feat: Improve error handling

Dockerfile

@@ -18,6 +18,7 @@ ENV LANG=C.UTF-8 \
     MODEL_NAME=bigscience/bloom \
     NUM_GPUS=8 \
     CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 \
+    NCCL_ASYNC_ERROR_HANDLING=1 \
     CUDA_HOME=/usr/local/cuda \
     LD_LIBRARY_PATH="/opt/miniconda/envs/text-generation/lib:/usr/local/cuda/lib64:/usr/local/cuda/extras/CUPTI/lib64:$LD_LIBRARY_PATH" \
     CONDA_DEFAULT_ENV=text-generation \
@@ -51,7 +52,7 @@ RUN cd server && \
     /opt/miniconda/envs/text-generation/bin/pip install . --no-cache-dir
 
 # Install router
-COPY --from=builder /usr/local/cargo/bin/bloom-inference /usr/local/bin/bloom-inference
+COPY --from=builder /usr/local/cargo/bin/text-generation-router /usr/local/bin/text-generation-router
 
 COPY run.sh .
 RUN chmod +x run.sh

README.md

@@ -48,5 +48,4 @@ python server/bloom_inference/main.py bigscience/bloom --num-gpus 8 --shard-dire
 - [ ] Add tests
 - [ ] Add shutdown logic in router and server
 - [ ] Improve multi-processing logic in server
-- [ ] Improve error handling everywhere
 - [ ] Improve past key layer indexing?

aml/deployment.yaml

@@ -8,7 +8,7 @@ environment_variables:
   MODEL_NAME: bigscience/bloom
   NUM_GPUS: 8
 environment:
-  image: db4c2190dd824d1f950f5d1555fbadf0.azurecr.io/text-generation:0.1
+  image: db4c2190dd824d1f950f5d1555fbadf0.azurecr.io/text-generation:0.3
   inference_config:
     liveness_route:
       port: 3000
@@ -24,15 +24,15 @@ request_settings:
   request_timeout_ms: 90000
   max_concurrent_requests_per_instance: 256
 liveness_probe:
-  initial_delay: 300
+  initial_delay: 600
   timeout: 20
-  period: 60
+  period: 120
   success_threshold: 1
-  failure_threshold: 60
+  failure_threshold: 3
 readiness_probe:
-  initial_delay: 300
+  initial_delay: 600
   timeout: 20
-  period: 60
+  period: 120
   success_threshold: 1
-  failure_threshold: 60
+  failure_threshold: 3
 instance_count: 1

router/Cargo.lock

@@ -149,22 +149,6 @@ dependencies = [
  "generic-array",
 ]
 
-[[package]]
-name = "bloom-inference"
-version = "0.1.0"
-dependencies = [
- "axum",
- "bloom-inference-client",
- "futures",
- "parking_lot",
- "serde",
- "serde_json",
- "tokenizers",
- "tokio",
- "tracing",
- "tracing-subscriber",
-]
-
 [[package]]
 name = "bloom-inference-client"
 version = "0.1.0"
@@ -1669,6 +1653,23 @@ dependencies = [
  "winapi",
 ]
 
+[[package]]
+name = "text-generation-router"
+version = "0.1.0"
+dependencies = [
+ "axum",
+ "bloom-inference-client",
+ "futures",
+ "parking_lot",
+ "serde",
+ "serde_json",
+ "thiserror",
+ "tokenizers",
+ "tokio",
+ "tracing",
+ "tracing-subscriber",
+]
+
 [[package]]
 name = "textwrap"
 version = "0.11.0"

router/Cargo.toml

@@ -1,8 +1,15 @@
 [package]
-name = "bloom-inference"
+name = "text-generation-router"
 version = "0.1.0"
 edition = "2021"
 
