| 1 | //! Token streaming for plain chat. |
| 2 | |
| 3 | use std::pin::Pin; |
| 4 | use std::task::{Context, Poll}; |
| 5 | |
| 6 | use futures_core::Stream; |
| 7 | use onde::inference::types::StreamChunk; |
| 8 | use onde::inference::InferenceError; |
| 9 | use tokio::sync::mpsc::Receiver; |
| 10 | |
| 11 | /// The reply to one message, as a stream of text deltas. |
| 12 | /// |
| 13 | /// Yields `Ok(delta)` for each piece and ends after the last one. A failure |
| 14 | /// mid-stream is a final `Err`. Dropping the stream stops generation: `onde` |
| 15 | /// notices the closed channel and keeps only what was delivered. |
| 16 | /// |
| 17 | /// The finished turn is added to the engine's history when the stream ends, |
| 18 | /// as with [`Ed::send`](crate::Ed::send). |
| 19 | pub struct TextStream { |
| 20 | rx: Receiver<StreamChunk>, |
| 21 | finished: bool, |
| 22 | } |
| 23 | |
| 24 | impl TextStream { |
| 25 | pub(crate) fn new(rx: Receiver<StreamChunk>) -> Self { |
| 26 | Self { |
| 27 | rx, |
| 28 | finished: false, |
| 29 | } |
| 30 | } |
| 31 | } |
| 32 | |
| 33 | impl TextStream { |
| 34 | /// The next delta, `None` when the reply is complete. For callers without a stream |
| 35 | /// combinator library. |
| 36 | pub async fn next(&mut self) -> Option<Result<String, InferenceError>> { |
| 37 | std::future::poll_fn(|cx| Pin::new(&mut *self).poll_next(cx)).await |
| 38 | } |
| 39 | } |
| 40 | |
| 41 | /// `onde` reports a failed stream as a final chunk whose `finish_reason` |
| 42 | /// starts with `error`. |
| 43 | pub(crate) fn chunk_error(chunk: &StreamChunk) -> Option<InferenceError> { |
| 44 | chunk |
| 45 | .finish_reason |
| 46 | .as_deref() |
| 47 | .filter(|reason| reason.starts_with("error")) |
| 48 | .map(|reason| InferenceError::Inference { |
| 49 | reason: reason.trim_start_matches("error:").trim().to_owned(), |
| 50 | }) |
| 51 | } |
| 52 | |
| 53 | impl Stream for TextStream { |
| 54 | type Item = Result<String, InferenceError>; |
| 55 | |
| 56 | fn poll_next(mut self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<Option<Self::Item>> { |
| 57 | loop { |
| 58 | if self.finished { |
| 59 | return Poll::Ready(None); |
| 60 | } |
| 61 | match self.rx.poll_recv(cx) { |
| 62 | Poll::Pending => return Poll::Pending, |
| 63 | Poll::Ready(None) => { |
| 64 | self.finished = true; |
| 65 | return Poll::Ready(None); |
| 66 | } |
| 67 | Poll::Ready(Some(chunk)) => { |
| 68 | if chunk.done { |
| 69 | self.finished = true; |
| 70 | return Poll::Ready(chunk_error(&chunk).map(Err)); |
| 71 | } |
| 72 | if chunk.delta.is_empty() { |
| 73 | continue; |
| 74 | } |
| 75 | return Poll::Ready(Some(Ok(chunk.delta))); |
| 76 | } |
| 77 | } |
| 78 | } |
| 79 | } |
| 80 | } |
| 81 | |
| 82 | #[cfg(test)] |
| 83 | mod tests { |
| 84 | use super::*; |
| 85 | |
| 86 | fn chunk(delta: &str, done: bool, reason: Option<&str>) -> StreamChunk { |
| 87 | StreamChunk { |
| 88 | delta: delta.to_owned(), |
| 89 | done, |
| 90 | finish_reason: reason.map(str::to_owned), |
| 91 | } |
| 92 | } |
| 93 | |
| 94 | #[tokio::test] |
| 95 | async fn yields_deltas_then_ends() { |
| 96 | let (tx, rx) = tokio::sync::mpsc::channel(8); |
| 97 | let mut stream = TextStream::new(rx); |
| 98 | tx.send(chunk("he", false, None)).await.unwrap(); |
| 99 | tx.send(chunk("llo", false, None)).await.unwrap(); |
| 100 | tx.send(chunk("", true, Some("stop"))).await.unwrap(); |
| 101 | |
| 102 | assert_eq!(stream.next().await.unwrap().unwrap(), "he"); |
| 103 | assert_eq!(stream.next().await.unwrap().unwrap(), "llo"); |
| 104 | assert!(stream.next().await.is_none()); |
| 105 | } |
| 106 | |
| 107 | #[tokio::test] |
| 108 | async fn a_failed_stream_ends_with_an_error() { |
| 109 | let (tx, rx) = tokio::sync::mpsc::channel(8); |
| 110 | let mut stream = TextStream::new(rx); |
| 111 | tx.send(chunk("", true, Some("error: boom"))).await.unwrap(); |
| 112 | |
| 113 | let err = stream.next().await.unwrap().unwrap_err(); |
| 114 | assert!(err.to_string().contains("boom")); |
| 115 | assert!(stream.next().await.is_none()); |
| 116 | } |
| 117 | } |