[CI] Remove orphaned test/srt/ascend and test/srt/configs (#24145)
This commit is contained in:
Lianmin Zheng
@@ -60,7 +60,7 @@ jobs:
|
||||
- "python/sglang/!(multimodal_gen)/**/!(*.md)"
|
||||
- "python/pyproject_npu.toml"
|
||||
- "scripts/ci/npu/npu_ci_install_dependency.sh"
|
||||
- "test/srt/ascend/**"
|
||||
- "test/registered/ascend/**"
|
||||
- ".github/workflows/pr-test-npu.yml"
|
||||
multimodal_gen:
|
||||
- "python/sglang/multimodal_gen/**/!(*.md|*.ipynb)"
|
||||
|
||||
@@ -1,104 +0,0 @@
|
||||
import unittest
|
||||
|
||||
import torch
|
||||
|
||||
from sglang.srt.configs.qwen3_vl import Qwen3VLConfig
|
||||
from sglang.srt.distributed.parallel_state import (
|
||||
init_distributed_environment,
|
||||
initialize_model_parallel,
|
||||
)
|
||||
from sglang.srt.layers.dp_attention import initialize_dp_attention
|
||||
from sglang.srt.layers.quantization.unquant import (
|
||||
LinearMethodBase,
|
||||
UnquantizedLinearMethod,
|
||||
)
|
||||
from sglang.srt.models.qwen3_vl import Qwen3VLMoeVisionModel
|
||||
from sglang.srt.server_args import ServerArgs, set_global_server_args_for_scheduler
|
||||
|
||||
|
||||
def unpack(tensor, dim_len, pack_len):
|
||||
dim_part = dim_len // pack_len
|
||||
ret_val = tensor.reshape(dim_part, dim_part, pack_len, pack_len, -1)
|
||||
ret_val = ret_val.permute(4, 0, 2, 1, 3).reshape(1, -1, dim_len, dim_len)
|
||||
return ret_val
|
||||
|
||||
|
||||
class TestEmbedInterpolate(unittest.TestCase):
|
||||
@classmethod
|
||||
def setUpClass(cls):
|
||||
cls.pDevice = torch.get_default_device()
|
||||
torch.set_default_device("npu")
|
||||
|
||||
@classmethod
|
||||
def tearDownClass(cls):
|
||||
torch.set_default_device(cls.pDevice)
|
||||
|
||||
def test_embed_interpolate(self):
|
||||
self.assertTrue(issubclass(UnquantizedLinearMethod, LinearMethodBase))
|
||||
t_dim = [16, 32]
|
||||
s_dim = [192, 574]
|
||||
sarg = ServerArgs(model_path="dummy", device="npu")
|
||||
mconf = Qwen3VLConfig(
|
||||
hidden_size=64,
|
||||
num_heads=1,
|
||||
num_position_embeddings=2304,
|
||||
patch_size=16,
|
||||
spatial_merge_size=2,
|
||||
temporal_patch_size=2,
|
||||
deepstack_visual_indexes=[5, 11, 17],
|
||||
in_channels=3,
|
||||
depth=24,
|
||||
intermediate_size=256,
|
||||
hidden_act="gelu_pytorch_tanh",
|
||||
out_hidden_size=2560,
|
||||
)
|
||||
set_global_server_args_for_scheduler(sarg)
|
||||
init_distributed_environment(
|
||||
backend="gloo",
|
||||
world_size=1,
|
||||
rank=0,
|
||||
local_rank=0,
|
||||
distributed_init_method="tcp://127.0.0.1:2646",
|
||||
)
|
||||
initialize_model_parallel()
|
||||
initialize_dp_attention(
|
||||
server_args=sarg,
|
||||
model_config=mconf,
|
||||
)
|
||||
model = Qwen3VLMoeVisionModel(
|
||||
mconf,
|
||||
quant_config=None,
|
||||
norm_eps=1e-6,
|
||||
prefix="visual",
|
||||
)
|
||||
grid_thw = torch.tensor(
|
||||
[(t, s, s) for t, s in zip(t_dim, s_dim)], dtype=torch.int32
|
||||
)
|
||||
embeddings = model.fast_pos_embed_interpolate(grid_thw)
|
||||
|
||||
embeddings_s0 = embeddings[: s_dim[0] * s_dim[0], :]
|
||||
embeddings_s1 = embeddings[s_dim[0] * s_dim[0] : 2 * s_dim[0] * s_dim[0], :]
|
||||
