for agents API, provider data from the header is not parsed as for agents there is no provider_data_validator meta-reference implementation. Added Together data validator as the provider_data_validator for now. Did some code changes accordingly.

llama_stack/distribution/request_headers.py

@@ -18,6 +18,7 @@ def get_request_provider_data() -> Any:
 
 
 def set_request_provider_data(headers: Dict[str, str], validator_classes: List[str]):
+
     if not validator_classes:
         return
 

llama_stack/distribution/server/server.py

@@ -199,9 +199,12 @@ async def lifespan(app: FastAPI):
 
 
 def create_dynamic_passthrough(
-    downstream_url: str, downstream_headers: Optional[Dict[str, str]] = None
+    downstream_url: str,
+    provider_data_validators: List[str],
+    downstream_headers: Optional[Dict[str, str]] = None,
 ):
     async def endpoint(request: Request):
+        set_request_provider_data(request.headers, provider_data_validators)
         return await passthrough(request, downstream_url, downstream_headers)
 
     return endpoint
@@ -223,7 +226,6 @@ def create_dynamic_typed_route(
 
         async def endpoint(request: Request, **kwargs):
             await start_trace(func.__name__)
-
             set_request_provider_data(request.headers, provider_data_validators)
 
             async def sse_generator(event_gen):
@@ -446,6 +448,16 @@ def main(yaml_config: str, port: int = 5000, disable_ipv6: bool = False):
         impl = impls[api]
 
         provider_spec = specs[api]
+        validators = []
+        if isinstance(provider_spec, AutoRoutedProviderSpec):
+            inner_specs = specs[provider_spec.routing_table_api].inner_specs
+            for spec in inner_specs:
+                if spec.provider_data_validator:
+                    validators.append(spec.provider_data_validator)
+        elif not isinstance(provider_spec, RoutingTableProviderSpec):
+            if provider_spec.provider_data_validator:
+                validators.append(provider_spec.provider_data_validator)
+
         if (
             isinstance(provider_spec, RemoteProviderSpec)
             and provider_spec.adapter is None
@@ -453,7 +465,7 @@ def main(yaml_config: str, port: int = 5000, disable_ipv6: bool = False):
             for endpoint in endpoints:
                 url = impl.__provider_config__.url.rstrip("/") + endpoint.route
                 getattr(app, endpoint.method)(endpoint.route)(
-                    create_dynamic_passthrough(url)
+                    create_dynamic_passthrough(url, validators)
                 )
         else:
             for endpoint in endpoints:
@@ -465,16 +477,6 @@ def main(yaml_config: str, port: int = 5000, disable_ipv6: bool = False):
 
                 impl_method = getattr(impl, endpoint.name)
 
-                validators = []
-                if isinstance(provider_spec, AutoRoutedProviderSpec):
-                    inner_specs = specs[provider_spec.routing_table_api].inner_specs
-                    for spec in inner_specs:
-                        if spec.provider_data_validator:
-                            validators.append(spec.provider_data_validator)
-                elif not isinstance(provider_spec, RoutingTableProviderSpec):
-                    if provider_spec.provider_data_validator:
-                        validators.append(provider_spec.provider_data_validator)
-
                 getattr(app, endpoint.method)(endpoint.route, response_model=None)(
                     create_dynamic_typed_route(
                         impl_method,

llama_stack/providers/adapters/inference/together/together.py

@@ -3,7 +3,6 @@
 #
 # This source code is licensed under the terms described in the LICENSE file in
 # the root directory of this source tree.
-
 from typing import AsyncGenerator
 
 from llama_models.llama3.api.chat_format import ChatFormat
@@ -61,6 +60,20 @@ class TogetherInferenceAdapter(Inference):
                 role = "tool"
             else:
                 role = message.role
+
+            if role == "user" and type(message.content) == list:
+                contents = []
+                for content in message.content:
+                    if type(content) == str:
+                        contents.append({"type": "text", "text": content})
+                    elif type(content) == ImageMedia:
+                        contents.append(
+                            {
+                                "type": "image_url",
+                                "image_url": {"url": content.image.uri},
+                            }
+                        )
+                message.content = contents
             together_messages.append({"role": role, "content": message.content})
 
         return together_messages

llama_stack/providers/impls/meta_reference/agents/agent_instance.py

@@ -541,6 +541,7 @@ class ChatAgent(ShieldRunnerMixin):
                             )
                         )
                     )
+
                     yield AgentTurnResponseStreamChunk(
                         event=AgentTurnResponseEvent(
                             payload=AgentTurnResponseStepProgressPayload(

llama_stack/providers/registry/agents.py

@@ -29,6 +29,7 @@ def available_providers() -> List[ProviderSpec]:
                 Api.safety,
                 Api.memory,
             ],
+            provider_data_validator="llama_stack.providers.adapters.safety.together.TogetherProviderDataValidator",
         ),
         remote_provider_spec(
             api=Api.agents,