[Hardware][CPU] using current_platform.is_cpu (#9536)
This commit is contained in:
@@ -5,7 +5,7 @@ Run `pytest tests/models/test_phimoe.py`.
|
||||
import pytest
|
||||
import torch
|
||||
|
||||
from vllm.utils import is_cpu
|
||||
from vllm.platforms import current_platform
|
||||
|
||||
from ....utils import large_gpu_test
|
||||
from ...utils import check_logprobs_close
|
||||
@@ -70,7 +70,7 @@ def test_phimoe_routing_function():
|
||||
assert torch.equal(topk_ids, ground_truth[test_id]["topk_ids"])
|
||||
|
||||
|
||||
@pytest.mark.skipif(condition=is_cpu(),
|
||||
@pytest.mark.skipif(condition=current_platform.is_cpu(),
|
||||
reason="This test takes a lot time to run on CPU, "
|
||||
"and vllm CI's disk space is not enough for this model.")
|
||||
@large_gpu_test(min_gb=80)
|
||||
|
||||
@@ -3,8 +3,8 @@ from typing import List, Optional, Tuple, Type
|
||||
import pytest
|
||||
|
||||
from vllm.multimodal.utils import rescale_image_size
|
||||
from vllm.platforms import current_platform
|
||||
from vllm.sequence import SampleLogprobs
|
||||
from vllm.utils import is_cpu
|
||||
|
||||
from ....conftest import IMAGE_ASSETS, HfRunner, VllmRunner, _ImageAssets
|
||||
from ...utils import check_logprobs_close
|
||||
@@ -46,7 +46,7 @@ def run_test(
|
||||
|
||||
All the image fixtures for the test are from IMAGE_ASSETS.
|
||||
For huggingface runner, we provide the PIL images as input.
|
||||
For vllm runner, we provide MultiModalDataDict objects
|
||||
For vllm runner, we provide MultiModalDataDict objects
|
||||
and corresponding MultiModalConfig as input.
|
||||
Note, the text input is also adjusted to abide by vllm contract.
|
||||
The text output is sanitized to be able to compare with hf.
|
||||
@@ -103,7 +103,7 @@ def run_test(
|
||||
|
||||
|
||||
target_dtype = "half"
|
||||
if is_cpu():
|
||||
if current_platform.is_cpu():
|
||||
target_dtype = "bfloat16"
|
||||
|
||||
|
||||
|
||||
@@ -7,7 +7,7 @@ from PIL.Image import Image
|
||||
from transformers import AutoConfig
|
||||
|
||||
from vllm.multimodal.utils import rescale_image_size
|
||||
from vllm.utils import is_cpu
|
||||
from vllm.platforms import current_platform
|
||||
|
||||
from ....conftest import (IMAGE_ASSETS, HfRunner, PromptImageInput, VllmRunner,
|
||||
_ImageAssets)
|
||||
@@ -78,7 +78,7 @@ def run_test(
|
||||
|
||||
All the image fixtures for the test are from IMAGE_ASSETS.
|
||||
For huggingface runner, we provide the PIL images as input.
|
||||
For vllm runner, we provide MultiModalDataDict objects
|
||||
For vllm runner, we provide MultiModalDataDict objects
|
||||
and corresponding MultiModalConfig as input.
|
||||
Note, the text input is also adjusted to abide by vllm contract.
|
||||
The text output is sanitized to be able to compare with hf.
|
||||
@@ -244,7 +244,7 @@ def run_awq_test(
|
||||
|
||||
|
||||
target_dtype = "half"
|
||||
if is_cpu():
|
||||
if current_platform.is_cpu():
|
||||
target_dtype = "bfloat16"
|
||||
|
||||
|
||||
|
||||
@@ -10,8 +10,9 @@ from vllm.inputs import InputContext, token_inputs
|
||||
from vllm.model_executor.models.phi3v import _IMAGE_TOKEN_ID
|
||||
from vllm.multimodal import MultiModalRegistry
|
||||
from vllm.multimodal.utils import rescale_image_size
|
||||
from vllm.platforms import current_platform
|
||||
from vllm.sequence import SampleLogprobs
|
||||
from vllm.utils import is_cpu, is_hip
|
||||
from vllm.utils import is_hip
|
||||
|
||||
from ....conftest import (IMAGE_ASSETS, HfRunner, PromptImageInput, VllmRunner,
|
||||
_ImageAssets)
|
||||
@@ -49,7 +50,7 @@ def vllm_to_hf_output(vllm_output: Tuple[List[int], str,
|
||||
|
||||
|
||||
target_dtype = "half"
|
||||
if is_cpu():
|
||||
if current_platform.is_cpu():
|
||||
target_dtype = "bfloat16"
|
||||
|
||||
# ROCm Triton FA can run into shared memory issues with these models,
|
||||
|
||||
@@ -5,8 +5,8 @@ import torch
|
||||
|
||||
from vllm.config import ModelConfig, TaskOption
|
||||
from vllm.inputs import InputContext
|
||||
from vllm.platforms import current_platform
|
||||
from vllm.sequence import Logprob, PromptLogprobs, SampleLogprobs
|
||||
from vllm.utils import is_cpu
|
||||
|
||||
TokensText = Tuple[List[int], str]
|
||||
|
||||
@@ -19,7 +19,7 @@ def check_outputs_equal(
|
||||
name_1: str,
|
||||
):
|
||||
"""
|
||||
Compare the two sequences generated by different models,
|
||||
Compare the two sequences generated by different models,
|
||||
which should be equal.
|
||||
"""
|
||||
assert len(outputs_0_lst) == len(outputs_1_lst)
|
||||
@@ -255,7 +255,7 @@ def build_model_context(model_name: str,
|
||||
mm_processor_kwargs: Optional[Dict] = None,
|
||||
limit_mm_per_prompt: Optional[Dict] = None):
|
||||
"""Creates an InputContext for a given model.
|
||||
|
||||
|
||||
Args:
|
||||
model_name: Name of the model being considered.
|
||||
tokenizer_name: Name of the tokenizer being considered.
|
||||
@@ -270,7 +270,7 @@ def build_model_context(model_name: str,
|
||||
if tokenizer_name is None:
|
||||
tokenizer_name = model_name
|
||||
if dtype is None:
|
||||
dtype = "bfloat16" if is_cpu() else "half"
|
||||
dtype = "bfloat16" if current_platform.is_cpu() else "half"
|
||||
|
||||
model_config = ModelConfig(
|
||||
model_name,
|
||||
|
||||
Reference in New Issue
Block a user