+[lib]
+path = "src/lib.rs"
+
+[[bin]]
+name = "text-generation-router"
+path = "src/main.rs"
+
 [dependencies]
 axum = { version = "0.5.16", features = ["json", "serde_json"] }
 bloom-inference-client = { path = "client" }
@@ -10,6 +17,7 @@ futures = "0.3.24"
 parking_lot = "0.12.1"
 serde = "1.0.145"
 serde_json = "1.0.85"
+thiserror = "1.0.37"
 tokenizers = "0.13.0"
 tokio = { version = "1.21.1", features = ["rt-multi-thread", "parking_lot", "sync"] }
 tracing = "0.1.36"

router/client/src/client.rs

@@ -1,45 +1,37 @@
 use crate::pb::generate::v1::text_generation_service_client::TextGenerationServiceClient;
 use crate::pb::generate::v1::*;
 use crate::Result;
-use std::time::Duration;
 use tonic::transport::{Channel, Uri};
-use tower::timeout::Timeout;
 use tracing::*;
 
 /// BLOOM Inference gRPC client
 #[derive(Clone)]
 pub struct Client {
-    stub: TextGenerationServiceClient<Timeout<Channel>>,
+    stub: TextGenerationServiceClient<Channel>,
 }
 
 impl Client {
-    /// Returns a client connected to the given url. Requests exceeding timeout will fail.
-    pub async fn connect(uri: Uri, timeout: Duration) -> Self {
-        let channel = Channel::builder(uri)
-            .connect()
-            .await
-            .expect("Transport error");
-        let timeout_channel = Timeout::new(channel, timeout);
+    /// Returns a client connected to the given url
+    pub async fn connect(uri: Uri) -> Result<Self> {
+        let channel = Channel::builder(uri).connect().await?;
 
-        Self {
-            stub: TextGenerationServiceClient::new(timeout_channel),
-        }
+        Ok(Self {
+            stub: TextGenerationServiceClient::new(channel),
+        })
     }
 
-    /// Returns a client connected to the given unix socket. Requests exceeding timeout will fail.
-    pub async fn connect_uds(path: String, timeout: Duration) -> Self {
+    /// Returns a client connected to the given unix socket
+    pub async fn connect_uds(path: String) -> Result<Self> {
         let channel = Channel::from_shared("http://[::]:50051".to_string())
             .unwrap()
             .connect_with_connector(tower::service_fn(move |_: Uri| {
                 tokio::net::UnixStream::connect(path.clone())
             }))
-            .await
-            .expect("Transport error");
-        let timeout_channel = Timeout::new(channel, timeout);
+            .await?;
 
-        Self {
-            stub: TextGenerationServiceClient::new(timeout_channel),
-        }
+        Ok(Self {
+            stub: TextGenerationServiceClient::new(channel),
+        })
     }
 
     #[instrument(skip(self))]

router/client/src/lib.rs

@@ -8,23 +8,27 @@ pub use client::Client;
 pub use pb::generate::v1::{Batch, GeneratedText, LogitsWarperParameters, Request};
 pub use sharded_client::ShardedClient;
 use thiserror::Error;
-pub use tonic::transport::Uri;
+pub use tonic::transport;
 use tonic::Status;
 
 #[derive(Error, Debug, Clone)]
-#[error("Text generation client error: {msg:?}")]
-pub struct ClientError {
-    msg: String,
-    // source: Status,
+pub enum ClientError {
+    #[error("Could not connect to Text Generation server: {0:?}")]
+    Connection(String),
+    #[error("Server error: {0:?}")]
+    Generation(String),
 }
 
 impl From<Status> for ClientError {
     fn from(err: Status) -> Self {
-        Self {
-            msg: err.to_string(),
-            // source: err,
+        Self::Generation(err.to_string())
     }
 }
+
+impl From<transport::Error> for ClientError {
+    fn from(err: transport::Error) -> Self {
+        Self::Connection(err.to_string())
+    }
 }
 
 pub type Result<T> = std::result::Result<T, ClientError>;

router/client/src/sharded_client.rs

@@ -1,7 +1,6 @@
 use crate::Result;
 use crate::{Batch, Client, GeneratedText};
 use futures::future::join_all;
-use std::time::Duration;
 use tokio::sync::{broadcast, mpsc};
 use tonic::transport::Uri;
 
@@ -69,24 +68,22 @@ impl ShardedClient {
         Self { request_tx }
     }
 
-    async fn from_master_client(mut master_client: Client) -> Self {
+    async fn from_master_client(mut master_client: Client) -> Result<Self> {
         let uris = master_client.service_discovery().await.unwrap();
-        let futures = uris
-            .into_iter()
-            .map(|path| Client::connect_uds(path, Duration::from_secs(5)));
-        let clients = join_all(futures).await;
-        Self::new(clients)
+        let futures = uris.into_iter().map(|path| Client::connect_uds(path));
+        let clients: Result<Vec<Client>> = join_all(futures).await.into_iter().collect();
+        Ok(Self::new(clients?))
     }
 
-    /// Returns a client connected to the given url. Requests exceeding timeout will fail.
-    pub async fn connect(uri: Uri, timeout: Duration) -> Self {
-        let master_client = Client::connect(uri, timeout).await;
+    /// Returns a client connected to the given url
+    pub async fn connect(uri: Uri) -> Result<Self> {
+        let master_client = Client::connect(uri).await?;
         Self::from_master_client(master_client).await
     }
 
-    /// Returns a client connected to the given unix socket. Requests exceeding timeout will fail.
-    pub async fn connect_uds(path: String, timeout: Duration) -> Self {
-        let master_client = Client::connect_uds(path, timeout).await;
+    /// Returns a client connected to the given unix socket
+    pub async fn connect_uds(path: String) -> Result<Self> {
+        let master_client = Client::connect_uds(path).await?;
         Self::from_master_client(master_client).await
     }
 

router/src/batcher.rs

@@ -1,13 +1,30 @@
 use crate::server::GenerateRequest;
-use crate::Db;
+use crate::{Db, Entry};
+use axum::http::StatusCode;
 use bloom_inference_client::{Batch, ClientError, GeneratedText, ShardedClient};
 use std::future::Future;
 use std::sync::Arc;
+use thiserror::Error;
 use tokio::sync::{oneshot, Notify};
 
 const MAX_LENGTH: usize = 128;
 
-pub struct InferError {}
+#[derive(Debug, Error)]
+pub enum InferError {
+    #[error("Request failed during generation: {0}")]
+    GenerationError(String),
+    #[error("Model is overloaded")]
+    Overloaded,
+}
+
+impl From<InferError> for (StatusCode, String) {
+    fn from(err: InferError) -> Self {
+        match err {
+            InferError::GenerationError(_) => (StatusCode::INTERNAL_SERVER_ERROR, err.to_string()),
+            InferError::Overloaded => (StatusCode::TOO_MANY_REQUESTS, err.to_string()),
+        }
+    }
+}
 
 #[derive(Clone)]
 pub(crate) struct Batcher {
@@ -37,14 +54,18 @@ impl Batcher {
         request: GenerateRequest,
     ) -> Result<String, InferError> {
         if self.db.len() > MAX_LENGTH {
-            return Err(InferError {});
+            return Err(InferError::Overloaded);
         }
         let (request_tx, request_rx) = oneshot::channel();
-        self.db.append(input_length, request, request_tx);
+        self.db.append(Entry {
+            request,
+            response_tx: request_tx,
+            input_length,
+        });
         self.shared.batching_task.notify_waiters();
         match request_rx.await.unwrap() {
             Ok(output) => Ok(output),
-            Err(_) => Err(InferError {}),
+            Err(err) => Err(InferError::GenerationError(err.to_string())),
         }
     }
 }
@@ -108,7 +129,6 @@ async fn wrap_future(
             next_batch
         }
         Err(err) => {
-            println!("{:?}", err);
             send_error(err, request_ids, db);
             None
         }
@@ -117,14 +137,18 @@ async fn wrap_future(
 
 fn send_error(error: ClientError, request_ids: Vec<u64>, db: &Db) {
     request_ids.into_iter().for_each(|id| {
-        let (_, response_tx) = db.remove(&id).unwrap();
-        response_tx.send(Err(error.clone())).unwrap_or(());
+        let entry = db.remove(&id).expect("ID not found in db. This is a bug.");
+        // unwrap_or is valid here as we don't care if the receiver is gone.
+        entry.response_tx.send(Err(error.clone())).unwrap_or(());
     });
 }
 
 fn send_generated(finished: Vec<GeneratedText>, db: &Db) {
     finished.into_iter().for_each(|output| {
-        let (_, response_tx) = db.remove(&output.request.unwrap().id).unwrap();
-        response_tx.send(Ok(output.output)).unwrap_or(());
+        let entry = db
+            .remove(&output.request.unwrap().id)
+            .expect("ID not found in db. This is a bug.");
+        // unwrap_or is valid here as we don't care if the receiver is gone.
+        entry.response_tx.send(Ok(output.output)).unwrap_or(());
     });
 }

router/src/db.rs

@@ -1,11 +1,29 @@
 /// This code is massively inspired by Tokio mini-redis
-use crate::server::GenerateRequest;
+use crate::server::{GenerateParameters, GenerateRequest};
 use bloom_inference_client::{Batch, ClientError, LogitsWarperParameters, Request};
 use parking_lot::RwLock;
 use std::collections::BTreeMap;
 use std::sync::Arc;
 use tokio::sync::oneshot::Sender;
 
+#[derive(Debug)]
+pub(crate) struct Entry {
+    pub request: GenerateRequest,
+    pub response_tx: Sender<Result<String, ClientError>>,
+    pub input_length: usize,
+}
+
+impl From<GenerateParameters> for LogitsWarperParameters {
+    fn from(parameters: GenerateParameters) -> Self {
+        Self {
+            temperature: parameters.temperature,
+            top_k: parameters.top_k as u32,
+            top_p: parameters.top_p,
+            do_sample: parameters.do_sample,
+        }
+    }
+}
+
 #[derive(Debug, Clone)]
 pub(crate) struct Db {
     pub shared: Arc<Shared>,
@@ -18,7 +36,7 @@ pub struct Shared {
 
 #[derive(Debug)]
 struct State {
-    entries: BTreeMap<u64, (Request, Sender<Result<String, ClientError>>)>,
+    entries: BTreeMap<u64, Entry>,
 
     /// Identifier to use for the next expiration. Each expiration is associated
     /// with a unique identifier. See above for why.
@@ -44,37 +62,16 @@ impl Db {
         Self { shared }
     }
 
-    pub(crate) fn append(
-        &self,
-        input_length: usize,
-        request: GenerateRequest,
-        sender: Sender<Result<String, ClientError>>,
-    ) {
+    pub(crate) fn append(&self, entry: Entry) {
         let mut state = self.shared.state.write();
 
         let id = state.next_id;
         state.next_id += 1;
 
-        let parameters = Some(LogitsWarperParameters {
-            temperature: request.parameters.temperature,
-            top_k: request.parameters.top_k,
-            top_p: request.parameters.top_p,
-            do_sample: request.parameters.do_sample,
-        });
-        let request = Request {
-            id,
-            inputs: request.inputs,
-            input_length: input_length as u32,
-            parameters,
-            max_new_tokens: request.parameters.max_new_tokens,
-        };
-        state.entries.insert(id, (request, sender));
+        state.entries.insert(id, entry);
     }
 
-    pub(crate) fn remove(
-        &self,
-        id: &u64,
-    ) -> Option<(Request, Sender<Result<String, ClientError>>)> {
+    pub(crate) fn remove(&self, id: &u64) -> Option<Entry> {
         let mut state = self.shared.state.write();
         state.entries.remove(id)
     }
@@ -91,7 +88,15 @@ impl Db {
             .entries
             .range(state.next_batch_start_id..)
             .take(max_size)
-            .map(|(_, (request, _))| request.clone())
+            .map(|(id, entry)| Request {
+                id: *id,
+                inputs: entry.request.inputs.clone(),
+                input_length: entry.input_length as u32,
+                parameters: Some(LogitsWarperParameters::from(
+                    entry.request.parameters.clone(),
+                )),
+                max_new_tokens: entry.request.parameters.max_new_tokens,
+            })
             .collect();
 
         if requests.is_empty() {

router/src/lib.rs

@@ -0,0 +1,8 @@
+mod batcher;
+mod db;
+pub mod server;
+mod validation;
+
+use batcher::Batcher;
+use db::{Db, Entry};
+use validation::Validation;

router/src/main.rs

@@ -1,21 +1,8 @@
 use bloom_inference_client::ShardedClient;
 use std::net::SocketAddr;
-use std::time::Duration;
+use text_generation_router::server;
 use tokenizers::Tokenizer;
 
-mod server;
-mod validation;
-
-use validation::Validation;
-
-mod db;
-
-use db::Db;
-
-mod batcher;
-
-use batcher::Batcher;
-
 fn main() -> Result<(), std::io::Error> {
     let tokenizer = Tokenizer::from_pretrained("bigscience/bloom", None).unwrap();
 
@@ -26,11 +13,9 @@ fn main() -> Result<(), std::io::Error> {
         .block_on(async {
             tracing_subscriber::fmt::init();
 
-            let sharded_client = ShardedClient::connect_uds(
-                "/tmp/bloom-inference-0".to_string(),
-                Duration::from_secs(5),
-            )
-            .await;
+            let sharded_client = ShardedClient::connect_uds("/tmp/bloom-inference-0".to_string())
+                .await
+                .expect("Could not connect to server");
             sharded_client
                 .clear_cache()
                 .await

router/src/server.rs

@@ -1,9 +1,9 @@
-use bloom_inference_client::ShardedClient;
 use crate::{Batcher, Validation};
 use axum::extract::Extension;
 use axum::http::StatusCode;
 use axum::routing::{get, post};
 use axum::{Json, Router};
+use bloom_inference_client::ShardedClient;
 use serde::Deserialize;
 use std::net::SocketAddr;
 use tokenizers::Tokenizer;
@@ -15,7 +15,7 @@ pub(crate) struct GenerateParameters {
     #[serde(default = "default_temperature")]
     pub temperature: f32,
     #[serde(default = "default_top_k")]
-    pub top_k: u32,
+    pub top_k: i32,
     #[serde(default = "default_top_p")]
     pub top_p: f32,
     #[serde(default = "default_do_sample")]
@@ -28,7 +28,7 @@ fn default_temperature() -> f32 {
     1.0
 }
 
-fn default_top_k() -> u32 {
+fn default_top_k() -> i32 {
     0
 }
 
@@ -62,8 +62,8 @@ pub(crate) struct GenerateRequest {
 }
 
 #[instrument(skip(state), fields(time, time_per_token))]
-async fn liveness(state: Extension<ServerState>) -> Result<(), StatusCode> {
-    let output = state
+async fn liveness(state: Extension<ServerState>) -> Result<(), (StatusCode, String)> {
+    state
         .infer
         .infer(
             1,
@@ -78,37 +78,27 @@ async fn liveness(state: Extension<ServerState>) -> Result<(), StatusCode> {
                 },
             },
         )
-        .await;
-
-    match output {
-        Ok(_) => Ok(()),
-        Err(_) => Err(StatusCode::INTERNAL_SERVER_ERROR),
-    }
+        .await?;
+    Ok(())
 }
 
 #[instrument(skip(state), fields(time, time_per_token))]
 async fn generate(
     state: Extension<ServerState>,
     req: Json<GenerateRequest>,
-) -> Result<Json<serde_json::Value>, StatusCode> {
+) -> Result<Json<serde_json::Value>, (StatusCode, String)> {
     let start = Instant::now();
 
-    let (input_length, validated_request) = match state
+    let (input_length, validated_request) = state
         .validation
         .validate(GenerateRequest {
             inputs: req.inputs.clone(),
             parameters: req.parameters.clone(),
         })
-        .await
-    {
-        Ok(result) => result,
-        Err(_) => return Err(StatusCode::INTERNAL_SERVER_ERROR),
-    };
+        .await?;
 
-    let output = state.infer.infer(input_length, validated_request).await;
+    let generated_text = state.infer.infer(input_length, validated_request).await?;
 
-    match output {
-        Ok(generated_text) => {
     tracing::Span::current().record("time", format!("{:?}", start.elapsed()));
     tracing::Span::current().record(
         "time_per_token",
@@ -120,9 +110,6 @@ async fn generate(
         "generated_text": generated_text,
     })))
 }
-        Err(_) => Err(StatusCode::INTERNAL_SERVER_ERROR),
-    }
-}
 
 #[derive(Clone)]
 struct ServerState {

router/src/validation.rs

@@ -1,9 +1,28 @@
 use crate::server::GenerateRequest;
+use axum::http::StatusCode;
+use thiserror::Error;
 use tokenizers::tokenizer::Tokenizer;
 use tokio::sync::{mpsc, oneshot};
 
-#[derive(Debug)]
-pub struct ValidationError {}
+#[derive(Error, Debug)]
+pub enum ValidationError {
+    #[error("Temperature must be strictly positive")]
+    Temperature,
+    #[error("Top p must be <= 0.0 or > 1.0")]
+    TopP,
+    #[error("Top k must be strictly positive")]
+    TopK,
+    #[error("Max New Tokens must be < 512")]
+    MaxNewTokens,
+    #[error("Inputs must have less than 512 tokens. Given: {0}")]
+    InputLength(usize),
+}
+
+impl From<ValidationError> for (StatusCode, String) {
+    fn from(err: ValidationError) -> Self {
+        (StatusCode::BAD_REQUEST, err.to_string())
+    }
+}
 
 type ValidationRequest = (
     GenerateRequest,
@@ -39,15 +58,23 @@ impl Validation {
 async fn validation_task(tokenizer: Tokenizer, mut receiver: mpsc::Receiver<ValidationRequest>) {
     while let Some((request, response_tx)) = receiver.recv().await {
         if request.parameters.temperature < 0.0 {
-            response_tx.send(Err(ValidationError {})).unwrap_or(());
+            response_tx
+                .send(Err(ValidationError::Temperature))
+                .unwrap_or(());
             continue;
         }
         if request.parameters.top_p <= 0.0 || request.parameters.top_p > 1.0 {
-            response_tx.send(Err(ValidationError {})).unwrap_or(());
+            response_tx.send(Err(ValidationError::TopP)).unwrap_or(());
+            continue;
+        }
+        if request.parameters.top_k < 0 {
+            response_tx.send(Err(ValidationError::TopK)).unwrap_or(());
             continue;
         }
         if request.parameters.max_new_tokens > 512 {
-            response_tx.send(Err(ValidationError {})).unwrap_or(());
+            response_tx
+                .send(Err(ValidationError::MaxNewTokens))
+                .unwrap_or(());
             continue;
         }
 
@@ -55,11 +82,12 @@ async fn validation_task(tokenizer: Tokenizer, mut receiver: mpsc::Receiver<Vali
         let input_length = inputs.len();
 
         if input_length > 512 {
-            response_tx.send(Err(ValidationError {})).unwrap_or(());
+            response_tx
+                .send(Err(ValidationError::InputLength(input_length)))
+                .unwrap_or(());
             continue;
         }
 
         response_tx.send(Ok((input_length, request))).unwrap_or(());
     }
-    println!("drop here");
 }

run.sh

@@ -1,10 +1,12 @@
 #!/usr/bin/env bash
 
-server_cmd="python server/bloom_inference/main.py $MODEL_NAME --num-gpus $NUM_GPUS --shard-directory $MODEL_BASE_PATH"
-$server_cmd &
+server_cmd="bloom-inference-server launcher $MODEL_NAME --num-gpus $NUM_GPUS --shard-directory $MODEL_BASE_PATH"
 
