Skip to content

GitLab

Commit 129fce94 authored by zhuwenwen's avatar zhuwenwen
Browse files

update DCU info and skip tests same to nv

parent 3f78216a
Show whitespace changes
Inline Side-by-side
Showing with 44 additions and 251 deletions
tests/kernels/test_cache.py
View file @ 129fce94
......@@ -6,8 +6,7 @@ import torch
from tests.kernels.utils import DEFAULT_OPCHECK_TEST_UTILS
from vllm import _custom_ops as ops
from vllm.utils import seed_everything
from vllm.utils import is_hip
from vllm.utils import seed_everything, is_hip
from .utils import torch_version
COPYING_DIRECTION = [('cuda', 'cpu'), ('cuda', 'cuda'), ('cpu', 'cuda')]
......
tests/models/decoder_only/vision_language/test_llava_next_video.py
View file @ 129fce94
......@@ -12,6 +12,7 @@ from vllm.sequence import SampleLogprobs
from ....conftest import VIDEO_ASSETS, HfRunner, VllmRunner, _VideoAssets
from ...utils import check_logprobs_close
from ....utils import models_path_prefix
from vllm.utils import is_hip
_PREFACE = (
"A chat between a curious human and an artificial intelligence assistant. "
......@@ -160,12 +161,11 @@ def run_test(
)
@pytest.mark.skipif(transformers.__version__ < "4.45",
reason="Waiting for next transformers release")
@pytest.mark.parametrize("model", models)
@pytest.mark.parametrize(
"size_factors",
[
def get_size_factors():
if is_hip():
return [[],]
else:
return [
# No video
[],
# Single-scale
......@@ -175,6 +175,14 @@ def run_test(
# Multi-scale
[0.25, 0.5, 1.0],
],
@pytest.mark.skipif(transformers.__version__ < "4.45",
reason="Waiting for next transformers release")
@pytest.mark.parametrize("model", models)
@pytest.mark.parametrize(
"size_factors",
get_size_factors()
)
@pytest.mark.parametrize("dtype", ["half"])
@pytest.mark.parametrize("max_tokens", [128])
......@@ -205,6 +213,8 @@ def test_models(hf_runner, vllm_runner, video_assets, model, size_factors,
)
@pytest.mark.skipif(is_hip(),
reason="Consistent with NV.")
@pytest.mark.skipif(transformers.__version__ < "4.45",
reason="Waiting for next transformers release")
@pytest.mark.parametrize("model", models)
......
tests/models/decoder_only/vision_language/test_paligemma.py
View file @ 129fce94
......@@ -125,10 +125,11 @@ def run_test(
)
@pytest.mark.parametrize("model", models)
@pytest.mark.parametrize(
"size_factors",
[
def get_size_factors():
if is_hip():
return [[],]
else:
return [
# No image
[],
# Single-scale
......@@ -137,7 +138,12 @@ def run_test(
[1.0, 1.0, 1.0],
# Multi-scale
[0.25, 0.5, 1.0],
],
]
@pytest.mark.parametrize("model", models)
@pytest.mark.parametrize(
"size_factors",
get_size_factors()
)
@pytest.mark.parametrize("dtype", [
pytest.param(
......
tests/samplers/test_rejection_sampler.py
View file @ 129fce94
......@@ -201,6 +201,8 @@ def test_deterministic_when_seeded(k: int, vocab_size: int, batch_size: int,
assert torch.equal(results[j][i], results[0][i])
@pytest.mark.skipif(is_hip(),
reason="Consistent with NV.")
@pytest.mark.parametrize("k", [1, 3, 6])
@pytest.mark.parametrize("vocab_size", [30_000, 50_000])
@pytest.mark.parametrize("batch_size", [1, 8, 32, 128])
......
tests/tokenization/test_get_eos.py
View file @ 129fce94
......@@ -7,8 +7,12 @@ from vllm.transformers_utils.config import try_get_generation_config
from vllm.transformers_utils.tokenizer import get_tokenizer
from ..utils import models_path_prefix
import os
import pytest
from vllm.utils import is_hip
@pytest.mark.skipif(is_hip(),
reason="Consistent with NV.")
def test_get_llama3_eos_token():
model_name = os.path.join(models_path_prefix, "meta-llama/Meta-Llama-3-8B-Instruct")
......
vllm/model_executor/layers/fused_moe/configs/E=64,N=1408,device_name=DCU_K100_AI.json deleted 100644 → 0
View file @ 3f78216a
{
"1": {
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 32,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 8,
"num_warps": 8,
"num_stages": 2
},
"2": {
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 32,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 6,
"num_warps": 8,
"num_stages": 2
},
"4": {
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 32,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1,
"num_warps": 8,
"num_stages": 2
},
"8": {
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 32,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1,
"num_warps": 8,
"num_stages": 2
},
"16": {
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 32,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1,
"num_warps": 8,
"num_stages": 2
},
"32": {
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 32,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1,
"num_warps": 8,
"num_stages": 2
},
"1024": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1,
"num_warps": 4,
"num_stages": 2
},
"2048": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 4,
"num_warps": 4,
"num_stages": 2
},
"4096": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 6,
"num_warps": 4,
"num_stages": 1
},
"6144": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 6,
"num_warps": 8,
"num_stages": 1
},
"8192": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 8,
"num_warps": 4,
"num_stages": 1
},
"12288": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 8,
"num_warps": 8,
"num_stages": 1
},
"16384": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 4,
"num_warps": 4,
"num_stages": 1
},
"32786": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 6,
"num_warps": 4,
"num_stages": 1
}
}
vllm/model_executor/layers/fused_moe/configs/E=64,N=704,device_name=DCU_K100_AI.json deleted 100644 → 0
View file @ 3f78216a
{
"1": {
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 8,
"num_warps": 4,
"num_stages": 2
},
"2": {
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 32,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 8,
"num_warps": 8,
"num_stages": 2
},
"4": {
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 32,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 4,
"num_warps": 8,
"num_stages": 2
},
"8": {
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 32,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 4,
"num_warps": 8,
"num_stages": 2
},
"16": {
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 32,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1,
"num_warps": 8,
"num_stages": 2
},
"32": {
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 32,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1,
"num_warps": 8,
"num_stages": 2
},
"1024": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1,
"num_warps": 4,
"num_stages": 2
},
"2048": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1,
"num_warps": 4,
"num_stages": 2
},
"4096": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1,
"num_warps": 4,
"num_stages": 1
},
"6144": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1,
"num_warps": 4,
"num_stages": 1
},
"8192": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 4,
"num_warps": 4,
"num_stages": 1
},
"12288": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 6,
"num_warps": 4,
"num_stages": 1
},
"16384": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1,
"num_warps": 4,
"num_stages": 1
},
"32786": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1,
"num_warps": 4,
"num_stages": 1
}
}
vllm/utils.py
View file @ 129fce94
......@@ -1418,5 +1418,5 @@ class W8a8GetCacheJSON:
def get_w8a8json_name(self,n,k):
device_name = current_platform.get_device_name().replace(" ", "_")
return self.triton_json_dir+f"/W8A8_{n}_{k}_DCU{device_name}.json"
return self.triton_json_dir+f"/W8A8_{n}_{k}_HCU{device_name}.json"
\ No newline at end of file
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment