[Core] Pipeline Parallel Support (#4412)

Signed-off-by: Muralidhar Andoorveedu <muralidhar.andoorveedu@centml.ai>
This commit is contained in:
Murali Andoorveedu
2024-07-02 10:58:08 -07:00
committed by GitHub
parent 15aba081f3
commit c5832d2ae9
82 changed files with 1096 additions and 400 deletions
+3 -1
View File
@@ -18,7 +18,7 @@ from vllm.model_executor.models.clip import CLIPVisionModel
from vllm.model_executor.models.llama import LlamaModel
from vllm.model_executor.sampling_metadata import SamplingMetadata
from vllm.multimodal import MULTIMODAL_REGISTRY
from vllm.sequence import SamplerOutput
from vllm.sequence import IntermediateTensors, SamplerOutput
from .clip import dummy_image_for_clip, dummy_seq_data_for_clip
from .interfaces import SupportsVision
@@ -202,6 +202,7 @@ class LlavaForConditionalGeneration(nn.Module, SupportsVision):
positions: torch.Tensor,
kv_caches: List[torch.Tensor],
attn_metadata: AttentionMetadata,
intermediate_tensors: Optional[IntermediateTensors] = None,
**kwargs: object,
) -> SamplerOutput:
"""Run forward pass for LLaVA-1.5.
@@ -247,6 +248,7 @@ class LlavaForConditionalGeneration(nn.Module, SupportsVision):
positions,
kv_caches,
attn_metadata,
None,
inputs_embeds=inputs_embeds)
return hidden_states