[VLM] Remove image_input_type from VLM config (#5852)

Signed-off-by: Xiaowei Jiang <xwjiang2010@gmail.com>
Co-authored-by: Cyrus Leung <cyrus.tl.leung@gmail.com>
Co-authored-by: Roger Wang <ywang@roblox.com>
This commit is contained in:
xwjiang2010
2024-07-02 07:57:09 +00:00
committed by GitHub
co-authored by Cyrus Leung Roger Wang
parent 2c37540aa6
commit 98d6682cd1
35 changed files with 329 additions and 751 deletions
+7 -6
View File
@@ -8,7 +8,7 @@ from PIL import Image
from vllm.config import ModelConfig
from vllm.envs import VLLM_IMAGE_FETCH_TIMEOUT
from vllm.multimodal.image import ImagePixelData
from vllm.multimodal.base import MultiModalDataDict
class ImageFetchAiohttp:
@@ -53,14 +53,10 @@ class ImageFetchAiohttp:
"Invalid 'image_url': A valid 'image_url' must start "
"with either 'data:image' or 'http'.")
image.load()
return image
async def async_get_and_parse_image(image_url: str) -> ImagePixelData:
with await ImageFetchAiohttp.fetch_image(image_url) as image:
return ImagePixelData(image)
def encode_image_base64(image: Image.Image, format: str = 'JPEG') -> str:
"""Encode a pillow image to base64 format."""
@@ -91,3 +87,8 @@ def get_full_image_text_prompt(image_prompt: str, text_prompt: str,
raise ValueError(
f"Unsupported model type: {config.hf_config.model_type}")
return full_prompt
async def async_get_and_parse_image(image_url: str) -> MultiModalDataDict:
image = await ImageFetchAiohttp.fetch_image(image_url)
return {"image": image}