apunkt/plano
1
0
Fork 0
mirror of https://github.com/katanemo/plano.git synced 2026-05-24 14:05:14 +02:00
Code Issues Projects Releases Packages Wiki Activity

fix: route Perplexity OpenAI paths without /v1

Browse source
This commit is contained in:
Spherrrical 2026-03-30 14:17:23 -07:00
parent 3a531ce22a
commit 1e2013bffc
3 changed files with 86 additions and 32 deletions
Download patch file Download diff file Expand all files Collapse all files

16
crates/brightstaff/src/handlers/llm/mod.rs
View file

@ -386,7 +386,7 @@ async fn parse_and_validate_request(
let temperature = client_request.get_temperature(); let temperature = client_request.get_temperature();
let is_streaming_request = client_request.is_streaming(); let is_streaming_request = client_request.is_streaming();
let alias_resolved_model = resolve_model_alias(&model_from_request, model_aliases); let alias_resolved_model = resolve_model_alias(&model_from_request, model_aliases);
let (provider_id, _) = get_provider_info(llm_providers, &alias_resolved_model).await; let (provider_id, _, _) = get_provider_info(llm_providers, &alias_resolved_model).await;
// Validate model exists in configuration // Validate model exists in configuration
if llm_providers if llm_providers
@ -741,7 +741,8 @@ async fn get_upstream_path(
resolved_model: &str, resolved_model: &str,
is_streaming: bool, is_streaming: bool,
) -> String { ) -> String {
let (provider_id, base_url_path_prefix) = get_provider_info(llm_providers, model_name).await; let (provider_id, base_url_path_prefix, use_unversioned_paths) =
get_provider_info(llm_providers, model_name).await;
let Some(client_api) = SupportedAPIsFromClient::from_endpoint(request_path) else { let Some(client_api) = SupportedAPIsFromClient::from_endpoint(request_path) else {
return request_path.to_string(); return request_path.to_string();
@ -753,6 +754,7 @@ async fn get_upstream_path(
resolved_model, resolved_model,
is_streaming, is_streaming,
base_url_path_prefix.as_deref(), base_url_path_prefix.as_deref(),
use_unversioned_paths,
) )
} }
@ -760,21 +762,23 @@ async fn get_upstream_path(
async fn get_provider_info( async fn get_provider_info(
llm_providers: &Arc<RwLock<LlmProviders>>, llm_providers: &Arc<RwLock<LlmProviders>>,
model_name: &str, model_name: &str,
) -> (hermesllm::ProviderId, Option<String>) { ) -> (hermesllm::ProviderId, Option<String>, bool) {
let providers_lock = llm_providers.read().await; let providers_lock = llm_providers.read().await;
if let Some(provider) = providers_lock.get(model_name) { if let Some(provider) = providers_lock.get(model_name) {
let provider_id = provider.provider_interface.to_provider_id(); let provider_id = provider.provider_interface.to_provider_id();
let prefix = provider.base_url_path_prefix.clone(); let prefix = provider.base_url_path_prefix.clone();
return (provider_id, prefix); let use_unversioned_paths = provider.name.starts_with("perplexity/");
return (provider_id, prefix, use_unversioned_paths);
} }
if let Some(provider) = providers_lock.default() { if let Some(provider) = providers_lock.default() {
let provider_id = provider.provider_interface.to_provider_id(); let provider_id = provider.provider_interface.to_provider_id();
let prefix = provider.base_url_path_prefix.clone(); let prefix = provider.base_url_path_prefix.clone();
(provider_id, prefix) let use_unversioned_paths = provider.name.starts_with("perplexity/");
(provider_id, prefix, use_unversioned_paths)
} else { } else {
warn!("No default provider found, falling back to OpenAI"); warn!("No default provider found, falling back to OpenAI");
(hermesllm::ProviderId::OpenAI, None) (hermesllm::ProviderId::OpenAI, None, false)
} }
} }

