pub trait DraftProvider: Send + Sync {
// Required methods
fn draft<'life0, 'life1, 'async_trait>(
&'life0 self,
messages: &'life1 [ChatMessage],
n_tokens: usize,
) -> Pin<Box<dyn Future<Output = Result<Vec<String>, DraftError>> + Send + 'async_trait>>
where Self: 'async_trait,
'life0: 'async_trait,
'life1: 'async_trait;
fn model_name(&self) -> &str;
}Expand description
Trait for a small/fast draft model that proposes candidate tokens.
The draft provider generates N candidate token strings given the current conversation context. These candidates are then verified by the target model.
Required Methods§
Sourcefn draft<'life0, 'life1, 'async_trait>(
&'life0 self,
messages: &'life1 [ChatMessage],
n_tokens: usize,
) -> Pin<Box<dyn Future<Output = Result<Vec<String>, DraftError>> + Send + 'async_trait>>where
Self: 'async_trait,
'life0: 'async_trait,
'life1: 'async_trait,
fn draft<'life0, 'life1, 'async_trait>(
&'life0 self,
messages: &'life1 [ChatMessage],
n_tokens: usize,
) -> Pin<Box<dyn Future<Output = Result<Vec<String>, DraftError>> + Send + 'async_trait>>where
Self: 'async_trait,
'life0: 'async_trait,
'life1: 'async_trait,
Generate n_tokens candidate continuations given the conversation so far.
Returns a Vec of token strings (one per draft position).
Sourcefn model_name(&self) -> &str
fn model_name(&self) -> &str
Name of the draft model.