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

[V0 Deprecation] Remove V0 sampling metadata (#25345)

Browse Source 
Signed-off-by: Woosuk Kwon <woosuk@thinkingmachines.ai>
This commit is contained in:
Woosuk Kwon
2025-09-21 10:37:11 -07:00
committed by GitHub
parent c438b2951c
commit 1c3ffdbecc
141 changed files with 172 additions and 583 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
tests/plugins/vllm_add_dummy_model/vllm_add_dummy_model/my_llava.py
View File

@@ -9,7 +9,6 @@ from vllm.model_executor.models.llava import (LlavaDummyInputsBuilder,
LlavaForConditionalGeneration,
LlavaMultiModalProcessor,
LlavaProcessingInfo)
from vllm.model_executor.sampling_metadata import SamplingMetadata
from vllm.multimodal import MULTIMODAL_REGISTRY
@@ -18,11 +17,10 @@ from vllm.multimodal import MULTIMODAL_REGISTRY
dummy_inputs=LlavaDummyInputsBuilder)
class MyLlava(LlavaForConditionalGeneration):
def compute_logits(
self, hidden_states: torch.Tensor,
sampling_metadata: SamplingMetadata) -> Optional[torch.Tensor]:
def compute_logits(self,
hidden_states: torch.Tensor) -> Optional[torch.Tensor]:
# this dummy model always predicts the first token
logits = super().compute_logits(hidden_states, sampling_metadata)
logits = super().compute_logits(hidden_states)
if logits is not None:
logits.zero_()
logits[:, 0] += 1.0

8
tests/plugins/vllm_add_dummy_model/vllm_add_dummy_model/my_opt.py
View File

@@ -6,16 +6,14 @@ from typing import Optional
import torch
from vllm.model_executor.models.opt import OPTForCausalLM
from vllm.model_executor.sampling_metadata import SamplingMetadata
class MyOPTForCausalLM(OPTForCausalLM):
def compute_logits(
self, hidden_states: torch.Tensor,
sampling_metadata: SamplingMetadata) -> Optional[torch.Tensor]:
def compute_logits(self,
hidden_states: torch.Tensor) -> Optional[torch.Tensor]:
# this dummy model always predicts the first token
logits = super().compute_logits(hidden_states, sampling_metadata)
logits = super().compute_logits(hidden_states)
if logits is not None:
logits.zero_()
logits[:, 0] += 1.0