EngineX-Ascend/enginex-ascend-910-vllm
10
0
Fork 1
Code Issues Pull Requests Actions Projects Releases Wiki Activity

init v0.11.0rc0

Browse Source
This commit is contained in:
luopingyi
2025-10-14 10:38:28 +08:00
parent 67afd0ea78
commit 66dc16f966
278 changed files with 28130 additions and 11708 deletions
Download Patch File Download Diff File Expand all files Collapse all files

42
tests/e2e/multicard/test_prefix_caching.py
View File

@@ -116,20 +116,22 @@ def test_prefix_cache_with_ascend_scheduler(model: str,
prefix_cache_output = vllm_model.generate_greedy(
INPUT_PROMPTS, max_tokens)
with VllmRunner(model,
additional_config={
'ascend_scheduler_config': {
'enabled': True,
'enable_prefix_caching': True,
"enable_chunked_prefill": True,
},
},
enforce_eager=True,
max_model_len=2048,
tensor_parallel_size=2,
gpu_memory_utilization=0.7) as vllm_model:
chunk_prefill_prefix_cache_output = vllm_model.generate_greedy(
INPUT_PROMPTS, max_tokens)
# TODO: enable apc and chunked prefill with ascend scheduler will lead accuracy problem.
# Disable it now. Fix it or drop the ascend scheduler in the future.
# with VllmRunner(model,
# additional_config={
# 'ascend_scheduler_config': {
# 'enabled': True,
# 'enable_prefix_caching': True,
# "enable_chunked_prefill": True,
# },
# },
# enforce_eager=True,
# max_model_len=2048,
# tensor_parallel_size=2,
# gpu_memory_utilization=0.7) as vllm_model:
# chunk_prefill_prefix_cache_output = vllm_model.generate_greedy(
# INPUT_PROMPTS, max_tokens)
check_outputs_equal(
outputs_0_lst=vllm_output,
@@ -138,9 +140,9 @@ def test_prefix_cache_with_ascend_scheduler(model: str,
name_1="prefix_cache_output",
)
check_outputs_equal(
outputs_0_lst=chunk_prefill_prefix_cache_output,
outputs_1_lst=prefix_cache_output,
name_0="chunk_prefill_prefix_cache_output",
name_1="prefix_cache_output",
)
# check_outputs_equal(
# outputs_0_lst=chunk_prefill_prefix_cache_output,
# outputs_1_lst=prefix_cache_output,
# name_0="chunk_prefill_prefix_cache_output",
# name_1="prefix_cache_output",
# )