101
crates/hermesllm/src/clients/endpoints.rs
View file

@ -92,6 +92,7 @@ impl SupportedAPIsFromClient {
model_id: &str, model_id: &str,
is_streaming: bool, is_streaming: bool,
base_url_path_prefix: Option<&str>, base_url_path_prefix: Option<&str>,
use_unversioned_paths: bool,
) -> String { ) -> String {
// Helper function to build endpoint with optional prefix override // Helper function to build endpoint with optional prefix override
let build_endpoint = |provider_prefix: &str, suffix: &str| -> String { let build_endpoint = |provider_prefix: &str, suffix: &str| -> String {
@ -161,7 +162,13 @@ impl SupportedAPIsFromClient {
build_endpoint("/v1", endpoint_suffix) build_endpoint("/v1", endpoint_suffix)
} }
} }
_ => build_endpoint("/v1", endpoint_suffix), _ => {
if use_unversioned_paths {
build_endpoint("", endpoint_suffix)
} else {
build_endpoint("/v1", endpoint_suffix)
}
}
} }
}; };
@ -343,7 +350,8 @@ mod tests {
"/v1/chat/completions", "/v1/chat/completions",
"gpt-4", "gpt-4",
false, false,
None None,
false
), ),
"/v1/chat/completions" "/v1/chat/completions"
); );
@ -355,7 +363,8 @@ mod tests {
"/v1/chat/completions", "/v1/chat/completions",
"llama2", "llama2",
false, false,
None None,
false
), ),
"/openai/v1/chat/completions" "/openai/v1/chat/completions"
); );
@ -367,7 +376,8 @@ mod tests {
"/v1/chat/completions", "/v1/chat/completions",
"chatglm", "chatglm",
false, false,
None None,
false
), ),
"/api/paas/v4/chat/completions" "/api/paas/v4/chat/completions"
); );
@ -379,7 +389,8 @@ mod tests {
"/v1/chat/completions", "/v1/chat/completions",
"qwen-turbo", "qwen-turbo",
false, false,
None None,
false
), ),
"/compatible-mode/v1/chat/completions" "/compatible-mode/v1/chat/completions"
); );
@ -391,7 +402,8 @@ mod tests {
"/v1/chat/completions", "/v1/chat/completions",
"gpt-4", "gpt-4",
false, false,
None None,
false
), ),
"/openai/deployments/gpt-4/chat/completions?api-version=2025-01-01-preview" "/openai/deployments/gpt-4/chat/completions?api-version=2025-01-01-preview"
); );
@ -403,12 +415,30 @@ mod tests {
"/v1/chat/completions", "/v1/chat/completions",
"gemini-pro", "gemini-pro",
false, false,
None None,
false
), ),
"/v1beta/openai/chat/completions" "/v1beta/openai/chat/completions"
); );
} }
#[test]
fn test_target_endpoint_unversioned_paths() {
let api = SupportedAPIsFromClient::OpenAIChatCompletions(OpenAIApi::ChatCompletions);
assert_eq!(
api.target_endpoint_for_provider(
&ProviderId::OpenAI,
"/v1/chat/completions",
"sonar-pro",
false,
None,
true
),
"/chat/completions"
);
}
#[test] #[test]
fn test_target_endpoint_with_base_url_prefix() { fn test_target_endpoint_with_base_url_prefix() {
let api = SupportedAPIsFromClient::OpenAIChatCompletions(OpenAIApi::ChatCompletions); let api = SupportedAPIsFromClient::OpenAIChatCompletions(OpenAIApi::ChatCompletions);
@ -420,7 +450,8 @@ mod tests {
"/v1/chat/completions", "/v1/chat/completions",
"chatglm", "chatglm",
false, false,
Some("/api/coding/paas/v4") Some("/api/coding/paas/v4"),
false
), ),
"/api/coding/paas/v4/chat/completions" "/api/coding/paas/v4/chat/completions"
); );
@ -432,7 +463,8 @@ mod tests {
"/v1/chat/completions", "/v1/chat/completions",
"chatglm", "chatglm",
false, false,
Some("api/coding/paas/v4") Some("api/coding/paas/v4"),
false
), ),
"/api/coding/paas/v4/chat/completions" "/api/coding/paas/v4/chat/completions"
); );
@ -444,7 +476,8 @@ mod tests {
"/v1/chat/completions", "/v1/chat/completions",
"chatglm", "chatglm",
false, false,
Some("/api/coding/paas/v4/") Some("/api/coding/paas/v4/"),
false
), ),
"/api/coding/paas/v4/chat/completions" "/api/coding/paas/v4/chat/completions"
); );
@ -456,7 +489,8 @@ mod tests {
"/v1/chat/completions", "/v1/chat/completions",
"gpt-4", "gpt-4",
false, false,
Some("/custom/api/v2") Some("/custom/api/v2"),
false
), ),
"/custom/api/v2/chat/completions" "/custom/api/v2/chat/completions"
); );
@ -468,7 +502,8 @@ mod tests {
"/v1/chat/completions", "/v1/chat/completions",
"llama2", "llama2",
false, false,
Some("/api/v2") Some("/api/v2"),
false
), ),
"/api/v2/v1/chat/completions" "/api/v2/v1/chat/completions"
); );
@ -485,7 +520,8 @@ mod tests {
"/v1/chat/completions", "/v1/chat/completions",
"chatglm", "chatglm",
false, false,
Some("/") Some("/"),
false
), ),
"/api/paas/v4/chat/completions" "/api/paas/v4/chat/completions"
); );
@ -497,7 +533,8 @@ mod tests {
"/v1/chat/completions", "/v1/chat/completions",
"chatglm", "chatglm",
false, false,
None None,
false
), ),
"/api/paas/v4/chat/completions" "/api/paas/v4/chat/completions"
); );
@ -514,7 +551,8 @@ mod tests {
"/v1/messages", "/v1/messages",
"us.amazon.nova-pro-v1:0", "us.amazon.nova-pro-v1:0",
false, false,
None None,
false
), ),
"/model/us.amazon.nova-pro-v1:0/converse" "/model/us.amazon.nova-pro-v1:0/converse"
); );
@ -526,7 +564,8 @@ mod tests {
"/v1/messages", "/v1/messages",
"us.amazon.nova-pro-v1:0", "us.amazon.nova-pro-v1:0",
true, true,
None None,
false
), ),
"/model/us.amazon.nova-pro-v1:0/converse-stream" "/model/us.amazon.nova-pro-v1:0/converse-stream"
); );
@ -538,7 +577,8 @@ mod tests {
"/v1/messages", "/v1/messages",
"us.amazon.nova-pro-v1:0", "us.amazon.nova-pro-v1:0",
false, false,
Some("/custom/path") Some("/custom/path"),
false
), ),
"/custom/path/model/us.amazon.nova-pro-v1:0/converse" "/custom/path/model/us.amazon.nova-pro-v1:0/converse"
); );
@ -550,7 +590,8 @@ mod tests {
"/v1/messages", "/v1/messages",
"us.amazon.nova-pro-v1:0", "us.amazon.nova-pro-v1:0",
true, true,
Some("/custom/path") Some("/custom/path"),
false
), ),
"/custom/path/model/us.amazon.nova-pro-v1:0/converse-stream" "/custom/path/model/us.amazon.nova-pro-v1:0/converse-stream"
); );
@ -567,7 +608,8 @@ mod tests {
"/v1/messages", "/v1/messages",
"claude-3-opus", "claude-3-opus",
false, false,
None None,
false
), ),
"/v1/messages" "/v1/messages"
); );
@ -579,7 +621,8 @@ mod tests {
"/v1/messages", "/v1/messages",
"claude-3-opus", "claude-3-opus",
false, false,
Some("/api/v2") Some("/api/v2"),
false
), ),
"/api/v2/messages" "/api/v2/messages"
); );
@ -596,7 +639,8 @@ mod tests {
"/custom/path", "/custom/path",
"llama2", "llama2",
false, false,
None None,
false
), ),
"/v1/chat/completions" "/v1/chat/completions"
); );
@ -608,7 +652,8 @@ mod tests {
"/custom/path", "/custom/path",
"chatglm", "chatglm",
false, false,
None None,
false
), ),
"/v1/chat/completions" "/v1/chat/completions"
); );
@ -620,7 +665,8 @@ mod tests {
"/custom/path", "/custom/path",
"chatglm", "chatglm",
false, false,
Some("/api/v2") Some("/api/v2"),
false
), ),
"/api/v2/chat/completions" "/api/v2/chat/completions"
); );
@ -637,7 +683,8 @@ mod tests {
"/v1/chat/completions", "/v1/chat/completions",
"gpt-4-deployment", "gpt-4-deployment",
false, false,
None None,
false
), ),
"/openai/deployments/gpt-4-deployment/chat/completions?api-version=2025-01-01-preview" "/openai/deployments/gpt-4-deployment/chat/completions?api-version=2025-01-01-preview"
); );
@ -649,7 +696,8 @@ mod tests {
"/v1/chat/completions", "/v1/chat/completions",
"gpt-4-deployment", "gpt-4-deployment",
false, false,
Some("/custom/azure/path") Some("/custom/azure/path"),
false
), ),
"/custom/azure/path/gpt-4-deployment/chat/completions?api-version=2025-01-01-preview" "/custom/azure/path/gpt-4-deployment/chat/completions?api-version=2025-01-01-preview"
); );
@ -664,7 +712,8 @@ mod tests {
"/v1/responses", "/v1/responses",
"grok-4-1-fast-reasoning", "grok-4-1-fast-reasoning",
false, false,
None None,
false
), ),
"/v1/responses" "/v1/responses"
); );

1
crates/llm_gateway/src/stream_context.rs
View file

@ -122,6 +122,7 @@ impl StreamContext {
.unwrap_or(&"".to_string()), .unwrap_or(&"".to_string()),
self.streaming_response, self.streaming_response,
self.llm_provider().base_url_path_prefix.as_deref(), self.llm_provider().base_url_path_prefix.as_deref(),
self.llm_provider().name.starts_with("perplexity/"),
); );
if target_endpoint != request_path { if target_endpoint != request_path {
self.set_http_request_header(":path", Some(&target_endpoint)); self.set_http_request_header(":path", Some(&target_endpoint));