self.assertTrue(torch.allclose(embeddings_s0, embeddings_s1, atol=5e-5))
|
||||
|
||||
embeddings_l = embeddings[
|
||||
t_dim[0] * s_dim[0] * s_dim[0] : t_dim[0] * s_dim[0] * s_dim[0]
|
||||
+ s_dim[1] * s_dim[1],
|
||||
:,
|
||||
]
|
||||
embeddings_s0 = torch.nn.functional.interpolate(
|
||||
unpack(embeddings_s0, s_dim[0], 2),
|
||||
size=(48, 48),
|
||||
mode="area",
|
||||
)
|
||||
embeddings_r = torch.nn.functional.interpolate(
|
||||
unpack(embeddings_l, s_dim[1], 2),
|
||||
size=(48, 48),
|
||||
mode="area",
|
||||
)
|
||||
self.assertTrue(
|
||||
torch.allclose(embeddings_s0, embeddings_r, atol=5e-1, rtol=5e-1)
|
||||
)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main()
|
||||
@@ -1,28 +0,0 @@
|
||||
tasks:
|
||||
- name: sglang-8192-1024-concurrency1
|
||||
server_cmd: python3 -m sglang.launch_server --model deepseek-ai/DeepSeek-V3-0324 --tp 8 --trust-remote-code --disable-radix-cache --max-prefill-tokens 32768
|
||||
client_cmd: python3 -m sglang.bench_serving --dataset-name random --random-range-ratio 1 --random-input-len 8192 --random-output-len 1024 --max-concurrency 1 --num-prompts 5 --output-file deepseek_v3_results.jsonl
|
||||
|
||||
- name: sglang-8192-1024-concurrency2
|
||||
server_cmd: python3 -m sglang.launch_server --model deepseek-ai/DeepSeek-V3-0324 --tp 8 --trust-remote-code --disable-radix-cache --max-prefill-tokens 32768
|
||||
client_cmd: python3 -m sglang.bench_serving --dataset-name random --random-range-ratio 1 --random-input-len 8192 --random-output-len 1024 --max-concurrency 2 --num-prompts 10 --output-file deepseek_v3_results.jsonl
|
||||
|
||||
- name: sglang-8192-1024-concurrency4
|
||||
server_cmd: python3 -m sglang.launch_server --model deepseek-ai/DeepSeek-V3-0324 --tp 8 --trust-remote-code --disable-radix-cache --max-prefill-tokens 32768
|
||||
client_cmd: python3 -m sglang.bench_serving --dataset-name random --random-range-ratio 1 --random-input-len 8192 --random-output-len 1024 --max-concurrency 4 --num-prompts 20 --output-file deepseek_v3_results.jsonl
|
||||
|
||||
- name: sglang-8192-1024-concurrency8
|
||||
server_cmd: python3 -m sglang.launch_server --model deepseek-ai/DeepSeek-V3-0324 --tp 8 --trust-remote-code --disable-radix-cache --max-prefill-tokens 32768
|
||||
client_cmd: python3 -m sglang.bench_serving --dataset-name random --random-range-ratio 1 --random-input-len 8192 --random-output-len 1024 --max-concurrency 8 --num-prompts 32 --output-file deepseek_v3_results.jsonl
|
||||
|
||||
- name: sglang-8192-1024-concurrency16
|
||||
server_cmd: python3 -m sglang.launch_server --model deepseek-ai/DeepSeek-V3-0324 --tp 8 --trust-remote-code --disable-radix-cache --max-prefill-tokens 32768
|
||||
client_cmd: python3 -m sglang.bench_serving --dataset-name random --random-range-ratio 1 --random-input-len 8192 --random-output-len 1024 --max-concurrency 16 --num-prompts 48 --output-file deepseek_v3_results.jsonl
|
||||
|
||||
- name: sglang-8192-1024-concurrency24
|
||||
server_cmd: python3 -m sglang.launch_server --model deepseek-ai/DeepSeek-V3-0324 --tp 8 --trust-remote-code --disable-radix-cache --max-prefill-tokens 32768
|
||||
