Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[VLM] Improve consistency between feature size calculation and dummy data for profiling #6146

Merged
merged 1 commit into from
Jul 5, 2024
Merged

[VLM] Improve consistency between feature size calculation and dummy data for profiling #6146

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
21 changes: 8 additions & 13 deletions vllm/model_executor/models/llava_next.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -37,6 +37,9 @@
"language_model.model": "language_model",
}

# Result in the max possible feature size (2x2 grid of 336x336px tiles)
MAX_IMAGE_FEATURE_SIZE_HEIGHT = MAX_IMAGE_FEATURE_SIZE_WIDTH = 448


class LlavaNextImagePixelInputs(TypedDict):
type: Literal["pixel_values"]
Expand Down Expand Up @@ -128,27 +131,19 @@ def get_llava_next_image_feature_size(


def get_max_llava_next_image_tokens(ctx: InputContext):
# Result in the max possible feature size (2x2 grid of 336x336px tiles)
dummy_height = dummy_width = 448

return get_llava_next_image_feature_size(
ctx.get_hf_config(LlavaNextConfig),
input_height=dummy_height,
input_width=dummy_width,
input_height=MAX_IMAGE_FEATURE_SIZE_HEIGHT,
input_width=MAX_IMAGE_FEATURE_SIZE_WIDTH,
)


def dummy_data_for_llava_next(ctx: InputContext, seq_len: int):
hf_config = ctx.get_hf_config(LlavaNextConfig)
vision_config = hf_config.vision_config

# Result in the max possible feature size (2x2 grid of 336x336px tiles)
dummy_height = dummy_width = 448
image_feature_size = get_llava_next_image_feature_size(
hf_config,
input_height=dummy_height,
input_width=dummy_width,
)
image_feature_size = get_max_llava_next_image_tokens(ctx)

if isinstance(vision_config, CLIPVisionConfig):
seq_data = dummy_seq_data_for_clip(
Expand All @@ -160,8 +155,8 @@ def dummy_data_for_llava_next(ctx: InputContext, seq_len: int):

mm_data = dummy_image_for_clip(
vision_config,
image_width_override=dummy_width,
image_height_override=dummy_height,
image_width_override=MAX_IMAGE_FEATURE_SIZE_WIDTH,
image_height_override=MAX_IMAGE_FEATURE_SIZE_HEIGHT,
)

return seq_data, mm_data
Expand Down
23 changes: 10 additions & 13 deletions vllm/model_executor/models/phi3v.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -53,6 +53,10 @@
# Cannot find the following 2 numbers from hf config.
_IMAGE_TOKEN_ID = 32044

# Result in the max possible feature size (h:w = 16:1)
MAX_IMAGE_FEATURE_SIZE_HEIGHT = 8000
MAX_IMAGE_FEATURE_SIZE_WIDTH = 50

CLIP_VIT_LARGE_PATCH14_336_CONFIG = CLIPVisionConfig(dropout=0.0,
hidden_act="quick_gelu",
hidden_size=1024,
Expand Down Expand Up @@ -322,24 +326,17 @@ def get_phi3v_image_feature_size(


def get_max_phi3v_image_tokens(ctx: InputContext):
# Result in the max possible feature size (h:w = 16:1)
dummy_height, dummy_width = 8000, 50

return get_phi3v_image_feature_size(
ctx.get_hf_config(PretrainedConfig),
input_height=dummy_height,
input_width=dummy_width,
input_height=MAX_IMAGE_FEATURE_SIZE_HEIGHT,
input_width=MAX_IMAGE_FEATURE_SIZE_WIDTH,
)


def dummy_data_for_phi3v(ctx: InputContext, seq_len: int):
# Result in the max possible feature size (h:w = 16:1)
dummy_height, dummy_width = 8000, 50
image_feature_size = get_phi3v_image_feature_size(
ctx.get_hf_config(PretrainedConfig),
input_height=dummy_height,
input_width=dummy_width,
)

image_feature_size = get_max_phi3v_image_tokens(ctx)

seq_data = dummy_seq_data_for_clip(
CLIP_VIT_LARGE_PATCH14_336_CONFIG,
Expand All @@ -349,8 +346,8 @@ def dummy_data_for_phi3v(ctx: InputContext, seq_len: int):
)
mm_data = dummy_image_for_clip(
CLIP_VIT_LARGE_PATCH14_336_CONFIG,
image_width_override=dummy_width,
image_height_override=dummy_height,
image_width_override=MAX_IMAGE_FEATURE_SIZE_WIDTH,
image_height_override=MAX_IMAGE_FEATURE_SIZE_HEIGHT,
)

return seq_data, mm_data
Expand Down
Loading