huggingface/text-generation-inference
5
0
Fork 0
mirror of https://github.com/huggingface/text-generation-inference.git synced 2025-09-10 20:04:52 +00:00
Code Issues Packages Projects Releases Wiki Activity

use relaxed

Browse Source
This commit is contained in:
OlivierDehaene 2023-10-20 12:41:21 +02:00
parent 3756a5f1e2
commit df419a21e0
2 changed files with 6 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
router/src/health.rs
View File

@ -23,7 +23,7 @@ impl Health {
}
pub(crate) async fn check(&mut self) -> bool {
if self.generation_health.load(Ordering::SeqCst) {
if self.generation_health.load(Ordering::Relaxed) {
// Generation is healthy, we only check that the shards are answering gRPC calls
self.client.health().await.is_ok()
} else {
@ -61,7 +61,7 @@ impl Health {
// Skips the queue
let value = self.client.prefill(batch).await.is_ok();
// Update generation health
self.generation_health.store(value, Ordering::SeqCst);
self.generation_health.store(value, Ordering::Relaxed);
value
}
}

8
router/src/infer.rs
View File

@ -385,7 +385,7 @@ async fn prefill(
match client.prefill(batch).await {
Ok((generations, next_batch)) => {
// Update health
generation_health.store(true, Ordering::SeqCst);
generation_health.store(true, Ordering::Relaxed);
// Send generated tokens and filter stopped entries
filter_send_generations(generations, entries);
@ -399,7 +399,7 @@ async fn prefill(
// If we have an error, we discard the whole batch
Err(err) => {
// Update health
generation_health.store(false, Ordering::SeqCst);
generation_health.store(false, Ordering::Relaxed);
let _ = client.clear_cache(Some(batch_id)).await;
send_errors(err, entries);
metrics::increment_counter!("tgi_batch_inference_failure", "method" => "prefill");
@ -424,7 +424,7 @@ async fn decode(
match client.decode(batches).await {
Ok((generations, next_batch)) => {
// Update health
generation_health.store(true, Ordering::SeqCst);
generation_health.store(true, Ordering::Relaxed);
// Send generated tokens and filter stopped entries
filter_send_generations(generations, entries);
@ -437,7 +437,7 @@ async fn decode(
}
// If we have an error, we discard the whole batch
Err(err) => {
generation_health.store(false, Ordering::SeqCst);
generation_health.store(false, Ordering::Relaxed);
for id in batch_ids {
let _ = client.clear_cache(Some(id)).await;
}