Find answers from the community

Updated 9 months ago

Im following the notebook only,

PPujari12

Im following the notebook only, currently using this -
base64str = base64.b64encode(response.content).decode("utf-8")

base64str2 = base64.b64encode(requests.get(image_url).content)

image_document = ImageDocument(image=base64str, image_mimetype="image/jpeg")

10 comments

DDeleted User

so when you pass the base64 string it complains about image url not being supported?

PPujari12

yeah

DDeleted User

what does the beginning of the base64 string look like

PPujari12

'iVBORw0KGgoAAAANSUhEUgAABEgAAAU4CAYAAAC8LuKjAAAMPmlDQ1BJQ0MgUHJv

DDeleted User

are you using llamaindex's SimpleDirectoryLoader to get the image document(s)?

azure is likely complaining that you're not sending the right format of image data. the content of image_document

DDeleted User

try image_documents and send the array returned from llamaindex's directory loader (it should handle whatever type their API is looking for)

DDeleted User

i think in python it would be like:

Plain Text

image_documents = SimpleDirectoryReader("./your_img_directory/").load_data()

response = azure_openai_mm_llm.complete(
  prompt="Describe the images as an alternative text",
  image_documents=image_documents
)

DDeleted User

here's where llamaindex formats the image data in their reader, so it actually should be the same as your base64 https://github.com/run-llama/llama_index/blob/main/llama-index-legacy/llama_index/legacy/readers/file/image_reader.py#L71

DDeleted User

ahhh if it uses image_path in that case it is actually a path (URL) not the image data

PPujari12

I deployed a new LLM and its working 😛

Add a reply