biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Chore] Move tokenizer initialization methods (#29793)

Browse Source 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2025-12-02 13:33:37 +08:00
committed by GitHub
parent e2fbfc955e
commit 653591d5e7
51 changed files with 150 additions and 129 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
vllm/model_executor/models/granite_speech.py
View File

@@ -59,8 +59,8 @@ from vllm.multimodal.processing import (
)
from vllm.multimodal.profiling import BaseDummyInputsBuilder
from vllm.sequence import IntermediateTensors
from vllm.transformers_utils.processor import cached_get_processor
from vllm.transformers_utils.tokenizer import cached_get_tokenizer
from vllm.tokenizers import cached_tokenizer_from_config
from vllm.transformers_utils.processor import cached_processor_from_config
from vllm.utils.tensor_schema import TensorSchema, TensorShape
from .blip2 import Blip2QFormerModel
@@ -862,7 +862,7 @@ class GraniteSpeechForConditionalGeneration(
else:
raise ValueError(f"Unsupported task type {task_type}")
tokenizer = cached_get_tokenizer(model_config.model)
tokenizer = cached_tokenizer_from_config(model_config)
chat = [dict(role="user", content=user_prompt)]
prompt = tokenizer.apply_chat_template(
chat,
@@ -886,7 +886,7 @@ class GraniteSpeechForConditionalGeneration(
model_config: ModelConfig,
) -> int | None:
"""Get the number of audio tokens for an audio duration in sec."""
processor = cached_get_processor(model_config.model)
processor = cached_processor_from_config(model_config)
hop_length = processor.audio_processor.melspec_kwargs["hop_length"]
proj_win_size = processor.audio_processor.projector_window_size
ds_rate = processor.audio_processor.projector_downsample_rate