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

[Bugfix] Fix dtype mismatch in PaliGemma (#6367)

Browse Source
This commit is contained in:
Cyrus Leung 2024-07-12 23:22:18 +08:00 committed by GitHub
parent aea19f0989
commit 024ad87cdc
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
3 changed files with 12 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
tests/models/test_paligemma.py
View File

@ -129,7 +129,7 @@ def run_test(
[0.25, 0.5, 1.0], [0.25, 0.5, 1.0],
], ],
) )
@pytest.mark.parametrize("dtype", ["float"]) @pytest.mark.parametrize("dtype", ["float", "half"])
@pytest.mark.parametrize("max_tokens", [128]) @pytest.mark.parametrize("max_tokens", [128])
@pytest.mark.parametrize("num_logprobs", [5]) @pytest.mark.parametrize("num_logprobs", [5])
def test_models(hf_runner, vllm_runner, image_assets, model, size_factors, def test_models(hf_runner, vllm_runner, image_assets, model, size_factors,

1
vllm/model_executor/models/gemma.py
View File

@ -277,6 +277,7 @@ class GemmaModel(nn.Module):
positions: torch.Tensor, positions: torch.Tensor,
kv_caches: List[torch.Tensor], kv_caches: List[torch.Tensor],
attn_metadata: AttentionMetadata, attn_metadata: AttentionMetadata,
intermediate_tensors: Optional[IntermediateTensors] = None,
inputs_embeds: Optional[torch.Tensor] = None, inputs_embeds: Optional[torch.Tensor] = None,
) -> torch.Tensor: ) -> torch.Tensor:
if inputs_embeds is not None: if inputs_embeds is not None:

14
vllm/model_executor/models/paligemma.py
View File

@ -19,7 +19,7 @@ from vllm.model_executor.models.gemma import GemmaModel
from vllm.model_executor.sampling_metadata import SamplingMetadata from vllm.model_executor.sampling_metadata import SamplingMetadata
from vllm.multimodal import MULTIMODAL_REGISTRY from vllm.multimodal import MULTIMODAL_REGISTRY
from vllm.multimodal.image import cached_get_tokenizer from vllm.multimodal.image import cached_get_tokenizer
from vllm.sequence import SamplerOutput, SequenceData from vllm.sequence import IntermediateTensors, SamplerOutput, SequenceData
from .interfaces import SupportsVision from .interfaces import SupportsVision
from .utils import merge_vision_embeddings from .utils import merge_vision_embeddings
@ -111,7 +111,7 @@ def input_processor_for_paligemma(ctx: InputContext, llm_inputs: LLMInputs):
orig_prompt = llm_inputs.get("prompt") orig_prompt = llm_inputs.get("prompt")
orig_prompt_ids = llm_inputs.get("prompt_token_ids") orig_prompt_ids = llm_inputs.get("prompt_token_ids")
if image_token_str in orig_prompt: if orig_prompt is not None and image_token_str in orig_prompt:
logger.warning( logger.warning(
"The image token '%s' was detected in the prompt and " "The image token '%s' was detected in the prompt and "
"will be removed. Please follow the proper prompt format" "will be removed. Please follow the proper prompt format"
@ -214,7 +214,9 @@ class PaliGemmaForConditionalGeneration(nn.Module, SupportsVision):
def _image_pixels_to_features(self, vision_tower: SiglipVisionModel, def _image_pixels_to_features(self, vision_tower: SiglipVisionModel,
pixel_values: torch.Tensor) -> torch.Tensor: pixel_values: torch.Tensor) -> torch.Tensor:
image_outputs = vision_tower(pixel_values, output_hidden_states=True) target_dtype = vision_tower.get_input_embeddings().weight.dtype
image_outputs = vision_tower(pixel_values.to(dtype=target_dtype),
output_hidden_states=True)
selected_image_features = image_outputs.last_hidden_state selected_image_features = image_outputs.last_hidden_state
@ -236,9 +238,12 @@ class PaliGemmaForConditionalGeneration(nn.Module, SupportsVision):
return self.multi_modal_projector(image_features) return self.multi_modal_projector(image_features)
def forward(self, input_ids: torch.Tensor, positions: torch.Tensor, def forward(self,
input_ids: torch.Tensor,
positions: torch.Tensor,
kv_caches: List[torch.Tensor], kv_caches: List[torch.Tensor],
attn_metadata: AttentionMetadata, attn_metadata: AttentionMetadata,
intermediate_tensors: Optional[IntermediateTensors] = None,
**kwargs: object) -> SamplerOutput: **kwargs: object) -> SamplerOutput:
parsed_image_input = self._parse_and_validate_image_input(**kwargs) parsed_image_input = self._parse_and_validate_image_input(**kwargs)
@ -263,6 +268,7 @@ class PaliGemmaForConditionalGeneration(nn.Module, SupportsVision):
positions, positions,
kv_caches, kv_caches,
attn_metadata, attn_metadata,
None,
inputs_embeds=inputs_embeds) inputs_embeds=inputs_embeds)
return hidden_states return hidden_states