[Enhancement] Core changes to add support for custom proxy models

sygra/config/models.yaml

@@ -136,18 +136,47 @@ gemini_2_5_pro:
 # Bedrock model
 bedrock_model:
   model_type: bedrock
-  model: anthropic.claude-sonnet-4-5-20250929-v1:0
+  model: us.anthropic.claude-sonnet-4-5-20250929-v1:0
   # aws_access_key_id, aws_secret_access_key, aws_region_name should be defined at .env file as SYGRA_BEDROCK_MODEL_AWS_ACCESS_KEY_ID, SYGRA_BEDROCK_MODEL_AWS_SECRET_ACCESS_KEY, SYGRA_BEDROCK_MODEL_AWS_REGION_NAME
   parameters:
     max_tokens: 5000
     temperature: 0.5
 
 
-#gpt 5
+openai_gpt41:
+  model_type: azure_openai
+  model: gpt-4.1
+  api_version: 2024-02-15-preview
+  parameters:
+    max_tokens: 5000
+    temperature: 0.1
+
 gpt-5:
   model_type: azure_openai
   model: gpt-5
-  # URL and auth_token should be defined at .env file as SYGRA_GPT-4O_URL and SYGRA_GPT-4O_TOKEN
-  api_version: 2024-08-01-preview
+  # URL and auth_token should be defined at .env file as SYGRA_GPT-5_URL and SYGRA_GPT-5_TOKEN
+  api_version: 2024-02-15-preview
   parameters:
     max_completion_tokens: 5000
+    temperature: 1
+
+claude_large:
+  model_type: claude_proxy
+  client_type: http
+  backend: proxy
+  additional_params:
+    thinking:
+      type: disabled
+    anthropic_beta:
+      - computer-use-2025-01-24
+  parameters:
+    maxTokens: 2500
+    temperature: 0
+
+gemini_2.5_proxy:
+  model_type: gemini_proxy
+  client_type: http
+  backend: proxy
+  parameters:
+    maxOutputTokens: 2500
+    temperature: 0

sygra/core/dataset/dataset_processor.py

@@ -441,10 +441,10 @@ async def _process_and_store_results(self):
                     metadata = {"output_file": output_file}
                     processor = utils.get_func_from_str(post_processor)
                     processor_name = post_processor.split(".")[-1]
-                    processed_output_data = processor().process(output_data, metadata)
+                    processed_output_data = await processor().process(output_data, metadata)
                     new_output_file = output_file[: output_file.rfind("/") + 1] + output_file[
                         output_file.rfind("/") + 1 :
-                    ].replace("output_", processor_name + "_", 1)
+                    ].replace("output", processor_name, 1)
                     with open(new_output_file, "w") as f:
                         logger.info(f"Writing metrics output to file {new_output_file}")
                         json.dump(processed_output_data, f, indent=4)

sygra/core/graph/graph_postprocessor.py

@@ -8,6 +8,6 @@ class GraphPostProcessor(ABC):
     """
 
     @abstractmethod
-    def process(self, data: list, metadata: dict) -> list:
+    async def process(self, data: list, metadata: dict) -> list:
         # implement post processing logic with whole data, return the final data list
         pass

sygra/core/graph/langgraph/langchain_callback.py

@@ -59,7 +59,7 @@ def calculate_cost(model_name: str, prompt_tokens: int, completion_tokens: int)
                 return _get_anthropic_claude_token_cost(
                     prompt_tokens, completion_tokens, model_name
                 )
-            except (KeyError, ValueError):
+            except (KeyError, ValueError, IndexError):
                 # Claude model but not in Bedrock pricing table
                 pass
 

sygra/core/graph/nodes/llm_node.py

@@ -240,13 +240,13 @@ def _generate_prompt_tmpl_from_msg(
         for message in chat_frmt_messages:
             contents = message["content"]
             # if it's a normal text conversation then no need to expand
-            if isinstance(contents, str):
+            if isinstance(contents, str) or message.get("role", "") == "tool":
                 continue
             expanded_contents = []
             for item in contents:
-                if item["type"] == "image_url":
+                if item.get("type") and item.get("type") == "image_url":
                     expanded_contents.extend(expand_image_item(item, state))
-                elif item["type"] == "audio_url":
+                elif item.get("type") and item.get("type") == "audio_url":
                     expanded_contents.extend(expand_audio_item(item, state))
                 else:
                     expanded_contents.append(item)

sygra/core/models/client/client_factory.py

@@ -17,6 +17,7 @@
 class ModelType(Enum):
     """Enum representing the supported model types for client creation."""
 
+    # model type or server type
     VLLM = "vllm"
     OPENAI = "openai"
     AZURE_OPENAI = "azure_openai"
@@ -27,8 +28,14 @@ class ModelType(Enum):
     TRITON = "triton"
 
 
+class ClientType(Enum):
+    # client type
+    HTTP = "http"
+
+
 # Define which model types do not require a AUTH_TOKEN
 NO_AUTH_TOKEN_NEEDED_MODEL_TYPES = [ModelType.OLLAMA.value]
+NO_AUTH_TOKEN_NEEDED_CLIENT_TYPES = [ClientType.HTTP.value]
 
 
 class ClientFactory:
@@ -65,22 +72,49 @@ def create_client(
         # Validate model_type is present
         utils.validate_required_keys(["model_type"], model_config, "model")
 
+        # read model type or server type
         model_type = model_config["model_type"].lower()
 
+        # read client type
+        client_type = (
+            model_config["client_type"].lower() if model_config.get("client_type") else None
+        )
+
         # Validate if url is present
         if url is None:
             logger.error("URL is required for client creation.")
             raise ValueError("URL is required for client creation.")
 
         # Validate if auth_token is present
-        if auth_token is None and model_type not in NO_AUTH_TOKEN_NEEDED_MODEL_TYPES:
+        if auth_token is None and (
+            model_type not in NO_AUTH_TOKEN_NEEDED_MODEL_TYPES
+            and client_type not in NO_AUTH_TOKEN_NEEDED_CLIENT_TYPES
+        ):
             logger.error("Auth token/API key is required for client creation.")
             raise ValueError("Auth token/API key is required for client creation.")
 
+        # TODO: add backend condition
+
         # Validate that the model_type is supported
-        supported_types = [type_enum.value for type_enum in ModelType]
-        if model_type not in supported_types:
-            supported_types_str = ", ".join(supported_types)
+        supported_model_types = [type_enum.value for type_enum in ModelType]
+        if client_type is not None:
+            supported_client_types = [type_enum.value for type_enum in ClientType]
+            if client_type not in supported_client_types:
+                supported_types_str = ", ".join(supported_client_types)
+                logger.error(
+                    f"Unsupported client type: {client_type}. Supported types: {supported_types_str}"
+                )
+                raise ValueError(
+                    f"Unsupported client type: {client_type}. Must be one of: {supported_types_str}"
+                )
+
+            if client_type == ClientType.HTTP.value:
+                return cls._create_http_client(model_config, url, auth_token)
+            else:
+                logger.error(f"Unsupported client type: {client_type}")
+            # Add more client type which can be common
+        elif model_type not in supported_model_types:
+            supported_types_str = ", ".join(supported_model_types)
             logger.error(
                 f"Unsupported model type: {model_type}. Supported types: {supported_types_str}"
             )
@@ -89,8 +123,9 @@ def create_client(
             )
 
         # Create client based on model type
-        if model_type == ModelType.VLLM.value or model_type == ModelType.OPENAI.value:
+        elif model_type == ModelType.VLLM.value or model_type == ModelType.OPENAI.value:
             # Initialize the client with default chat_completions_api
+            # for VLLM server type, client type is openai ONLY
             return cls._create_openai_client(
                 model_config,
                 url,
@@ -99,6 +134,7 @@ def create_client(
                 not model_config.get("completions_api", False),
             )
         elif model_type == ModelType.AZURE_OPENAI.value:
+            # for AZURE_OPENAI server type, client type is openai_azure ONLY
             return cls._create_openai_azure_client(
                 model_config,
                 url,
@@ -111,10 +147,13 @@ def create_client(
             or model_type == ModelType.TGI.value
             or model_type == ModelType.TRITON.value
         ):
+            # for AZURE/TGI/TRITON server type, client type is http ONLY
             return cls._create_http_client(model_config, url, auth_token)
         elif model_type == ModelType.MISTRALAI.value:
+            # for Mistral server type, client type is mistral ONLY
             return cls._create_mistral_client(model_config, url, auth_token, async_client)
         elif model_type == ModelType.OLLAMA.value:
+            # for OLLAMA server type, client type is ollama
             return cls._create_ollama_client(
                 model_config,
                 url,
@@ -306,6 +345,12 @@ def _create_http_client(
             auth_token = auth_token.replace("Bearer ", "")
             headers["Authorization"] = f"Bearer {auth_token}"
 
+        if model_config.get("extra_headers"):
+            extra_headers = model_config.get("extra_headers")
+            if extra_headers:
+                for k, v in extra_headers.items():
+                    headers[k] = v
+
         timeout = model_config.get("timeout", constants.DEFAULT_TIMEOUT)
         max_retries = model_config.get("max_retries", 3)