+# Run in background
+$server_cmd 2>&1 > /dev/null &
+
+# Check if server is running by checking if the unix socket is created
 FILE=/tmp/bloom-inference-0
-
 while :
   do
     if test -S "$FILE"; then
@@ -18,4 +20,11 @@ while :
 
 sleep 1
 
-exec "bloom-inference"
+# Run in background
+text-generation-router &
+
+# Wait for any process to exit
+wait -n
+
+# Exit with status of process that exited first
+exit $?

server/bloom_inference/cli.py

@@ -0,0 +1,42 @@
+import typer
+
+from pathlib import Path
+from torch.distributed.launcher import launch_agent, LaunchConfig
+from typing import Optional
+
+from bloom_inference import server
+
+app = typer.Typer()
+
+
+@app.command()
+def launcher(
+        model_name: str,
+        num_gpus: int = 1,
+        shard_directory: Optional[Path] = None,
+):
+    if num_gpus == 1:
+        serve(model_name, False, shard_directory)
+
+    else:
+        config = LaunchConfig(
+            min_nodes=1,
+            max_nodes=1,
+            nproc_per_node=num_gpus,
+            rdzv_backend="c10d",
+            max_restarts=0,
+        )
+        launch_agent(config, server.serve, [model_name, True, shard_directory])
+
+
+@app.command()
+def serve(
+        model_name: str,
+        sharded: bool = False,
+        shard_directory: Optional[Path] = None,
+):
+    server.serve(model_name, sharded, shard_directory)
+
+
+if __name__ == "__main__":
+    app()

server/bloom_inference/main.py

@@ -1,30 +0,0 @@
-import typer
-
-from pathlib import Path
-from torch.distributed.launcher import launch_agent, LaunchConfig
-from typing import Optional
-
-from bloom_inference.server import serve
-
-
-def main(
-    model_name: str,
-    num_gpus: int = 1,
-    shard_directory: Optional[Path] = None,
-):
-    if num_gpus == 1:
-        serve(model_name, False, shard_directory)
-
-    else:
-        config = LaunchConfig(
-            min_nodes=1,
-            max_nodes=1,
-            nproc_per_node=num_gpus,
-            rdzv_backend="c10d",
-            max_restarts=0,
-        )
-        launch_agent(config, serve, [model_name, True, shard_directory])
-
-
-if __name__ == "__main__":
-    typer.run(main)

server/bloom_inference/server.py

@@ -1,4 +1,6 @@
 import asyncio
+import os
+
 from grpc import aio
 
 from grpc_reflection.v1alpha import reflection
@@ -143,7 +145,3 @@ def serve(model_name, sharded, shard_directory):
         await server.wait_for_termination()
 
     asyncio.run(serve_inner(model_name, sharded, shard_directory))
-
-
-if __name__ == "__main__":
-    serve("bigscience/bloom-560m", True, Path("/tmp/models"))

server/bloom_inference/utils.py

@@ -2,6 +2,8 @@ import os
 import contextlib
 import torch
 import torch.distributed
+
+from datetime import timedelta
 from transformers.generation_logits_process import (
     LogitsProcessorList,
     TemperatureLogitsWarper,
@@ -79,6 +81,7 @@ def initialize_torch_distributed():
         backend=backend,
         world_size=world_size,
         rank=rank,
+        timeout=timedelta(seconds=60),
         init_method="tcp://localhost:6000",
     )
 

server/pyproject.toml

@@ -4,6 +4,9 @@ version = "0.1.0"
 description = "BLOOM Inference Python gRPC Server"
 authors = ["Olivier Dehaene <olivier@huggingface.co>"]
 
+[tool.poetry.scripts]
+bloom-inference-server = 'bloom_inference.cli:app'
+
 [tool.poetry.dependencies]
 python = "^3.9"
 protobuf = "^4.21.7"