Struct StreamingCompletionResponse 

pub struct StreamingCompletionResponse<U = Usage> {
    pub usage: U,
    pub finish_reason: Option<FinishReason>,
    pub response_id: Option<String>,
    pub model: Option<String>,
    pub provider_request_id: Option<String>,
    pub logprobs: Option<Value>,
    pub additional_params: Option<AdditionalParams>,
}

Final streaming response. U is the provider’s streaming usage payload (Usage for OpenAI itself; providers with richer usage accounting, e.g. Mistral and DeepSeek, substitute their own via OpenAICompatibleProvider::StreamingUsage).

This is the provider-native terminal record yielded by GenericCompletionModel::raw_stream. The normalized path maps it into a StreamFinal exactly once, through normalize_stream.

Fields

usage: U

Usage reported on the stream’s terminal event.

finish_reason: Option<FinishReason>

Why the model stopped generating, when the stream reported it.

Normalized out of the OpenAI-compatible finish_reason vocabulary, with unrecognized values preserved verbatim. The Stop -> ToolCalls upgrade is deliberately not applied here: it belongs to normalize_stream, the only place that sees which tool calls the stream actually emitted.

response_id: Option<String>

Provider-assigned response identifier, when the stream emitted one.

model: Option<String>

Provider-reported model identifier, when the stream emitted one.

provider_request_id: Option<String>

The transport request id from the SSE connection’s x-request-id response header — not part of any stream frame; stamped by the transport. None when the provider did not report one.

logprobs: Option<Value>

Token log probabilities accumulated from all primary-choice chunks.

This stays provider-native on GenericCompletionModel::raw_stream: normalized completions do not currently model log probabilities, just as the blocking normalized path omits Choice::logprobs while its raw response retains them.

additional_params: Option<AdditionalParams>

Provider-specific top-level fields accumulated from the stream’s chunks, such as OpenAI’s service_tier and system_fingerprint or OpenRouter’s routed provider.

Implementations

impl<U> StreamingCompletionResponse<U>

pub fn new(usage: U) -> StreamingCompletionResponse<U>

Create a terminal record carrying usage; the optional metadata starts unset.

Trait Implementations

impl<U> Clone for StreamingCompletionResponse<U>

where U: Clone,

fn clone(&self) -> StreamingCompletionResponse<U>

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl<U> Debug for StreamingCompletionResponse<U>

where U: Debug,

fn fmt(&self, f: &mut Formatter<'_>) -> Result<(), Error>

Formats the value using the given formatter.

impl<'de, U> Deserialize<'de> for StreamingCompletionResponse<U>

where U: Deserialize<'de>,

fn deserialize<__D>( __deserializer: __D, ) -> Result<StreamingCompletionResponse<U>, <__D as Deserializer<'de>>::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl<U> Serialize for StreamingCompletionResponse<U>

where U: Serialize,

fn serialize<__S>( &self, __serializer: __S, ) -> Result<<__S as Serializer>::Ok, <__S as Serializer>::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.