From 067d886b8dca93ac752ccbac1e4ffb8ad1832a4a Mon Sep 17 00:00:00 2001
From: Salman Paracha <salmanparacha@MacBook-Pro-288.local>
Date: Mon, 29 Sep 2025 17:43:04 -0700
Subject: [PATCH] fixes based on PR

---
 arch/envoy.template.yaml                    | 10 +++++-----
 arch/tools/cli/config_generator.py          |  2 +-
 arch/tools/cli/utils.py                     |  4 ++--
 crates/hermesllm/src/apis/openai.rs         |  8 ++++++++
 crates/hermesllm/src/clients/transformer.rs |  3 ++-
 5 files changed, 18 insertions(+), 9 deletions(-)

diff --git a/arch/envoy.template.yaml b/arch/envoy.template.yaml
index 5ee4c899..16b14343 100644
--- a/arch/envoy.template.yaml
+++ b/arch/envoy.template.yaml
@@ -140,7 +140,7 @@ static_resources:
                           route:
                             auto_host_rewrite: true
                             cluster: {{ llm_cluster_name }}
-                            timeout: 60s
+                            timeout: 300s
                       {% endfor %}
 
                       {% if agent_orchestrator %}
@@ -153,7 +153,7 @@ static_resources:
                           route:
                             auto_host_rewrite: true
                             cluster: {{ agent_orchestrator }}
-                            timeout: 60s
+                            timeout: 300s
                       {% endif %}
                 http_filters:
                   - name: envoy.filters.http.compressor
@@ -266,7 +266,7 @@ static_resources:
                           route:
                             auto_host_rewrite: true
                             cluster: {{ internal_cluster }}
-                            timeout: 60s
+                            timeout: 300s
                         {% endfor %}
 
                         {% for cluster_name, cluster in arch_clusters.items() %}
@@ -279,7 +279,7 @@ static_resources:
                           route:
                             auto_host_rewrite: true
                             cluster: {{ cluster_name }}
-                            timeout: 60s
+                            timeout: 300s
                         {% endfor %}
                 http_filters:
                   - name: envoy.filters.http.router
@@ -434,7 +434,7 @@ static_resources:
                           route:
                             auto_host_rewrite: true
                             cluster: {{ llm_cluster_name }}
-                            timeout: 60s
+                            timeout: 300s
                       {% endfor %}
                         - match:
                             prefix: "/"
diff --git a/arch/tools/cli/config_generator.py b/arch/tools/cli/config_generator.py
index 8f0dcefd..965bf040 100644
--- a/arch/tools/cli/config_generator.py
+++ b/arch/tools/cli/config_generator.py
@@ -242,7 +242,7 @@ def validate_and_render_schema():
     if llm_gateway_listener.get("address") == None:
         llm_gateway_listener["address"] = "127.0.0.1"
     if llm_gateway_listener.get("timeout") == None:
-        llm_gateway_listener["timeout"] = "10s"
+        llm_gateway_listener["timeout"] = "300s"
 
     use_agent_orchestrator = config_yaml.get("overrides", {}).get(
         "use_agent_orchestrator", False
diff --git a/arch/tools/cli/utils.py b/arch/tools/cli/utils.py
index d7adca60..c7d39d66 100644
--- a/arch/tools/cli/utils.py
+++ b/arch/tools/cli/utils.py
@@ -95,9 +95,9 @@ def find_config_file(path=".", file=None):
         return os.path.abspath(file)
     else:
         # If no file is provided, use the path and look for arch_config.yaml first, then config.yaml for convenience
-        arch_config_file = os.path.abspath(os.path.join(path, "arch_config.yaml"))
+        arch_config_file = os.path.abspath(os.path.join(path, "config.yaml"))
         if not os.path.exists(arch_config_file):
-            arch_config_file = os.path.abspath(os.path.join(path, "config.yaml"))
+            arch_config_file = os.path.abspath(os.path.join(path, "arch_config.yaml"))
         return arch_config_file
 
 
diff --git a/crates/hermesllm/src/apis/openai.rs b/crates/hermesllm/src/apis/openai.rs
index c4181768..d7d6ea70 100644
--- a/crates/hermesllm/src/apis/openai.rs
+++ b/crates/hermesllm/src/apis/openai.rs
@@ -117,6 +117,13 @@ impl ChatCompletionsRequest {
             self.max_tokens = None;
         }
     }
+
+    pub fn fix_temperature_if_gpt5(&mut self) {
+        let model = self.model.as_str();
+        if model.starts_with("gpt-5") {
+            self.temperature = Some(1.0);
+        }
+    }
 }
 
 // ============================================================================
@@ -599,6 +606,7 @@ impl TryFrom<&[u8]> for ChatCompletionsRequest {
        let mut req: ChatCompletionsRequest = serde_json::from_slice(bytes).map_err(OpenAIStreamError::from)?;
         // Use the centralized suppression logic
         req.suppress_max_tokens_if_o3();
+        req.fix_temperature_if_gpt5();
         Ok(req)
     }
 }
diff --git a/crates/hermesllm/src/clients/transformer.rs b/crates/hermesllm/src/clients/transformer.rs
index 33d2b8c1..0856c359 100644
--- a/crates/hermesllm/src/clients/transformer.rs
+++ b/crates/hermesllm/src/clients/transformer.rs
@@ -111,6 +111,7 @@ impl TryFrom<AnthropicMessagesRequest> for ChatCompletionsRequest {
             ..Default::default()
         };
         _chat_completions_req.suppress_max_tokens_if_o3();
+        _chat_completions_req.fix_temperature_if_gpt5();
         Ok(_chat_completions_req)
     }
 }
@@ -1014,7 +1015,7 @@ fn convert_content_delta(delta: MessagesContentDelta) -> Result<ChatCompletionsS
                 "unknown",
                 MessageDelta {
                     role: None,
-                    content: Some(format!("[Thinking: {}]", thinking)),
+                    content: Some(format!("thinking: {}", thinking)),
                     refusal: None,
                     function_call: None,
                     tool_calls: None,