EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

[CI] Fix FusedMoEConfig and input batch failure to recover CI (#1602)

Browse Source 
Make CI happy

1.
c1909e7e8c
changed moeConfig init way
2.
48fb076cbc
changed input batch logic.

This PR address these change to vllm-ascend.

Closes: https://github.com/vllm-project/vllm-ascend/issues/1600

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
This commit is contained in:
wangxiyuan
2025-07-03 18:36:17 +08:00
committed by GitHub
parent d96da1f00c
commit a45dfde283
11 changed files with 173 additions and 134 deletions
Download Patch File Download Diff File Expand all files Collapse all files

60
tests/e2e/singlecard/sample/test_rejection_sampler.py
View File

@@ -9,6 +9,7 @@ from vllm.v1.spec_decode.metadata import SpecDecodeMetadata
from vllm_ascend.sample.rejection_sampler import (PLACEHOLDER_TOKEN_ID,
AscendRejectionSampler)
from vllm_ascend.utils import vllm_version_is
DEVICE = "npu"
@@ -49,27 +50,46 @@ def create_sampling_metadata(
temperature = None
else:
assert temperature is not None
if vllm_version_is("0.9.1"):
return SamplingMetadata(
temperature=temperature,
all_greedy=all_greedy,
all_random=not all_greedy,
top_p=top_p,
top_k=top_k,
min_p=torch.empty(1, ),
generators=generators,
max_num_logprobs=0,
no_penalties=False,
prompt_token_ids=None,
frequency_penalties=torch.tensor([]),
presence_penalties=torch.tensor([]),
repetition_penalties=torch.tensor([]),
output_token_ids=[],
min_tokens={},
logit_bias=[None],
allowed_token_ids_mask=None,
bad_words_token_ids={},
)
else:
from vllm.v1.sample.logits_processor import LogitsProcessorManager
return SamplingMetadata(
temperature=temperature,
all_greedy=all_greedy,
all_random=not all_greedy,
top_p=top_p,
top_k=top_k,
min_p=torch.empty(1, ),
generators=generators,
max_num_logprobs=0,
no_penalties=False,
prompt_token_ids=None,
frequency_penalties=torch.tensor([]),
presence_penalties=torch.tensor([]),
repetition_penalties=torch.tensor([]),
output_token_ids=[],
min_tokens={},
logit_bias=[None],
allowed_token_ids_mask=None,
bad_words_token_ids={},
)
return SamplingMetadata(temperature=temperature,
all_greedy=all_greedy,
all_random=not all_greedy,
top_p=top_p,
top_k=top_k,
generators=generators,
max_num_logprobs=0,
no_penalties=False,
prompt_token_ids=None,
frequency_penalties=torch.tensor([]),
presence_penalties=torch.tensor([]),
repetition_penalties=torch.tensor([]),
output_token_ids=[],
allowed_token_ids_mask=None,
bad_words_token_ids={},
logitsprocs=LogitsProcessorManager())
########################### Tests for Greedy Sampling ###################