Adapter

Adapter for guardrail components.

NeMoLMAdapter(model=None, lm_invoker=None, fallback_lms=None, **kwargs)

NeMo LLMModel protocol adapter that delegates to a gllm_inference LM invoker.

Pass lm_invoker to supply the primary invoker; the model identifier is derived from it automatically via lm_invoker.model_id. The model parameter is optional and exists so NeMo's framework can set it when the adapter is registered as a provider — you do not need to pass it directly.

Attributes:

Name	Type	Description
model	str | None	Model identifier string.
lm_invoker	BaseLMInvoker | None	Primary LM invoker.
fallback_lms	list[BaseLMInvoker]	Ordered fallback invokers tried on BaseInvokerError or TimeoutError from the primary.
model_name	str	Identifier used by NeMo for logging and serialization.
provider_name	str | None	Provider name registered with NeMo Guardrails.
provider_url	str | None	Provider URL; not applicable for this adapter.

Initializes a new instance of the NeMoLMAdapter class.

Parameters:

Name	Type	Description	Default
model	str | None	Model identifier string. Defaults to None.	None
lm_invoker	BaseLMInvoker | None	Primary LM invoker. Defaults to None.	None
fallback_lms	list[BaseLMInvoker] | None	Ordered fallback invokers tried on BaseInvokerError or TimeoutError from the primary. Defaults to None.	None
**kwargs	Any	Additional keyword arguments forwarded from the NeMo framework.	{}

model_name property

Identifier used by NeMo for logging and serialization.

provider_name property

Provider name registered with NeMo Guardrails.

provider_url property

Provider URL; not applicable for this adapter.

generate_async(prompt, *, stop=None, **kwargs) async

Asynchronously generate a response via the invoker.

Tries the primary invoker first. On BaseInvokerError or TimeoutError, falls back to each invoker in fallback_lms in order until one succeeds or the list is exhausted, then re-raises the last exception.

Parameters:

Name	Type	Description	Default
prompt	str | list[ChatMessage]	Input prompt as a string or list of NeMo chat messages.	required
stop	list[str] | None	Stop sequences. Defaults to None.	None
**kwargs	Any	Optional runtime parameters forwarded to the invoker.	{}

Returns:

Name	Type	Description
LLMResponse	LLMResponse	The generated response from the first successful invoker.

Raises:

Type	Description
BaseInvokerError	Re-raised from the last fallback invoker if all fail.
TimeoutError	Re-raised from the last fallback invoker if all fail.

stream_async(prompt, *, stop=None, **kwargs) async

Stream a response, yielding text deltas as they are generated.

Runs the invocation in the background while forwarding text deltas emitted by the invoker's event emitter as they arrive. A final chunk carrying the finish reason and token usage is yielded once the invocation completes. Note that the finish reason is always "stop", since LMOutput does not expose the underlying provider's finish reason.

Parameters:

Name	Type	Description	Default
prompt	str | list[ChatMessage]	Input prompt as a string or list of NeMo chat messages.	required
stop	list[str] | None	Stop sequences. Defaults to None.	None
**kwargs	Any	Optional runtime parameters forwarded to the invoker.	{}

Yields:

Name	Type	Description
LLMResponseChunk	AsyncIterator[LLMResponseChunk]	Incremental chunks of the response as they are generated, followed by a final chunk carrying the finish reason and token usage.