SDK returns empty text blocks, but does not accept them

Using anthropic sdk via langchain_anthropic integration

from langchain_core.messages import AIMessage, HumanMessage, ToolMessage, SystemMessage
from langchain_anthropic import ChatAnthropic
from langchain_core.tools import tool

@tool
def type_letter(letter: str) -> str:
    """Type the given letter."""
    return "OK"


model = ChatAnthropic(model="claude-3-haiku-20240307", temperature=0).bind_tools([type_letter])

messages = [
    SystemMessage(content="Repeat the given string using the provided tools. Do not write anything else or provide any explanations. For example, if the string is 'abc', you must print the letters 'a', 'b', and 'c' one at a time and in that order. "),
    HumanMessage(content="dog")
]

model.invoke(messages)

The SDK can return empty text blocks from models:

ToolsBetaMessage(id='msg_01D6BbZbwbPYRVMPVu5SZRTp', content=[TextBlock(text='', type='text'), ToolUseBlock(id='toolu_01YHFB1pLY3xNa1edzoV4HtL', input={'letter': 'd'}, name='type_letter', type='tool_use'), ToolUseBlock(id='toolu_011ewRdDypyE8yHMniYT2gzj', input={'letter': 'o'}, name='type_letter', type='tool_use'), ToolUseBlock(id='toolu_01YKV1QaDPePiAFQVwmrSmjk', input={'letter': 'g'}, name='type_letter', type='tool_use')], model='claude-3-haiku-20240307', role='assistant', stop_reason='tool_use', stop_sequence=None, type='message', usage=Usage(input_tokens=395, output_tokens=125))

However, the SDK does not accept empty text block if re-interacting with the model (e.g., if using an agent).

Here's an example of re-interaction with the model:

from langchain_core.messages import AIMessage, HumanMessage, ToolMessage, SystemMessage
from langchain_anthropic import ChatAnthropic
from langchain_core.tools import tool

@tool
def type_letter(letter: str) -> str:
    """Type the given letter."""
    return "OK"


model = ChatAnthropic(model="claude-3-opus-20240229", temperature=0).bind_tools([type_letter])


messages = [
    SystemMessage(content="Repeat the given string using the provided tools. Do not write anything else or provide any explanations. For example, if the string is 'abc', you must print the letters 'a', 'b', and 'c' one at a time and in that order. "),
    HumanMessage(content="dog"),
    AIMessage(
        content=[
            {"text": "", "type": "text"},
            {
                "id": "toolu_01V6d6W32QGGSmQm4BT98EKk",
                "input": {
                    "letter": "d"
                },
                "name": "type_letter",
                "type": "tool_use"
            },
        ],
        tool_calls=[
            {
                "name": "type_letter",
                "args": {
                    "letter": "d"
                },
                "id": "toolu_01V6d6W32QGGSmQm4BT98EKk"
            },
        ],
    ),
    ToolMessage(content="OK", tool_call_id="toolu_01V6d6W32QGGSmQm4BT98EKk"),
]

model.invoke(messages)

This results in a 4xx error:

BadRequestError: Error code: 400 - {'type': 'error', 'error': {'type': 'invalid_request_error', 'message': 'messages: text content blocks must be non-empty'}}

But this should be accepted since it's output by the sdk / model itself

Related code snippet:

import anthropic

client = anthropic.Client()

client.beta.tools.messages.create(
    model="claude-3-haiku-20240307",
    max_tokens=1000,
    tools=[
        {
            "name": "tavily_search_results_json",
            "description": "A search engine optimized for comprehensive, accurate, and trusted results. Useful for when you need to answer questions about current events. Input should be a search query.",
            "input_schema": {
                "type": "object",
                "properties": {
                    "query": {
                        "description": "search query to look up",
                        "type": "string",
                    }
                },
                "required": ["query"],
            },
        }
    ],
    messages=[
        {
            "role": "user",
            "content": [{"type": "text", "text": "What's langgraph designed to do?"}],
        },
        {
            "role": "assistant",
            "content": [
                {"text": "", "type": "text"},
                {
                    "id": "toolu_01QxzCHR55x5RigzDdwLuQ5V",
                    "input": {"query": "foo"},
                    "name": "tavily_search_results_json",
                    "type": "tool_use",
                },
            ],
        },
        {
            "role": "user",
            "content": [
                {
                    "type": "tool_result",
                    "tool_use_id": "toolu_01QxzCHR55x5RigzDdwLuQ5V",
                    "content": "[Foo bar baz]",
                },
            ],
        },
    ],
)

Thanks for reporting. Working on fixing this asap!

This is now fixed.

Unless I am missing something, I still have the same issue with the following code:

import anthropic

client = anthropic.Client()

messages = [
    {
        "content": [{"text": "How many appointments do I have today?", "type": "text"}],
        "role": "user",
    },
    {
        "content": [
            {
                "text": '<thinking>\nTo find out how many appointments the user has today, I should use the get_day_calendar_events tool. This tool requires a "day" parameter in the format YYYY-MM-DD.\n\nEmma did not specify a date in her request, so I first need to determine what the current date is. I can use the get_current_day tool to retrieve that information.\n\nOnce I have the current date, I can pass that into the get_day_calendar_events tool to retrieve Emma\'s appointments for today.\n</thinking>',
                "type": "text",
            },
            {
                "id": "toolu_01UsBRMJWT8xAssuztCGEZNu",
                "input": {},
                "name": "get_current_day",
                "type": "tool_use",
            },
        ],
        "role": "assistant",
    },
    {
        "content": [
            {
                "tool_use_id": "toolu_01UsBRMJWT8xAssuztCGEZNu",
                "type": "tool_result",
                "content": [{"text": "2024-05-15", "type": "text"}],
                "is_error": False,
            }
        ],
        "role": "user",
    },
    {
        "content": [
            {"text": "", "type": "text"},
            {
                "id": "toolu_01YRKLXccBZjeQfYU2M7YLY8",
                "input": {"day": "2024-05-15"},
                "name": "get_day_calendar_events",
                "type": "tool_use",
            },
        ],
        "role": "assistant",
    },
    {
        "content": [
            {
                "tool_use_id": "toolu_01YRKLXccBZjeQfYU2M7YLY8",
                "type": "tool_result",
                "content": [{"text": "<TOOL OUTPUT HERE>", "type": "text"}],
                "is_error": False,
            }
        ],
        "role": "user",
    },
]


client.messages.create(
    model="claude-3-opus-20240229",
    max_tokens=1000,
    messages=messages,
)

This gives me the following error:

anthropic.BadRequestError: Error code: 400 - {'type': 'error', 'error': {'type': 'invalid_request_error', 'message': 'messages: text content blocks must be non-empty'}}

@dedeswim

Hi @dedeswim, was this part of your request given as output by the API or did you put it together yourself?

    {
        "content": [
            {"text": "", "type": "text"},
            {
                "id": "toolu_01YRKLXccBZjeQfYU2M7YLY8",
                "input": {"day": "2024-05-15"},
                "name": "get_day_calendar_events",
                "type": "tool_use",
            },
        ],
        "role": "assistant",
    },

The API returning 400 for {"text": "", "type": "text"} is correct, but the API returning this as content is not.

@aaron-lerner

Hi @aaron-lerner, you are right, it's my fault! Sorry, I somewhat completely missed that line.

Thanks a lot for the quick reply!

BadRequestError: Error code: 400 - {'type': 'error', 'error': {'type': 'invalid_request_error', 'message': 'messages: text content blocks must be non-empty'}}

I'm getting this issue from AnthropicVertex.

fix(llm/anthropic): add workaround to Anthropic API glitch

POC: native LLM support (#9628)

POC: native LLM support (dagger#9628)

POC: native LLM support (#9628)

Navigation Menu

Repository navigation

SDK returns empty text blocks, but does not accept them #461

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

SDK returns empty text blocks, but does not accept them #461

Description

Activity

eyurtsev commented on Apr 17, 2024

hinthornw commented on Apr 17, 2024

aaron-lerner commented on Apr 17, 2024

aaron-lerner commented on Apr 17, 2024

dedeswim commented on May 31, 2024

aaron-lerner commented on May 31, 2024

dedeswim commented on May 31, 2024

mmichelli commented on Feb 12, 2025

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions