BerriAI/litellm · error · ValueError

Invalid invoke provider: {invoke_provider}, for model: {mode

Error message

Invalid invoke provider: {invoke_provider}, for model: {model}

What it means

On the streaming passthrough path for 'invoke' endpoints, AmazonInvokeConfig.get_bedrock_invoke_provider(model) is used to derive the provider from the model string. If it returns None (model string doesn't start with a known Bedrock provider prefix), litellm raises ValueError('Invalid invoke provider: None, for model: ...') because no event-stream decoder can be selected.

Source

Thrown at litellm/llms/bedrock/passthrough/transformation.py:200

        3. Return the model_response
        """

        from litellm.litellm_core_utils.streaming_handler import (
            convert_generic_chunk_to_model_response_stream,
            generic_chunk_has_all_required_fields,
        )
        from litellm.llms.bedrock.chat import get_bedrock_event_stream_decoder
        from litellm.llms.bedrock.chat.invoke_transformations.base_invoke_transformation import (
            AmazonInvokeConfig,
        )
        from litellm.main import stream_chunk_builder
        from litellm.types.utils import GenericStreamingChunk, ModelResponseStream

        all_translated_chunks: Final = []
        if "invoke" in endpoint:
            invoke_provider: Final = AmazonInvokeConfig.get_bedrock_invoke_provider(model)
            if invoke_provider is None:
                raise ValueError(f"Invalid invoke provider: {invoke_provider}, for model: {model}")
            obj = get_bedrock_event_stream_decoder(
                invoke_provider=invoke_provider,
                model=model,
                sync_stream=True,
                json_mode=False,
            )
        elif "converse" in endpoint:
            obj = get_bedrock_event_stream_decoder(
                invoke_provider=None,
                model=model,
                sync_stream=True,
                json_mode=False,
            )
        else:
            return None

        for chunk in all_chunks:
            message = json.loads(chunk)

View on GitHub (pinned to 6c2dcb801b)

Solutions

Use the canonical model identifier with the correct lowercase provider prefix (e.g. 'anthropic.claude-3-haiku-20240307-v1:0').
For converse endpoints, use /bedrock/converse/<model> which resolves the provider differently (invoke_provider=None).
Check get_bedrock_invoke_provider's supported prefixes in your litellm version and match the model string to one.
Upgrade litellm if the provider is newly supported.

Example fix

# before
stream = litellm.completion(model='bedrock/invoke/MyClaude-Model', messages=msgs, stream=True)

# after
stream = litellm.completion(model='bedrock/invoke/anthropic.claude-3-haiku-20240307-v1:0', messages=msgs, stream=True)

Defensive patterns

Strategy: type-guard

Validate before calling

import re

def is_invoke_streamable_model(model: str) -> bool:
    # get_bedrock_invoke_provider matches known provider prefixes; mirror that check
    return re.match(r"^(amazon|anthropic|ai21|cohere|meta|mistral|stability|deepseek)\.", model) is not None

Type guard

def has_invoke_provider_prefix(model: str) -> bool:
    return "." in model and model.split(".")[0].islower() and model.split(".")[0] in {
        "amazon", "anthropic", "ai21", "cohere", "meta", "mistral", "stability"
    }

Try / catch

try:
    stream = litellm.completion(model=f"bedrock/invoke/{model}", messages=msgs, stream=True)
except ValueError as e:
    if "Invalid invoke provider" in str(e):
        # fall back to converse route which does not need provider derivation
        stream = litellm.completion(model=f"bedrock/converse/{model}", messages=msgs, stream=True)
    else:
        raise

Prevention

Use lowercase canonical provider-prefixed model IDs for invoke streaming.
Prefer converse over invoke for new integrations — it is provider-agnostic.
Unit-test model strings against the prefix list at config load time.

When it happens

Trigger: Streaming from /bedrock/invoke/<model> where <model> lacks a recognized provider prefix — e.g. a bare custom model name, a provisioned-throughput ARN fragment, or a typo'd prefix like 'Anthropic.Claude...' (capitalized).

Common situations: Using provisioned custom names (e.g. 'my-custom-model') without provider prefix; copy-pasting model ARNs instead of the model name; case-sensitive prefix mismatches.

Related errors

AI-assisted analysis of BerriAI/litellm@6c2dcb801b (2026-08-15). Data as JSON: /api/errors/dd6a327c9d60cbba. Report an issue: GitHub.