[Feature][Frontend] Expose effective attention block size for DCP (#56538)

Co-authored-by: Bugen Zhao <i@bugenzhao.com>
Co-authored-by: OpenAI Codex <noreply@openai.com>
Signed-off-by: Julien Darve <jdarve@NVIDIA.com>
Signed-off-by: Bugen Zhao <i@bugenzhao.com>
This commit is contained in:
JulienDarve
2026-09-16 08:29:43 +00:00
committed by GitHub
co-authored by Bugen Zhao OpenAI Codex
parent 0d8173d153
commit f37c550bf6
18 changed files with 159 additions and 11 deletions
+1
View File
@@ -195,6 +195,7 @@ impl pb::control_server::Control for ControlServiceImpl {
max_batched_tokens: ready.max_num_batched_tokens,
max_loras: ready.max_loras,
rl_capabilities: Some(self.rl_capabilities()),
effective_attention_block_size: self.client().effective_attention_block_size(),
}))
}
+12 -1
View File
@@ -1911,8 +1911,16 @@ async fn control_abort_resolves_external_id_and_empty_is_noop() {
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
#[serial]
async fn control_reports_server_and_model_info() {
let mut ready = default_ready_response();
ready.effective_attention_block_size = Some(64);
let (generate_service, control_service, engine_health, _engine_task) =
setup_grpc_service(b"engine-grpc-info", default_stream_output_specs()).await;
setup_grpc_service_with_engine_script(
b"engine-grpc-info".to_vec(),
ready,
Arc::new(FakeTextBackend),
|_, _| boxed_test_future(async {}),
)
.await;
let (channel, server_task) = start_grpc_test_server(
generate_service,
control_service,
@@ -1928,6 +1936,7 @@ async fn control_reports_server_and_model_info() {
.expect("get server info")
.into_inner();
assert_eq!(server.engine_version, "test-vllm-version");
assert_eq!(server.effective_attention_block_size, Some(64));
assert_eq!(server.api_version, "vllm");
assert_eq!(server.instance_id, "test-instance");
assert_eq!(server.max_model_len, DEFAULT_MOCK_MAX_MODEL_LEN as u32);
@@ -2195,6 +2204,7 @@ async fn control_aggregates_multi_engine_capacity() {
ready_0.weight_transfer_backend = Some("nccl".to_string());
ready_0.enable_sleep_mode = true;
ready_0.supports_draft_weight_updates = true;
ready_0.effective_attention_block_size = Some(64);
let mut ready_1 = default_ready_response();
ready_1.max_model_len = 4_096;
@@ -2248,6 +2258,7 @@ async fn control_aggregates_multi_engine_capacity() {
.into_inner();
assert_eq!(server.max_model_len, 4_096);
assert_eq!(server.total_kv_blocks, 30);
assert!(server.effective_attention_block_size.is_none());
let rl = server.rl_capabilities.expect("RL capabilities");
assert!(!rl.weight_transfer_enabled);
assert!(rl.weight_transfer_backend.is_empty());