use provider/model to identify models

This commit is contained in:
Adil Hafeez 2025-07-08 13:47:56 -07:00
parent 5f18fee089
commit bcd7f9be45
No known key found for this signature in database
GPG key ID: 9B18EF7691369645
7 changed files with 33 additions and 68 deletions

View file

@ -104,7 +104,7 @@ pub async fn chat_completions(
debug!("usage preferences: {:?}", usage_preferences);
let mut selected_llm = match router_service
let mut determined_route = match router_service
.determine_route(
&chat_completion_request.messages,
trace_parent.clone(),
@ -121,14 +121,14 @@ pub async fn chat_completions(
}
};
if selected_llm.is_none() {
if determined_route.is_none() {
debug!("No LLM model selected, using default from request");
selected_llm = Some(chat_completion_request.model.clone());
determined_route = Some(chat_completion_request.model.clone());
}
info!(
"sending request to llm provider: {} with llm model: {:?}",
llm_provider_endpoint, selected_llm
llm_provider_endpoint, determined_route
);
if let Some(trace_parent) = trace_parent {
@ -138,10 +138,10 @@ pub async fn chat_completions(
);
}
if let Some(selected_llm) = selected_llm {
if let Some(selected_route) = determined_route {
request_headers.insert(
ARCH_PROVIDER_HINT_HEADER,
header::HeaderValue::from_str(&selected_llm).unwrap(),
header::HeaderValue::from_str(&selected_route).unwrap(),
);
}

View file

@ -45,8 +45,8 @@ async fn main() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
let bind_address = env::var("BIND_ADDRESS").unwrap_or_else(|_| BIND_ADDRESS.to_string());
// loading arch_config.yaml file
let arch_config_path =
env::var("ARCH_CONFIG_PATH").unwrap_or_else(|_| "./arch_config.yaml".to_string());
let arch_config_path = env::var("ARCH_CONFIG_PATH_RENDERED")
.unwrap_or_else(|_| "./arch_config_rendered.yaml".to_string());
info!("Loading arch_config.yaml from {}", arch_config_path);
let config_contents =

View file

@ -1,4 +1,4 @@
use std::{collections::HashMap, sync::Arc};
use std::sync::Arc;
use common::{
configuration::{LlmProvider, LlmRoute, ModelUsagePreference},
@ -19,7 +19,6 @@ pub struct RouterService {
router_model: Arc<dyn RouterModel>,
routing_provider_name: String,
llm_usage_defined: bool,
llm_provider_map: HashMap<String, LlmProvider>,
}
#[derive(Debug, Error)]
@ -57,18 +56,12 @@ impl RouterService {
router_model_v1::MAX_TOKEN_LEN,
));
let llm_provider_map: HashMap<String, LlmProvider> = providers
.into_iter()
.map(|provider| (provider.name.clone(), provider))
.collect();
RouterService {
router_url,
client: reqwest::Client::new(),
router_model,
routing_provider_name,
llm_usage_defined: !providers_with_usage.is_empty(),
llm_provider_map,
}
}
@ -155,40 +148,15 @@ impl RouterService {
if let Some(ContentType::Text(content)) =
&chat_completion_response.choices[0].message.content
{
let mut selected_model: Option<String> = None;
if let Some(selected_llm_name) = self.router_model.parse_response(content)? {
if selected_llm_name != "other" {
if let Some(usage_preferences) = usage_preferences {
for usage in usage_preferences {
if usage.name == selected_llm_name {
selected_model = Some(usage.model);
break;
}
}
if selected_model.is_none() {
warn!(
"Selected LLM model not found in usage preferences: {}",
selected_llm_name
);
}
} else if let Some(provider) = self.llm_provider_map.get(&selected_llm_name) {
selected_model = provider.model.clone();
} else {
warn!(
"Selected LLM model not found in provider map: {}",
selected_llm_name
);
}
}
}
let route_name = self.router_model.parse_response(content)?;
info!(
"router response: {}, selected_model: {:?}, response time: {}ms",
content.replace("\n", "\\n"),
selected_model,
route_name,
router_response_time.as_millis()
);
Ok(selected_model)
Ok(route_name)
} else {
Ok(None)
}