import unittest
import torch
from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
from sglang.test.runners import HFRunner, SRTRunner
from sglang.test.test_utils import CustomTestCase
# Reward model tests
register_cuda_ci(est_time=103, suite="stage-b-test-small-1-gpu")
register_cuda_ci(est_time=69, suite="stage-b-test-small-1-gpu-5090")
register_amd_ci(est_time=132, suite="stage-b-test-small-1-gpu-amd")
# Copyright 2023-2024 SGLang Team
# Licensed under the Apache License, Version 2.0 (the "License");
@@ -19,13 +23,9 @@ register_amd_ci(est_time=132, suite="stage-b-test-small-1-gpu-amd")
# limitations under the License.
# ==============================================================================
import multiprocessing as mp
import unittest
import torch
from sglang.test.runners import HFRunner, SRTRunner
from sglang.test.test_utils import CustomTestCase
register_cuda_ci(est_time=103, suite="stage-b-test-small-1-gpu")
register_amd_ci(est_time=132, suite="stage-b-test-small-1-gpu-amd")
MODELS = [
("LxzGordon/URM-LLaMa-3.1-8B", 1, 4e-2),
test/registered/models/test_transformers_models.py
+4 -5
View File
@@ -1,9 +1,4 @@
from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
# Transformers fallback model tests
register_cuda_ci(est_time=245, suite="stage-b-test-small-1-gpu")
register_cuda_ci(est_time=210, suite="stage-b-test-small-1-gpu-5090")
register_amd_ci(est_time=320, suite="stage-b-test-small-1-gpu-amd")
import dataclasses
import multiprocessing as mp
@@ -14,6 +9,7 @@ from typing import List
import torch
from sglang.srt.utils import is_hip, kill_process_tree
from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
from sglang.test.runners import DEFAULT_PROMPTS, SRTRunner, check_close_model_outputs
from sglang.test.test_utils import (
DEFAULT_MODEL_NAME_FOR_TEST,
@@ -24,6 +20,9 @@ from sglang.test.test_utils import (
popen_launch_server,
)
register_cuda_ci(est_time=245, suite="stage-b-test-small-1-gpu")
register_amd_ci(est_time=320, suite="stage-b-test-small-1-gpu-amd")
class TestTransformersFallbackEndpoint(CustomTestCase):
@classmethod
test/registered/models/test_vlm_models.py
+7 -6
View File
@@ -1,9 +1,3 @@
from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
# VLM (Vision Language Model) tests
register_cuda_ci(est_time=228, suite="stage-b-test-small-1-gpu")
register_amd_ci(est_time=420, suite="stage-b-test-small-1-gpu-amd")
import argparse
import random
import sys
@@ -11,12 +5,19 @@ import unittest
from types import SimpleNamespace
from sglang.srt.utils import is_hip
from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
from sglang.test.kits.mmmu_vlm_kit import (
DEFAULT_MEM_FRACTION_STATIC,
MMMUMultiModelTestBase,
)
from sglang.test.test_utils import is_in_ci
# VLM (Vision Language Model) tests
register_cuda_ci(est_time=228, suite="stage-b-test-large-1-gpu")
register_amd_ci(est_time=420, suite="stage-b-test-small-1-gpu-amd")
_is_hip = is_hip()
# VLM models for testing
if _is_hip: