Skip to content

planner inference timeout과 출력 예산을 모델 프로필 설정으로 이동 #57

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
Createyouracccount merged 1 commit into from
Sep 6, 2026
Merged

planner inference timeout과 출력 예산을 모델 프로필 설정으로 이동 #57

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
90 changes: 81 additions & 9 deletions crates/xgeny-cli/src/composition.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -69,15 +69,14 @@ use crate::material_catalog::{
WORKSPACE_READ_MATERIAL_PROVIDER_ID, WORKSPACE_READ_RECIPE_DOMAIN,
WORKSPACE_READ_RECIPE_FORMAT_VERSION, WorkspaceReadMaterialProvider, WorkspaceReadMaterializer,
};
use crate::model_profile::InferenceLimits;
use crate::run_layout::{RunLayout, discover_state_root, generate_run_id};

const WORKSPACE_ID: &str = "primary";
const WORKSPACE_IDENTITY_PROFILE: &str = "xgeny.fs.workspace-root-identity.v1";
const DEFAULT_PLANNER_ID: &str = "xgeny.cli.openai";
const MAX_GOAL_BYTES: usize = 16 * 1024;
const MAX_TICKS: u32 = 1_024;
const MAX_OUTPUT_TOKENS: u32 = 1_024;
const MODEL_TIMEOUT: Duration = Duration::from_secs(60);
const MODEL_CHECK_TIMEOUT: Duration = Duration::from_secs(10);
const LOCAL_EXECUTION_PROFILE_DOMAIN: &str = "xgeny.cli.local-execution-profile/v1";
const WORKSPACE_DISCOVERY_PROFILE_DOMAIN: &str =
Expand Down Expand Up @@ -111,6 +110,7 @@ pub struct LocalRunRequest {
pub model: String,
pub tokenizer: String,
pub credential: Option<BearerCredential>,
pub inference_limits: InferenceLimits,
pub allow_files: Vec<String>,
pub allow_dirs: Vec<String>,
pub allow_executables: Vec<String>,
Expand Down Expand Up @@ -139,6 +139,7 @@ impl LocalRunRequest {
model,
tokenizer,
credential: None,
inference_limits: InferenceLimits::default(),
allow_files,
allow_dirs: Vec::new(),
allow_executables: Vec::new(),
Expand All @@ -158,6 +159,7 @@ pub struct LocalResumeRequest {
pub workspace: Option<PathBuf>,
pub base_url: Option<String>,
pub credential: Option<BearerCredential>,
pub inference_limits: InferenceLimits,
pub allow_files: Vec<String>,
pub allow_dirs: Vec<String>,
pub allow_executables: Vec<String>,
Expand Down Expand Up @@ -224,6 +226,7 @@ pub struct ModelCheckRequest {
pub model: String,
pub tokenizer: String,
pub credential: Option<BearerCredential>,
pub inference_limits: InferenceLimits,
}

/// Stable, redacted failure classes returned by `xgeny model check`.
Expand Down Expand Up @@ -331,6 +334,7 @@ pub fn check_openai_model(request: ModelCheckRequest) -> Result<(), ModelCheckEr
model,
tokenizer,
credential,
inference_limits: _,
} = request;
let config = OpenAiPlannerConfig::new(&base_url, DEFAULT_PLANNER_ID, &model, &tokenizer)
.and_then(|config| config.with_timeout(MODEL_CHECK_TIMEOUT))
Expand All @@ -352,6 +356,7 @@ pub fn list_openai_models(request: ModelCheckRequest) -> Result<Vec<String>, Mod
model,
tokenizer,
credential,
inference_limits: _,
} = request;
let config = OpenAiPlannerConfig::new(&base_url, DEFAULT_PLANNER_ID, &model, &tokenizer)
.and_then(|config| config.with_timeout(MODEL_CHECK_TIMEOUT))
Expand All @@ -373,8 +378,9 @@ pub fn check_openai_compatibility(request: ModelCheckRequest) -> Result<(), Mode
model,
tokenizer,
credential,
inference_limits,
} = request;
let config = compatibility_probe_config(&base_url, &model, &tokenizer)?;
let config = compatibility_probe_config(&base_url, &model, &tokenizer, inference_limits)?;
OpenAiCompatibilityChecker::new(config, credential)
.map_err(map_model_check_config)?
.check()
Expand All @@ -391,10 +397,11 @@ fn compatibility_probe_config(
base_url: &str,
model: &str,
tokenizer: &str,
limits: InferenceLimits,
) -> Result<OpenAiPlannerConfig, ModelCheckError> {
OpenAiPlannerConfig::new(base_url, DEFAULT_PLANNER_ID, model, tokenizer)
.and_then(|config| config.with_max_output_tokens(MAX_OUTPUT_TOKENS))
.and_then(|config| config.with_timeout(MODEL_TIMEOUT))
.and_then(|config| config.with_max_output_tokens(limits.max_output_tokens()))
.and_then(|config| config.with_timeout(limits.timeout()))
.map_err(map_model_check_config)
}

Expand Down Expand Up @@ -709,6 +716,7 @@ where
&request.planner_id,
&request.model,
&request.tokenizer,
request.inference_limits,
planning_constraints_required,
)?;
let local_execution_profile_digest =
Expand Down Expand Up @@ -963,6 +971,7 @@ where
manifest.planner_id(),
manifest.model(),
manifest.tokenizer(),
request.inference_limits,
catalog.workspace_discovery() || process.is_some(),
)?;
if manifest.request_profile_digest() != config.request_profile_digest() {
Expand Down Expand Up @@ -1508,11 +1517,12 @@ fn planner_config(
planner_id: &str,
model: &str,
tokenizer: &str,
limits: InferenceLimits,
planning_constraints_required: bool,
) -> Result<OpenAiPlannerConfig, PublicRunError> {
let config = OpenAiPlannerConfig::new(base_url, planner_id, model, tokenizer)
.and_then(|config| config.with_max_output_tokens(MAX_OUTPUT_TOKENS))
.and_then(|config| config.with_timeout(MODEL_TIMEOUT))
.and_then(|config| config.with_max_output_tokens(limits.max_output_tokens()))
.and_then(|config| config.with_timeout(limits.timeout()))
.map_err(map_provider_config)?;
if planning_constraints_required {
config
Expand Down Expand Up @@ -2471,15 +2481,75 @@ mod tests {
assert_ne!(error.code(), ModelCheckError::InvalidResponse.code());
}

#[test]
fn inference_limits_default_to_the_measured_local_budget() {
let limits = InferenceLimits::default();
assert_eq!(limits.timeout(), Duration::from_secs(300));
assert_eq!(limits.max_output_tokens(), 1_024);
}

#[test]
fn inference_limits_reject_values_outside_the_provider_bounds() {
assert!(InferenceLimits::new(Duration::from_secs(0), 1_024).is_err());
assert!(InferenceLimits::new(Duration::from_secs(3_601), 1_024).is_err());
assert!(InferenceLimits::new(Duration::from_secs(300), 0).is_err());
assert!(InferenceLimits::new(Duration::from_secs(300), 65_537).is_err());
assert!(InferenceLimits::new(Duration::from_secs(1), 1).is_ok());
assert!(InferenceLimits::new(Duration::from_secs(3_600), 65_536).is_ok());
}

#[test]
fn probe_and_planner_follow_the_profile_limits_and_stay_digest_identical() {
let limits = InferenceLimits::new(Duration::from_secs(300), 2_048).unwrap();
let probe =
compatibility_probe_config("http://127.0.0.1:1/v1", "model", "tokenizer", limits)
.unwrap();
let production = planner_config(
"http://127.0.0.1:1/v1",
DEFAULT_PLANNER_ID,
"model",
"tokenizer",
limits,
false,
)
.unwrap();
assert_eq!(
probe.request_profile_digest(),
production.request_profile_digest()
);

let other = InferenceLimits::new(Duration::from_secs(60), 2_048).unwrap();
let shorter = planner_config(
"http://127.0.0.1:1/v1",
DEFAULT_PLANNER_ID,
"model",
"tokenizer",
other,
false,
)
.unwrap();
assert_ne!(
production.request_profile_digest(),
shorter.request_profile_digest(),
"timeout is a committed request-profile input, so a different limit is a different profile"
);
}

#[test]
fn compatibility_probe_config_matches_the_production_planner_profile() {
let probe = compatibility_probe_config("http://127.0.0.1:1/v1", "model", "tokenizer")
.expect("probe config should validate");
let probe = compatibility_probe_config(
"http://127.0.0.1:1/v1",
"model",
"tokenizer",
InferenceLimits::default(),
)
.expect("probe config should validate");
let production = planner_config(
"http://127.0.0.1:1/v1",
DEFAULT_PLANNER_ID,
"model",
"tokenizer",
InferenceLimits::default(),
false,
)
.expect("production config should validate");
Expand Down Expand Up @@ -2586,6 +2656,7 @@ mod tests {
model: "model".to_owned(),
tokenizer: "tokenizer".to_owned(),
credential: None,
inference_limits: InferenceLimits::default(),
allow_files: vec!["README.md".to_owned()],
allow_dirs: Vec::new(),
allow_executables: Vec::new(),
Expand Down Expand Up @@ -2639,6 +2710,7 @@ mod tests {
model: "model".to_owned(),
tokenizer: "tokenizer".to_owned(),
credential: None,
inference_limits: InferenceLimits::default(),
allow_files: Vec::new(),
allow_dirs: vec![".".to_owned()],
allow_executables: vec![specification],
Expand Down
Loading