xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-05-17 07:09:08 +08:00
Code Issues Packages Projects Releases Wiki Activity

[V1][Minor] Minor cleanup for GPU Model Runner (#13983)

Browse Source 
Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
This commit is contained in:
Woosuk Kwon 2025-02-27 13:11:40 -08:00 committed by GitHub
parent 38acae6e97
commit cd813c6d4d
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 7 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
vllm/v1/worker/gpu_model_runner.py
View File

@ -1187,8 +1187,9 @@ class GPUModelRunner(LoRAModelRunnerMixin):
# NOTE: Currently model is profiled with a single non-text # NOTE: Currently model is profiled with a single non-text
# modality with the max possible input tokens even when # modality with the max possible input tokens even when
# it supports multiple. # it supports multiple.
max_tokens_by_modality_dict = MULTIMODAL_REGISTRY.get_max_tokens_per_item_by_nonzero_modality( # noqa: E501 max_tokens_by_modality_dict = (
self.model_config) MULTIMODAL_REGISTRY.
get_max_tokens_per_item_by_nonzero_modality(self.model_config))
dummy_data_modality, max_tokens_per_mm_item = max( dummy_data_modality, max_tokens_per_mm_item = max(
max_tokens_by_modality_dict.items(), key=lambda item: item[1]) max_tokens_by_modality_dict.items(), key=lambda item: item[1])
@ -1275,15 +1276,15 @@ class GPUModelRunner(LoRAModelRunnerMixin):
# maximum num_tokens. # maximum num_tokens.
num_reqs = self.scheduler_config.max_num_seqs num_reqs = self.scheduler_config.max_num_seqs
num_tokens = self.max_num_tokens num_tokens = self.max_num_tokens
min_tokens_per_req: int = num_tokens // num_reqs min_tokens_per_req = num_tokens // num_reqs
num_scheduled_tokens_list: List[int] = [min_tokens_per_req] * num_reqs num_scheduled_tokens_list = [min_tokens_per_req] * num_reqs
num_scheduled_tokens_list[-1] += num_tokens % num_reqs num_scheduled_tokens_list[-1] += num_tokens % num_reqs
assert sum(num_scheduled_tokens_list) == num_tokens assert sum(num_scheduled_tokens_list) == num_tokens
assert len(num_scheduled_tokens_list) == num_reqs assert len(num_scheduled_tokens_list) == num_reqs
num_scheduled_tokens: np.ndarray = np.array(num_scheduled_tokens_list, num_scheduled_tokens = np.array(num_scheduled_tokens_list,
dtype=np.int32) dtype=np.int32)
logit_indices = np.cumsum(num_scheduled_tokens) - 1 logit_indices = np.cumsum(num_scheduled_tokens) - 1
with self.maybe_profile_with_lora(self.lora_config, with self.maybe_profile_with_lora(self.lora_config,