[Hardware][XPU] Register batch-invariant kernels for XPU (#41934)
Signed-off-by: tzielinski-habana <tomasz.zielinski@intel.com> Signed-off-by: Tomasz Zielinski <85164140+tzielinski-habana@users.noreply.github.com> Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> Co-authored-by: Chendi.Xue <chendi.xue@intel.com>
This commit is contained in:
co-authored by
gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
Chendi.Xue
parent
05eed72aec
commit
61141ed265
@@ -11,11 +11,13 @@ from utils import (
|
||||
TEST_MODEL,
|
||||
_extract_step_logprobs,
|
||||
_random_prompt,
|
||||
skip_if_not_cuda,
|
||||
skip_unsupported,
|
||||
)
|
||||
|
||||
import vllm.envs as envs
|
||||
from vllm import LLM, SamplingParams
|
||||
from vllm.platforms import current_platform
|
||||
|
||||
|
||||
@skip_unsupported
|
||||
@@ -49,6 +51,11 @@ def test_v1_generation_is_deterministic_across_batch_sizes_with_needle(
|
||||
seed.
|
||||
- Keep max_tokens and max_model_len bounded for speed and memory use.
|
||||
"""
|
||||
# Not all batch-invariant kernels are registered on XPU yet
|
||||
# (e.g. attention, custom ops), so e2e determinism is not guaranteed.
|
||||
if current_platform.is_xpu():
|
||||
pytest.xfail("Not all batch-invariant kernels registered on XPU yet")
|
||||
|
||||
seed = int(os.getenv("VLLM_TEST_SEED", "12345"))
|
||||
random.seed(seed)
|
||||
|
||||
@@ -157,6 +164,11 @@ def test_logprobs_bitwise_batch_invariance_bs1_vs_bsN(
|
||||
block_m,
|
||||
block_n,
|
||||
):
|
||||
# Not all batch-invariant kernels are registered on XPU yet
|
||||
# (e.g. attention, custom ops), so e2e determinism is not guaranteed.
|
||||
if current_platform.is_xpu():
|
||||
pytest.xfail("Not all batch-invariant kernels registered on XPU yet")
|
||||
|
||||
seed = int(os.getenv("VLLM_TEST_SEED", "12345"))
|
||||
random.seed(seed)
|
||||
tp_size = int(os.getenv("VLLM_TEST_TP_SIZE", "1"))
|
||||
@@ -641,7 +653,7 @@ def test_logprobs_without_batch_invariance_should_fail(
|
||||
pytest.fail(fail_msg)
|
||||
|
||||
|
||||
@skip_unsupported
|
||||
@skip_if_not_cuda
|
||||
@pytest.mark.parametrize("backend", ["FLASH_ATTN"])
|
||||
def test_decode_logprobs_match_prefill_logprobs(
|
||||
backend,
|
||||
|
||||
Reference in New Issue
Block a user