Merge remote-tracking branch 'origin/main' into TamiTakamiya/tool-param-definition-update

llama_stack/apis/inference/inference.py

@@ -1008,7 +1008,6 @@ class InferenceProvider(Protocol):
 
     model_store: ModelStore | None = None
 
-    @webmethod(route="/inference/completion", method="POST", level=LLAMA_STACK_API_V1)
     async def completion(
         self,
         model_id: str,

llama_stack/providers/inline/telemetry/meta_reference/telemetry.py

@@ -224,10 +224,6 @@ class TelemetryAdapter(TelemetryDatasetMixin, Telemetry):
         return _GLOBAL_STORAGE["gauges"][name]
 
     def _log_metric(self, event: MetricEvent) -> None:
-        # Always log to console if console sink is enabled (debug)
-        if TelemetrySink.CONSOLE in self.config.sinks:
-            logger.debug(f"METRIC: {event.metric}={event.value} {event.unit} {event.attributes}")
-
         # Add metric as an event to the current span
         try:
             with self._lock:

llama_stack/providers/remote/inference/fireworks/fireworks.py

@@ -61,6 +61,7 @@ logger = get_logger(name=__name__, category="inference::fireworks")
 class FireworksInferenceAdapter(OpenAIMixin, ModelRegistryHelper, Inference, NeedsRequestProviderData):
     embedding_model_metadata = {
         "nomic-ai/nomic-embed-text-v1.5": {"embedding_dimension": 768, "context_length": 8192},
+        "accounts/fireworks/models/qwen3-embedding-8b": {"embedding_dimension": 4096, "context_length": 40960},
     }
 
     def __init__(self, config: FireworksImplConfig) -> None:

llama_stack/providers/utils/inference/prompt_adapter.py

@@ -192,6 +192,14 @@ async def localize_image_content(uri: str) -> tuple[bytes, str] | None:
                 format = "png"
 
         return content, format
+    elif uri.startswith("data"):
+        # data:image/{format};base64,{data}
+        match = re.match(r"data:image/(\w+);base64,(.+)", uri)
+        if not match:
+            raise ValueError(f"Invalid data URL format, {uri[:40]}...")
+        fmt, image_data = match.groups()
+        content = base64.b64decode(image_data)
+        return content, fmt
     else:
         return None