client_cmd: python3 -m sglang.bench_serving --dataset-name random --random-range-ratio 1 --random-input-len 8192 --random-output-len 1024 --max-concurrency 24 --num-prompts 72 --output-file deepseek_v3_results.jsonl
|
||||
|
||||
- name: sglang-8192-1024-concurrency32
|
||||
server_cmd: python3 -m sglang.launch_server --model deepseek-ai/DeepSeek-V3-0324 --tp 8 --trust-remote-code --disable-radix-cache --max-prefill-tokens 32768
|
||||
client_cmd: python3 -m sglang.bench_serving --dataset-name random --random-range-ratio 1 --random-input-len 8192 --random-output-len 1024 --max-concurrency 32 --num-prompts 96 --output-file deepseek_v3_results.jsonl
|
||||
@@ -1,28 +0,0 @@
|
||||
tasks:
|
||||
- name: sglang-32000-100-concurrency1
|
||||
server_cmd: python3 -m sglang.launch_server --model deepseek-ai/DeepSeek-V3-0324 --tp 8 --trust-remote-code --disable-radix-cache --max-prefill-tokens 32768
|
||||
client_cmd: python3 -m sglang.bench_serving --dataset-name random --random-range-ratio 1 --random-input-len 32000 --random-output-len 100 --max-concurrency 1 --num-prompts 5 --output-file deepseek_v3_long_context_results.jsonl
|
||||
|
||||
- name: sglang-32000-100-concurrency2
|
||||
server_cmd: python3 -m sglang.launch_server --model deepseek-ai/DeepSeek-V3-0324 --tp 8 --trust-remote-code --disable-radix-cache --max-prefill-tokens 32768
|
||||
client_cmd: python3 -m sglang.bench_serving --dataset-name random --random-range-ratio 1 --random-input-len 32000 --random-output-len 100 --max-concurrency 2 --num-prompts 10 --output-file deepseek_v3_long_context_results.jsonl
|
||||
|
||||
- name: sglang-32000-100-concurrency4
|
||||
server_cmd: python3 -m sglang.launch_server --model deepseek-ai/DeepSeek-V3-0324 --tp 8 --trust-remote-code --disable-radix-cache --max-prefill-tokens 32768
|
||||
client_cmd: python3 -m sglang.bench_serving --dataset-name random --random-range-ratio 1 --random-input-len 32000 --random-output-len 100 --max-concurrency 4 --num-prompts 20 --output-file deepseek_v3_long_context_results.jsonl
|
||||
|
||||
- name: sglang-32000-100-concurrency8
|
||||
server_cmd: python3 -m sglang.launch_server --model deepseek-ai/DeepSeek-V3-0324 --tp 8 --trust-remote-code --disable-radix-cache --max-prefill-tokens 32768
|
||||
client_cmd: python3 -m sglang.bench_serving --dataset-name random --random-range-ratio 1 --random-input-len 32000 --random-output-len 100 --max-concurrency 8 --num-prompts 32 --output-file deepseek_v3_long_context_results.jsonl
|
||||
|
||||
- name: sglang-32000-100-concurrency16
|
||||
server_cmd: python3 -m sglang.launch_server --model deepseek-ai/DeepSeek-V3-0324 --tp 8 --trust-remote-code --disable-radix-cache --max-prefill-tokens 32768
|
||||
client_cmd: python3 -m sglang.bench_serving --dataset-name random --random-range-ratio 1 --random-input-len 32000 --random-output-len 100 --max-concurrency 16 --num-prompts 48 --output-file deepseek_v3_long_context_results.jsonl
|
||||
|
||||
- name: sglang-32000-100-concurrency24
|
||||
server_cmd: python3 -m sglang.launch_server --model deepseek-ai/DeepSeek-V3-0324 --tp 8 --trust-remote-code --disable-radix-cache --max-prefill-tokens 32768
|
||||
