engines.rs 6.63 KB
Newer Older
1
2
3
4
5
6
7
8
9
10
11
12
13
14
// SPDX-FileCopyrightText: Copyright (c) 2024-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
// SPDX-License-Identifier: Apache-2.0
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
Ryan Olson's avatar
Ryan Olson committed
15
16
17
18
19
20
21
22
23

#![allow(dead_code)]

use std::{future::Future, pin::Pin, sync::Arc};

use async_trait::async_trait;
use futures::Stream;
use tokio::sync::mpsc;

Neelay Shah's avatar
Neelay Shah committed
24
use triton_distributed_runtime::engine::{
Ryan Olson's avatar
Ryan Olson committed
25
26
27
28
    AsyncEngine, AsyncEngineContext, AsyncEngineContextProvider, AsyncEngineStream,
    Data as DataType, Engine, EngineStream,
};

Neelay Shah's avatar
Neelay Shah committed
29
use triton_distributed_runtime::pipeline::{
Ryan Olson's avatar
Ryan Olson committed
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
    context::{Context, StreamContext},
    Error, ManyOut, SingleIn,
};

pub type AsyncFn<T, U> = dyn Fn(T) -> Pin<Box<dyn Future<Output = U> + Send>> + Send + Sync;

#[derive(Clone)]
// Define a struct that holds an async closure
pub struct AsyncProcessor<T, U> {
    func: Arc<AsyncFn<T, U>>,
}

impl<T, U> AsyncProcessor<T, U>
where
    T: Send + 'static,
    U: Send + 'static,
{
    // Define a `new` method that captures the already pinned async block
    pub fn new<F, Fut>(f: F) -> Self
    where
        F: Fn(T) -> Fut + Send + Sync + 'static,
        Fut: Future<Output = U> + Send + 'static,
    {
        // Wrap the closure in Arc and Box it for internal management
        AsyncProcessor {
            func: Arc::new(move |input: T| Box::pin(f(input))),
        }
    }

    // Method to execute the captured async function
    pub async fn process(&self, input: T) -> U {
        (self.func)(input).await
    }
}

#[derive(Debug, Clone)]
pub struct ResponseSource<T: Send + Sync + 'static> {
    tx: mpsc::Sender<T>,
    ctx: StreamContext,
}

impl<T: Send + Sync + 'static> ResponseSource<T> {
    fn new(tx: mpsc::Sender<T>, ctx: StreamContext) -> Self {
        ResponseSource { tx, ctx }
    }

    /// Emit a response to the stream
    pub async fn emit(&self, data: T) -> Result<(), ()> {
        self.tx.send(data).await.map_err(|_| ())
    }

    /// Check if a stop has been requested
    pub fn stop_requested(&self) -> bool {
        self.ctx.is_stopped()
    }

    /// Yield control until a stop is requested
    /// This is useful in a tokio::select! block
    pub async fn stopped(&self) {
        self.ctx.stopped().await;
    }
}

pub type AsyncGenerator<Req, Resp> = AsyncProcessor<(Req, ResponseSource<Resp>), ()>;

pub struct ReceiverStream<Resp: DataType> {
    receiver: tokio::sync::mpsc::Receiver<Resp>,
    context: Arc<dyn AsyncEngineContext>,
}

impl<Resp: DataType> ReceiverStream<Resp> {
    pub fn new(
        receiver: tokio::sync::mpsc::Receiver<Resp>,
        context: Arc<dyn AsyncEngineContext>,
    ) -> Self {
        Self { receiver, context }
    }
}

impl<Resp: DataType> Stream for ReceiverStream<Resp> {
    type Item = Resp;

    fn poll_next(
        mut self: Pin<&mut Self>,
        cx: &mut std::task::Context<'_>,
    ) -> std::task::Poll<Option<Self::Item>> {
        // if self.context.stop_issued() {
        //     return std::task::Poll::Ready(None);
        // }

        // Pinning the receiver to safely call poll_recv
        Pin::new(&mut self.receiver).poll_recv(cx)
    }
}

impl<Resp: DataType> std::fmt::Debug for ReceiverStream<Resp> {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        f.debug_struct("ReceiverStream")
            .field("context", &self.context)
            .finish()
    }
}

impl<Resp: DataType> AsyncEngineStream<Resp> for ReceiverStream<Resp> {}

impl<Resp: DataType> AsyncEngineContextProvider for ReceiverStream<Resp> {
    fn context(&self) -> Arc<dyn AsyncEngineContext> {
        self.context.clone()
    }
}

pub struct LlmdbaEngine<Req: DataType, Resp: DataType> {
    lambda: Arc<AsyncGenerator<Req, Resp>>,
}

impl<Req: DataType, Resp: DataType> LlmdbaEngine<Req, Resp> {
    fn new(lambda: AsyncGenerator<Req, Resp>) -> Self {
        LlmdbaEngine {
            lambda: Arc::new(lambda),
        }
    }

    pub fn from_generator(
        generator: AsyncGenerator<Req, Resp>,
    ) -> Engine<SingleIn<Req>, ManyOut<Resp>, Error> {
        Arc::new(LlmdbaEngine::new(generator))
    }
}

#[async_trait]
impl<Req: DataType, Resp: DataType> AsyncEngine<SingleIn<Req>, ManyOut<Resp>, Error>
    for LlmdbaEngine<Req, Resp>
{
    async fn generate(&self, request: Context<Req>) -> Result<EngineStream<Resp>, Error> {
        let (tx, rx) = mpsc::channel::<Resp>(1);
        let (req, ctx) = request.transfer(());
        let ctx: StreamContext = ctx.into();
        let s = ResponseSource::new(tx, ctx.clone());

        let lambda = self.lambda.clone();
        let _handle = tokio::spawn(async move { lambda.process((req, s)).await });

        let ctx = Arc::new(ctx);
        let stream = ReceiverStream::<Resp>::new(rx, ctx);
        let stream = Box::pin(stream);
        Ok(stream)
    }
}

#[cfg(test)]
mod tests {

    use futures::StreamExt;

    use super::*;

    #[tokio::test]
    async fn test_async_processor() {
        let processor = AsyncProcessor::new(move |x: i32| {
            async move {
                // Simulate some async work
                tokio::time::sleep(std::time::Duration::from_millis(100)).await;
                format!("Processed value: {}", x)
            }
        });

        // Use the processor to run the async closure
        let result = processor.process(42).await;
        println!("{}", result); // Output: Processed value: 42

        let result2 = processor.process(100).await;
        println!("{}", result2); // Output: Processed value: 100
    }

    #[tokio::test]
    async fn test_generator() {
        let generator = AsyncGenerator::<String, String>::new(|(req, stream)| async move {
            let chars = req.chars().collect::<Vec<char>>();
            for c in chars {
                match stream.emit(c.to_string()).await {
                    Ok(_) => {}
                    Err(_) => break,
                }
                tokio::time::sleep(std::time::Duration::from_millis(100)).await;
            }
        });

        let engine = LlmdbaEngine::new(generator);

        let mut stream = engine.generate("test".to_string().into()).await.unwrap();

        let mut counter = 0;
        while let Some(_output) = stream.next().await {
            counter += 1;
        }

        assert_eq!(counter, 4);
    }
}