Struct PromptContext 

pub struct PromptContext { /* private fields */ }

Manages the cross-utterance prompt prefix fed to the streaming decoder.

Call update_after_eou immediately after an EOU fires, passing the full committed token list from Committer::committed_tokens. Then pass prompt_tokens() into DecodeContext::prompt_tokens for every window until the next EOU.

Implementations

impl PromptContext

pub fn new(max_prompt_tokens: usize, prevtext_token_id: Option<u32>) -> Self

prevtext_token_id: result of tokenizer.token_to_id("<|prevtext|>"). Pass None for English-only models that lack <|prevtext|> in their vocabulary.

pub fn update_after_eou( &mut self, committed_tokens: &[TokenEmit], max_new_tokens: usize, )

Update the stored prompt after an EOU.

Extracts the last max_prompt_tokens regular-token IDs (filter: !is_special), prepends <|prevtext|> when available and the tail is non-empty, and caps the total so that prompt.len() + sot(1) + lang(1) + transcribe(1) + max_new_tokens ≤ 448.

pub fn prompt_tokens(&self) -> &[u32]

Returns the prompt tokens to pass into DecodeContext::prompt_tokens. Empty before the first EOU or after reset.

pub fn reset(&mut self)

Clear the stored prompt (call on stream restart or explicit reset).