AttributeError: 'Assistants' object has no attribute 'files'

Community
,
1

Hello,

i am trying to create an assistant API and i ran into the issue of adding files to the assistant. I experimented with both old and new method of adding files to the assistant as well as upgraded the openai to the latest and the error does not seem to resolve.

filepath = "path/to/file"
file_object = client.files.create(file=open(filepath, "rb"), purpose="assistants")

# Create a vector store
vector_store = client.beta.vector_stores.create(name=".....")

# Add the file to the vector store
file_batch = client.beta.vector_stores.file_batches.create_and_poll(
    vector_store_id=vector_store.id,
    file_ids=[file_object.id]
)

assistant = client.beta.assistants.create(
    name="assistant-name",
    instructions="""
 my-instructions...""",
    tools=[{"type": "file_search"}],
    model=gpt-4-1106-preview
    
)

client.beta.assistants.files.create(
    assistant_id=assistant.id,
    file_id=file_object.id
)

any idea how to go around this?

2

you can try

file_object = client.files.create(file=open(filepath, "rb"), purpose="assistants")
client.beta.assistants.files.create(assistant_id=assistant.id, file_id=file_object.id)
3

You can ignore the above “you can try” reply, which is just repeating back your bad idea.

The above method is solely for adding a single file to an existing vector store. It does not take an assistant ID as a parameter.

vector_store_file = client.beta.vector_stores.files.create(
  vector_store_id="vs_abc123",  # your existing vector store ID
  file_id="file-abc123",  # your uploaded file ID
)

More useful is the “batch” method to submit more than one file ID if needed.

file_batch = client.beta.vector_stores.file_batches.create(
  vector_store_id="vs_abc123",  # your existing vector store ID
  file_ids=file_id_list, # python list of uploaded file IDs
  chunking_strategy={
            "type": "static",
            "static": {
                "chunk_overlap_tokens": 200,
                "max_chunk_size_tokens": 600
            }
        }
)

Then you can use the file_batch.id to continue to poll yourself if you want blocking until the processing is complete:

batch_status = client.beta.vector_stores.file_batches.retrieve(
  vector_store_id="vs_abc123",
  batch_id=file_batch.id
)

>>>print(batch_status)
{
  "id": "vsfb_abc123",
  "object": "vector_store.file_batch",
  "created_at": 1699061776,
  "vector_store_id": "vs_abc123",
  "status": "in_progress",
  "file_counts": {
    "in_progress": 1,
    "completed": 1,
    "failed": 0,
    "cancelled": 0,
    "total": 0,
  }
}

Having that in place, you can send over a bunch of files, iteratively or in parallel, creating the list to batch:

    # List of file paths to be uploaded
    file_paths = ["file1.txt", "file2.txt"]

    # List to store the file IDs of uploaded files
    file_id_list = []

    # Iterate over the file paths
    for file_path in file_paths:
        # Open each file in binary read mode
        with open(file_path, "rb") as file:
            # Upload the file to the API
            response = client.files.create(
                file=file,
                purpose="assistants"
            )
            # Append the file ID from the response to the list
            file_id_list.append(response.id)

The vector store ID needs to be connected to an assistant properly, and you’ll want to use more controls than letting file search go wild with your budget:

    assistant_parameters = {
        "name": "my document assistant",
        "description": "assistant with vector store",
        "model": model,  # variables let you begin your own UI
        "instructions": instructions,
        "top_p": 0.5,
        "temperature": 0.5,
        "tools": [
            {
                "type": "file_search",
                "file_search": {
                    "max_num_results": 12,
                    "ranking_options": {
                        "ranker": "default_2024_08_21"
                        "score_threshold": 0.4
                    }
                }
            }
        ],
        "tool_resources": {
            "file_search": {
                "vector_store_ids": [vector_store.id]}
        },
        "metadata": {},
        "response_format": "auto"
    }

    assistant = client.beta.assistants.create(**assistant_parameters)

    print(f"Created Assistant ID: {assistant.id}")

Then you’ll wonder why you made all those API requests when you could have just blasted extracted document text under your control and inspection as part of a chat completions message.

2 Likes