Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Feat: reimplement vllm backend beam search using logprobs #293

Draft
wants to merge 1 commit into
base: main
Choose a base branch
from
Draft

Feat: reimplement vllm backend beam search using logprobs #293

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
38 changes: 18 additions & 20 deletions optimum_benchmark/backends/vllm/backend.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -117,35 +117,33 @@ def batch_offline_engine_generate(self, inputs: Dict[str, Any], kwargs: Dict[str
self.pretrained_model.add_request(
inputs=prompt,
request_id=str(i),
params=SamplingParams(
ignore_eos=True,
detokenize=True,
seed=self.config.seed,
n=kwargs.get("num_return_sequences"),
max_tokens=kwargs.get("max_new_tokens"),
min_tokens=kwargs.get("min_new_tokens"),
use_beam_search=kwargs.get("num_beams") > 1,
logits_processors=kwargs.get("logits_processors", None),
),
params=self.get_sampling_params(kwargs),
)

while self.pretrained_model.has_unfinished_requests():
self.pretrained_model.step()

def get_sampling_params(self, kwargs: Dict[str, Any]) -> SamplingParams:
params = SamplingParams(
ignore_eos=True,
detokenize=True,
seed=self.config.seed,
n=kwargs.get("num_return_sequences"),
max_tokens=kwargs.get("max_new_tokens"),
min_tokens=kwargs.get("min_new_tokens"),
logits_processors=kwargs.get("logits_processors", None),
)
# following huggingface transformers implementation
# https://github.com/huggingface/transformers/blob/v4.45.2/src/transformers/generation/beam_search.py#L534
if kwargs.get("num_beams") > 1:
params.logprobs = 2 * kwargs.get("num_beams")
return params

async def single_online_engine_generate(self, prompt: str, request_id: str, kwargs: Dict[str, Any]) -> Any:
stream = await self.pretrained_model.add_request(
inputs=prompt,
request_id=request_id,
params=SamplingParams(
ignore_eos=True,
detokenize=True,
seed=self.config.seed,
n=kwargs.get("num_return_sequences"),
max_tokens=kwargs.get("max_new_tokens"),
min_tokens=kwargs.get("min_new_tokens"),
use_beam_search=kwargs.get("num_beams") > 1,
logits_processors=kwargs.get("logits_processors", None),
),
params=self.get_sampling_params(),
)

async for _ in stream:
Expand Down