|
|
@@ -57,6 +57,11 @@ fn trec() -> Recipe {
|
|
|
}
|
|
|
|
|
|
fn run(recipe: &Recipe, node: &Node, head: bool) -> String {
|
|
|
+ let tp = if recipe.required_nodes == Some(recipe.tensor_parallel_size) {
|
|
|
+ 2
|
|
|
+ } else {
|
|
|
+ recipe.tensor_parallel_size
|
|
|
+ };
|
|
|
build_docker_command(
|
|
|
&tcls(),
|
|
|
&tnccl(),
|
|
|
@@ -65,6 +70,7 @@ fn run(recipe: &Recipe, node: &Node, head: bool) -> String {
|
|
|
node,
|
|
|
head,
|
|
|
2,
|
|
|
+ tp,
|
|
|
2,
|
|
|
"/home/user/.cache/huggingface",
|
|
|
false,
|
|
|
@@ -74,6 +80,11 @@ fn run(recipe: &Recipe, node: &Node, head: bool) -> String {
|
|
|
}
|
|
|
|
|
|
fn run_keep(recipe: &Recipe, node: &Node, head: bool) -> String {
|
|
|
+ let tp = if recipe.required_nodes == Some(recipe.tensor_parallel_size) {
|
|
|
+ 2
|
|
|
+ } else {
|
|
|
+ recipe.tensor_parallel_size
|
|
|
+ };
|
|
|
build_docker_command(
|
|
|
&tcls(),
|
|
|
&tnccl(),
|
|
|
@@ -82,6 +93,7 @@ fn run_keep(recipe: &Recipe, node: &Node, head: bool) -> String {
|
|
|
node,
|
|
|
head,
|
|
|
2,
|
|
|
+ tp,
|
|
|
2,
|
|
|
"/home/user/.cache/huggingface",
|
|
|
true,
|
|
|
@@ -237,6 +249,11 @@ fn sglang_recipe() -> Recipe {
|
|
|
}
|
|
|
|
|
|
fn srun(recipe: &Recipe, node: &Node, head: bool) -> String {
|
|
|
+ let tp = if recipe.required_nodes == Some(recipe.tensor_parallel_size) {
|
|
|
+ 4
|
|
|
+ } else {
|
|
|
+ recipe.tensor_parallel_size
|
|
|
+ };
|
|
|
build_docker_command(
|
|
|
&tcls(),
|
|
|
&tnccl(),
|
|
|
@@ -245,6 +262,7 @@ fn srun(recipe: &Recipe, node: &Node, head: bool) -> String {
|
|
|
node,
|
|
|
head,
|
|
|
4,
|
|
|
+ tp,
|
|
|
4,
|
|
|
"/home/user/.cache/huggingface",
|
|
|
false,
|
|
|
@@ -372,14 +390,21 @@ fn recipe_env_overrides_cluster_defaults() {
|
|
|
fn custom_init_port() {
|
|
|
// Non-default init_port should propagate to vLLM master-port and
|
|
|
// SGLang dist-init-addr.
|
|
|
+ let r = trec();
|
|
|
+ let tp = if r.required_nodes == Some(r.tensor_parallel_size) {
|
|
|
+ 2
|
|
|
+ } else {
|
|
|
+ r.tensor_parallel_size
|
|
|
+ };
|
|
|
let cmd = build_docker_command(
|
|
|
&tcls(),
|
|
|
&tnccl(),
|
|
|
- &trec(),
|
|
|
+ &r,
|
|
|
"r",
|
|
|
&tcls().nodes[0],
|
|
|
true,
|
|
|
2,
|
|
|
+ tp,
|
|
|
2,
|
|
|
"/home/user/.cache/huggingface",
|
|
|
false,
|