168ad600b5
### What this PR does / why we need it?
For offline scenarios, adjust the scheduling process to prioritize the
prefill phase of all requests, then process the decode phase of all
requests.
### How was this patch tested?
```
max_num_seqs=24,
additional_config={
"ascend_scheduler_config":{
"enabled": True,
"enable_pd_transfer": True,
"decode_max_num_seqs": 24,
"enable_chunked_prefill": False
}
},
```
| input | output | num prompts | max_num_seqs | dp | tp | scheduler |
tps |
| ------ | ------ | ---------- | ---------------- | ---- | ---- |
---------------- | --------------- |
| dapo-math-17K | 2K | 384 | 24 | 2 | 1 | v1 | 234.06 |
| dapo-math-17K | 2K | 384 | 24 | 2 | 1 | pd transfer | 239.59(+2.4%) |
| dapo-math-17K| 2K | 384 | 24 | 4 | 1 | v1 | 222.85 |
| dapo-math-17K| 2K | 384 | 24 | 4 | 1 | pd transfer | 225.81(+1.3%) |
- vLLM version: v0.10.1.1
- vLLM main:
6fb2788163
---------
Signed-off-by: CaranLic <740821011@qq.com>
41 lines
1.8 KiB
Python
41 lines
1.8 KiB
Python
import torch
|
|
from pytest_mock import MockerFixture
|
|
from vllm.config import SchedulerConfig, VllmConfig
|
|
|
|
from tests.ut.base import PytestBase
|
|
from vllm_ascend.sample.logits_processor import AscendMinPLogitsProcessor
|
|
|
|
|
|
class TestMinPLogitsProcessorInitFunc(PytestBase):
|
|
|
|
def test_init_func_with_decode_max_num_seqs(self, mocker: MockerFixture):
|
|
device_cpu = torch.device("cpu")
|
|
device_npu = torch.device("npu")
|
|
is_pin_memory = False
|
|
mock_vllm_config = mocker.MagicMock(spec=VllmConfig)
|
|
mock_scheduler_config = mocker.MagicMock(spec=SchedulerConfig)
|
|
mock_scheduler_config.decode_max_num_seqs = 0
|
|
mock_scheduler_config.max_num_seqs = 128
|
|
mock_vllm_config.scheduler_config = mock_scheduler_config
|
|
# torch.zeros/torch.empty returns error on online ut machine, so mock it
|
|
mock_tensor = torch.zeros((256, ),
|
|
dtype=torch.float32,
|
|
pin_memory=False)
|
|
mocker.patch("torch.zeros", return_value=mock_tensor)
|
|
mock_empty_tensor = torch.empty((256, ), dtype=torch.float32)
|
|
mocker.patch("torch.empty", return_value=mock_empty_tensor)
|
|
|
|
processor_cpu = AscendMinPLogitsProcessor(mock_vllm_config, device_cpu,
|
|
is_pin_memory)
|
|
|
|
assert processor_cpu.min_p is not None
|
|
assert processor_cpu.use_double_tensor is False
|
|
assert processor_cpu.min_p_cpu.shape[0] == 256
|
|
|
|
processor_cpu = AscendMinPLogitsProcessor(mock_vllm_config, device_npu,
|
|
is_pin_memory)
|
|
|
|
assert processor_cpu.min_p is not None
|
|
assert processor_cpu.use_double_tensor is True
|
|
assert processor_cpu.min_p_cpu.shape[0] == 256
|