arrow-progressGLLM Inference v0.4 to v0.5

As you may have noticed, several legacy modules in GLLM Inference have been marked as deprecated for a while. If your application is still using them, you should have received warning logs.

Backward compatibility will be removed in the upcoming minor version v0.5.0. Please review this migration guide to ensure a smooth transition.

circle-info

Note: If you've set the GLLM Inference dependency in your app as ^0.4.0, you don't have to do this migration immediately, as you're locked to v0.4.x. You will only migrate to 0.5.0 when you choose to do so by updating your dependency to ^0.5.0.

However, its still recommended to do so ASAP to be able to access new features that will be added in the future.

Part 1: Prompt Builder

  1. The following modules are deprecated:

    1. AgnosticPromptBuilder

    2. HuggingFacePromptBuilder

    3. LlamaPromptBuilder

    4. MistralPromptBuilder

    5. OpenAIPromptBuilder

    6. MultimodalPromptBuilder

  2. All capability to build prompts to be consumable by LM invoker will be replaced by the new PromptBuilderarrow-up-right class. LMRP will only support this class. The prompt builder no longer has interface and subclasses.

  3. If you want to use the format_as_string capability previously provided by the deprecated subclasses, please use the new PromptFormatterarrow-up-right modules. However, you shouldnt need this to work with LM invoker and LMRP.

Part 2: LM Invoker

  1. All ...MultimodalLMInvoker are deprecated and are replaced with their ...LMInvoker counterpart.

    1. e.g., OpenAIMultimodalLMInvoker is replaced by OpenAILMInvoker.

  2. Google classes have been merged into a single class: GoogleLMInvoker.

  3. All LM invoker has been migrated to:

    1. Use native providers SDK instead of LangChain implementation.

    2. Support multimodality, tool calling, structured output, analytics, retry, timeout, and some provider-specific features.

  4. bind_tool_params param is deprecated and is replaced by tools param.

  5. with_structured_output_params param is deprecated and is replaced by response_schema param.

Part 3: EM Invoker

  1. All ...MultimodalEMInvoker are deprecated and are replaced with their ...LMInvoker counterpart.

    1. e.g., TwelveLabsMultimodalEMInvoker is replaced by TwelveLabsEMInvoker.

  2. Google classes have been merged into a single class: GoogleEMInvoker.

  3. All EM invoker has been migrated to use native providers SDK instead of LangChain implementation.

Part 4: Catalog

  1. Catalogs has been simplified.

  2. Prompt builder catalog will only require these columns:

    1. name

    2. system

    3. user

  3. LMRP catalog will only require these columns:

    1. name

    2. system_template

    3. user_template

    4. model_id, e.g. openai/gpt-4.1-nano

    5. credentials, will be loaded from env var, e.g. OPENAI_API_KEY

    6. config, dictionary to be passed to the LM invoker, e.g. {"default_hyperparameters": {"temperature": 0.7}}

    7. output_parser_type , either none or json

Part 5: Miscellaneous Stuff

  1. ModelId and ModelProvider should now be imported from gllm_inference.schema instead of gllm_inference.builder

  2. retry and RetryConfig should now be imported from gllm_core.utils.retry instead of gllm_inference.utils.retry

Last updated