client_cmd: python3 -m sglang.bench_serving --dataset-name random --random-range-ratio 1 --random-input-len 32000 --random-output-len 100 --max-concurrency 24 --num-prompts 72 --output-file deepseek_v3_long_context_results.jsonl
|
||||
|
||||
- name: sglang-32000-100-concurrency32
|
||||
server_cmd: python3 -m sglang.launch_server --model deepseek-ai/DeepSeek-V3-0324 --tp 8 --trust-remote-code --disable-radix-cache --max-prefill-tokens 32768
|
||||
client_cmd: python3 -m sglang.bench_serving --dataset-name random --random-range-ratio 1 --random-input-len 32000 --random-output-len 100 --max-concurrency 32 --num-prompts 96 --output-file deepseek_v3_long_context_results.jsonl
|
||||
@@ -1,28 +0,0 @@
|
||||
tasks:
|
||||
- name: sglang-8192-1024-concurrency1
|
||||
server_cmd: python3 -m sglang.launch_server --model nvidia/Llama-3.1-405B-Instruct-FP8 --tp 8
|
||||
client_cmd: python3 -m sglang.bench_serving --dataset-name random --random-range-ratio 1 --random-input-len 8192 --random-output-len 1024 --max-concurrency 1 --num-prompts 5 --output-file llama_405b_results.jsonl
|
||||
|
||||
- name: sglang-8192-1024-concurrency2
|
||||
server_cmd: python3 -m sglang.launch_server --model nvidia/Llama-3.1-405B-Instruct-FP8 --tp 8
|
||||
client_cmd: python3 -m sglang.bench_serving --dataset-name random --random-range-ratio 1 --random-input-len 8192 --random-output-len 1024 --max-concurrency 2 --num-prompts 10 --output-file llama_405b_results.jsonl
|
||||
|
||||
- name: sglang-8192-1024-concurrency4
|
||||
server_cmd: python3 -m sglang.launch_server --model nvidia/Llama-3.1-405B-Instruct-FP8 --tp 8
|
||||
client_cmd: python3 -m sglang.bench_serving --dataset-name random --random-range-ratio 1 --random-input-len 8192 --random-output-len 1024 --max-concurrency 4 --num-prompts 20 --output-file llama_405b_results.jsonl
|
||||
|
||||
- name: sglang-8192-1024-concurrency8
|
||||
server_cmd: python3 -m sglang.launch_server --model nvidia/Llama-3.1-405B-Instruct-FP8 --tp 8
|
||||
client_cmd: python3 -m sglang.bench_serving --dataset-name random --random-range-ratio 1 --random-input-len 8192 --random-output-len 1024 --max-concurrency 8 --num-prompts 32 --output-file llama_405b_results.jsonl
|
||||
|
||||
- name: sglang-8192-1024-concurrency16
|
||||
server_cmd: python3 -m sglang.launch_server --model nvidia/Llama-3.1-405B-Instruct-FP8 --tp 8
|
||||
client_cmd: python3 -m sglang.bench_serving --dataset-name random --random-range-ratio 1 --random-input-len 8192 --random-output-len 1024 --max-concurrency 16 --num-prompts 48 --output-file llama_405b_results.jsonl
|
||||
|
||||
- name: sglang-8192-1024-concurrency24
|
||||
server_cmd: python3 -m sglang.launch_server --model nvidia/Llama-3.1-405B-Instruct-FP8 --tp 8
|
||||
client_cmd: python3 -m sglang.bench_serving --dataset-name random --random-range-ratio 1 --random-input-len 8192 --random-output-len 1024 --max-concurrency 24 --num-prompts 72 --output-file llama_405b_results.jsonl
|
||||
|
||||
- name: sglang-8192-1024-concurrency32
|
||||
server_cmd: python3 -m sglang.launch_server --model nvidia/Llama-3.1-405B-Instruct-FP8 --tp 8
|
||||
client_cmd: python3 -m sglang.bench_serving --dataset-name random --random-range-ratio 1 --random-input-len 8192 --random-output-len 1024 --max-concurrency 32 --num-prompts 96 --output-file llama_405b_results.jsonl
|
||||
@@ -1,25 +0,0 @@
|
||||
tasks:
|
||||
- name: sglang-128-4
|
||||
server_cmd: python3 -m sglang.launch_server --model-path meta-llama/Llama-3.1-8B-Instruct --disable-radix-cache
|
||||
client_cmd: python3 -m sglang.bench_serving --backend sglang --dataset-name random --random-input 128 --random-output 4 --request-rate 24 --num-prompt 1440
|
||||
- name: vllm-128-4
|
||||
server_cmd: python3 -m vllm.entrypoints.openai.api_server --model meta-llama/Llama-3.1-8B-Instruct --disable-log-requests
|
||||
client_cmd: python3 -m sglang.bench_serving --backend vllm --dataset-name random --random-input 128 --random-output 4 --request-rate 24 --num-prompt 1440
|
||||
- name: sglang-2000-100
|
||||
server_cmd: python3 -m sglang.launch_server --model-path meta-llama/Llama-3.1-8B-Instruct --disable-radix-cache
|
||||
client_cmd: python3 -m sglang.bench_serving --backend sglang --dataset-name random --random-input 2000 --random-output 100 --request-rate 2 --num-prompt 120
|
||||
- name: vllm-2000-100
|
||||
server_cmd: python3 -m vllm.entrypoints.openai.api_server --model meta-llama/Llama-3.1-8B-Instruct --disable-log-requests
|
||||
client_cmd: python3 -m sglang.bench_serving --backend vllm --dataset-name random --random-input 2000 --random-output 100 --request-rate 2 --num-prompt 120
|
||||
- name: sglang-4000-200
|
||||
server_cmd: python3 -m sglang.launch_server --model-path meta-llama/Llama-3.1-8B-Instruct --disable-radix-cache
|
||||
client_cmd: python3 -m sglang.bench_serving --backend sglang --dataset-name random --random-input 4000 --random-output 200 --request-rate 8 --num-prompt 480
|
||||
- name: vllm-4000-200
|
||||
server_cmd: python3 -m vllm.entrypoints.openai.api_server --model meta-llama/Llama-3.1-8B-Instruct --disable-log-requests
|
||||
client_cmd: python3 -m sglang.bench_serving --backend vllm --dataset-name random --random-input 4000 --random-output 200 --request-rate 8 --num-prompt 480
|
||||
- name: sglang-32000-100
|
||||
server_cmd: python3 -m sglang.launch_server --model-path meta-llama/Llama-3.1-8B-Instruct --disable-radix-cache
|
||||
client_cmd: python3 -m sglang.bench_serving --backend sglang --dataset-name random --random-input 32000 --random-output 100 --request-rate 1 --num-prompt 60
|
||||
- name: vllm-32000-100
|
||||
server_cmd: python3 -m vllm.entrypoints.openai.api_server --model meta-llama/Llama-3.1-8B-Instruct --disable-log-requests
|
||||
client_cmd: python3 -m sglang.bench_serving --backend vllm --dataset-name random --random-input 32000 --random-output 100 --request-rate 1 --num-prompt 60
|
||||
@@ -1,25 +0,0 @@
|
||||
tasks:
|
||||
- name: sglang-128-4
|
||||
server_cmd: python3 -m sglang.launch_server --model-path meta-llama/Llama-3.1-8B-Instruct --disable-radix-cache
|
||||
client_cmd: python3 -m sglang.bench_serving --backend sglang --dataset-name random --random-input 128 --random-output 4 --request-rate 24 --num-prompt 1440
|
||||
- name: sglang-triton-128-4
|
||||
server_cmd: python3 -m sglang.launch_server --model-path meta-llama/Llama-3.1-8B-Instruct --disable-radix-cache --attention-backend triton
