--- /tmp/evalscope-sync/src-v191/evalscope/models/anthropic_compatible.py	2026-07-21 03:18:18.000000000 +0000
+++ /data1/syy/evalscope/evalstone/evalscope/evalscope/models/anthropic_compatible.py	2026-09-01 03:48:30.947904360 +0000
@@ -172,42 +172,44 @@
 
         self.validate_request_params(request)
 
-        try:
-            t_start = time.monotonic()
-            ttft: Optional[float] = None
-
-            # Generate completion
-            message = retry_call(
-                self.client.messages.create,
-                retries=config.retries,
-                sleep_interval=config.retry_interval,
-                **request,
-            )
-
-            # Handle streaming response
-            if not isinstance(message, Message):
-                message, ttft = collect_stream_response(message, request_start=t_start)
-
-            total_time = time.monotonic() - t_start
-
-            response = message.model_dump()
-            self.on_response(response)
-
-            # Build output and populate timing + perf metrics
-            choices = self.chat_choices_from_message(message, tools)
-            output = model_output_from_anthropic(message, choices)
-            output.time = total_time
-            usage = output.usage
-            output.message.perf_metrics = PerformanceMetrics(
-                latency=total_time,
-                ttft=ttft,
-                input_tokens=usage.input_tokens if usage else 0,
-                output_tokens=usage.output_tokens if usage else 0,
-            )
-            return output
+        with self._track_logical_request():
+            try:
+                t_start = time.monotonic()
+                ttft: Optional[float] = None
+
+                # Generate completion
+                message = retry_call(
+                    self.client.messages.create,
+                    retries=config.retries,
+                    sleep_interval=config.retry_interval,
+                    on_attempt=self.request_stats.on_attempt,
+                    **request,
+                )
+
+                # Handle streaming response
+                if not isinstance(message, Message):
+                    message, ttft = collect_stream_response(message, request_start=t_start)
+
+                total_time = time.monotonic() - t_start
+
+                response = message.model_dump()
+                self.on_response(response)
+
+                # Build output and populate timing + perf metrics
+                choices = self.chat_choices_from_message(message, tools)
+                output = model_output_from_anthropic(message, choices)
+                output.time = total_time
+                usage = output.usage
+                output.message.perf_metrics = PerformanceMetrics(
+                    latency=total_time,
+                    ttft=ttft,
+                    input_tokens=usage.input_tokens if usage else 0,
+                    output_tokens=usage.output_tokens if usage else 0,
+                )
+                return output
 
-        except (BadRequestError, PermissionDeniedError) as ex:
-            return self.handle_bad_request(ex)
+            except (BadRequestError, PermissionDeniedError) as ex:
+                return self.handle_bad_request(ex)
 
     async def generate_async(
         self,
@@ -256,41 +258,43 @@
 
         self.validate_request_params(request)
 
-        try:
-            t_start = time.monotonic()
-            ttft: Optional[float] = None
-
-            # Async generation with retry
-            message = await async_retry_call(
-                self.async_client.messages.create,
-                retries=config.retries,
-                sleep_interval=config.retry_interval,
-                **request,
-            )
-
-            # Handle streaming response
-            if not isinstance(message, Message):
-                message, ttft = await async_collect_stream_response(message, request_start=t_start)
-
-            total_time = time.monotonic() - t_start
-
-            response = message.model_dump()
-            self.on_response(response)
-
-            choices = self.chat_choices_from_message(message, tools)
-            output = model_output_from_anthropic(message, choices)
-            output.time = total_time
-            usage = output.usage
-            output.message.perf_metrics = PerformanceMetrics(
-                latency=total_time,
-                ttft=ttft,
-                input_tokens=usage.input_tokens if usage else 0,
-                output_tokens=usage.output_tokens if usage else 0,
-            )
-            return output
+        with self._track_logical_request():
+            try:
+                t_start = time.monotonic()
+                ttft: Optional[float] = None
+
+                # Async generation with retry
+                message = await async_retry_call(
+                    self.async_client.messages.create,
+                    retries=config.retries,
+                    sleep_interval=config.retry_interval,
+                    on_attempt=self.request_stats.on_attempt,
+                    **request,
+                )
+
+                # Handle streaming response
+                if not isinstance(message, Message):
+                    message, ttft = await async_collect_stream_response(message, request_start=t_start)
+
+                total_time = time.monotonic() - t_start
+
+                response = message.model_dump()
+                self.on_response(response)
+
+                choices = self.chat_choices_from_message(message, tools)
+                output = model_output_from_anthropic(message, choices)
+                output.time = total_time
+                usage = output.usage
+                output.message.perf_metrics = PerformanceMetrics(
+                    latency=total_time,
+                    ttft=ttft,
+                    input_tokens=usage.input_tokens if usage else 0,
+                    output_tokens=usage.output_tokens if usage else 0,
+                )
+                return output
 
-        except (BadRequestError, PermissionDeniedError) as ex:
-            return self.handle_bad_request(ex)
+            except (BadRequestError, PermissionDeniedError) as ex:
+                return self.handle_bad_request(ex)
 
     def resolve_tools(self, tools: List[ToolInfo], tool_choice: ToolChoice,
                       config: GenerateConfig) -> Tuple[List[ToolInfo], ToolChoice, GenerateConfig]: