[BUG] OpenAI Assistants API - Parallel Tool Use Broken

[hayescode]

API Spec link

https://github.com/Azure/azure-rest-api-specs/tree/main/specification/cognitiveservices/data-plane/AzureOpenAI/inference/preview/2024-12-01-preview

API Spec version

2024-12-01-preview

Describe the bug

With parallel_tool_calls set to True sometimes the required_action returns fake function_names and arguments like "recipient_name". The only tool I have is called ask_database but instead of submitting 2 seperate tool calls it combines into 1.

    run = await llm.beta.threads.runs.create_and_poll(
        assistant_id=assistant_id,
        thread_id=thread.id,
        parallel_tool_calls=True,
    )

Expected behavior

{
  "data": [
    {
      "step_details": {
        "tool_calls": [
          {
            "id": "call_U83VxNlfG2lZAv3GvtTDxQ50",
            "function": {
              "arguments": "{\"sql_query\":\"SELECT SUM(BARRELS) "}}\",
              "name": "ask_database",
              "output": null
            },
            "type": "function"
          },
          {
            "id": "call_U83VxNlfG2lZAv3GvtTDxQ51",
            "function": {
              "arguments": "{\"sql_query\":\"SELECT SUM(NEW_BARRELS) "}}\",
              "name": "ask_database",
              "output": null
            },
            "type": "function"
          }
        ],
        "type": "tool_calls"
      },
}

Actual behavior

{
  "data": [
    {
      "id": "step_d1lZKBXSLZbPbEca3xKUYCrN",
      "assistant_id": "asst_Y8fWym6bJMhDOZRocybNQZUm",
      "cancelled_at": null,
      "completed_at": null,
      "created_at": 1737740639,
      "expired_at": null,
      "failed_at": null,
      "last_error": null,
      "metadata": null,
      "object": "thread.run.step",
      "run_id": "run_9ZeZ8syaZySOWMEzZX7W7hR8",
      "status": "in_progress",
      "step_details": {
        "tool_calls": [
          {
            "id": "call_U83VxNlfG2lZAv3GvtTDxQ50",
            "function": {
              "arguments": "{\"tool_uses\":[{\"recipient_name\":\"functions.ask_database\",\"parameters\":{\"sql_query\":\"SELECT SUM(BARRELS) "}},{\"recipient_name\":\"functions.ask_database\",\"parameters\":{\"sql_query\":\"SELECT SUM(BARRELS) "}}]}",
              "name": "multi_tool_use.parallel",
              "output": null
            },
            "type": "function"
          }
        ],
        "type": "tool_calls"
      },
      "thread_id": "thread_B8PoIjgTfDr381kahAAsA2CR",
      "type": "tool_calls",
      "usage": null,
      "expires_at": 1737741237
    }
  ],
  "object": "list",
  "first_id": "step_d1lZKBXSLZbPbEca3xKUYCrN",
  "last_id": "step_d1lZKBXSLZbPbEca3xKUYCrN",
  "has_more": false
}

Reproduction Steps

run = await llm.beta.threads.runs.create_and_poll(
    assistant_id=assistant_id,
    thread_id=thread.id,
    parallel_tool_calls=True,
)
steps= await llm.beta.threads.runs.steps.list(thread_id=run.thread_id, run_id=run.id)
print(steps.model_dump_json(indent=2))

Environment

No response