|
||||
client_cmd: python3 -m sglang.bench_serving --backend sglang --dataset-name random --random-input 128 --random-output 4 --request-rate 24 --num-prompt 1440
|
||||
- name: sglang-2000-100
|
||||
server_cmd: python3 -m sglang.launch_server --model-path meta-llama/Llama-3.1-8B-Instruct --disable-radix-cache
|
||||
client_cmd: python3 -m sglang.bench_serving --backend sglang --dataset-name random --random-input 2000 --random-output 100 --request-rate 2 --num-prompt 120
|
||||
- name: sglang-triton-2000-100
|
||||
server_cmd: python3 -m sglang.launch_server --model-path meta-llama/Llama-3.1-8B-Instruct --disable-radix-cache --attention-backend triton
|
||||
client_cmd: python3 -m sglang.bench_serving --backend sglang --dataset-name random --random-input 2000 --random-output 100 --request-rate 2 --num-prompt 120
|
||||
- name: sglang-4000-200
|
||||
server_cmd: python3 -m sglang.launch_server --model-path meta-llama/Llama-3.1-8B-Instruct --disable-radix-cache
|
||||
client_cmd: python3 -m sglang.bench_serving --backend sglang --dataset-name random --random-input 4000 --random-output 200 --request-rate 8 --num-prompt 480
|
||||
- name: sglang-triton-4000-200
|
||||
server_cmd: python3 -m sglang.launch_server --model-path meta-llama/Llama-3.1-8B-Instruct --disable-radix-cache --attention-backend triton
|
||||
client_cmd: python3 -m sglang.bench_serving --backend sglang --dataset-name random --random-input 4000 --random-output 200 --request-rate 8 --num-prompt 480
|
||||
- name: sglang-32000-100
|
||||
server_cmd: python3 -m sglang.launch_server --model-path meta-llama/Llama-3.1-8B-Instruct --disable-radix-cache
|
||||
client_cmd: python3 -m sglang.bench_serving --backend sglang --dataset-name random --random-input 32000 --random-output 100 --request-rate 1 --num-prompt 60
|
||||
- name: sglang-triton-32000-100
|
||||
server_cmd: python3 -m sglang.launch_server --model-path meta-llama/Llama-3.1-8B-Instruct --disable-radix-cache --attention-backend triton
|
||||
client_cmd: python3 -m sglang.bench_serving --backend sglang --dataset-name random --random-input 32000 --random-output 100 --request-rate 1 --num-prompt 60
|
||||
@@ -1,7 +0,0 @@
|
||||
tasks:
|
||||
- name: sglang-benchmark
|
||||
server_cmd: python3 -m sglang.launch_server --model-path meta-llama/Llama-3.1-8B-Instruct --disable-radix-cache
|
||||
client_cmd: python3 -m sglang.bench_serving --backend sglang --dataset-name sharegpt --request-rate 16
|
||||
- name: vllm-benchmark
|
||||
server_cmd: python3 -m vllm.entrypoints.openai.api_server --model meta-llama/Llama-3.1-8B-Instruct --disable-log-requests
|
||||
client_cmd: python3 -m sglang.bench_serving --backend vllm --dataset-name sharegpt --request-rate 16
|
||||
@@ -13,9 +13,7 @@ from sglang.test.ci.ci_utils import TestFile, run_unittest_files
|
||||
suites = {
|
||||
# quantization_test suite migrated to test/registered/quant/
|
||||
# All CUDA tests migrated to test/registered/
|
||||
"__not_in_ci__": [
|
||||
TestFile("ascend/test_embed_interpolate_unittest.py"),
|
||||
],
|
||||
"__not_in_ci__": [],
|
||||
}
|
||||
|
||||
# Add AMD tests
|
||||
|
||||
Reference in New Issue
Block a user