caikit_nlp/config/config.yml

@@ -56,3 +56,6 @@ embedding:
 
 runtime:
   library: caikit_nlp
+
+# Configure request timeout for TGIS backend (in seconds)
+tgis_request_timeout: 60

caikit_nlp/toolkit/text_generation/tgis_utils.py

@@ -20,6 +20,7 @@
 import grpc
 
 # First Party
+from caikit import get_config
 from caikit.core.exceptions import error_handler
 from caikit.core.exceptions.caikit_core_exception import (
     CaikitCoreException,
@@ -326,6 +327,8 @@ def __init__(
         self.producer_id = producer_id
         self.prefix_id = prefix_id
 
+        self.tgis_req_timeout = get_config().tgis_request_timeout
+
     def unary_generate(
         self,
         text,
@@ -432,7 +435,9 @@ def unary_generate(
         # Currently, we send a batch request of len(x)==1, so we expect one response back
         with alog.ContextTimer(log.trace, "TGIS request duration: "):
             try:
-                batch_response = self.tgis_client.Generate(request)
+                batch_response = self.tgis_client.Generate(
+                    request, timeout=self.tgis_req_timeout
+                )
             except grpc.RpcError as err:
                 raise_caikit_core_exception(err)
 
@@ -576,7 +581,9 @@ def stream_generate(
 
         # stream GenerationResponse
         try:
-            stream_response = self.tgis_client.GenerateStream(request)
+            stream_response = self.tgis_client.GenerateStream(
+                request, timeout=self.tgis_req_timeout
+            )
 
             for stream_part in stream_response:
                 details = TokenStreamDetails(
@@ -645,7 +652,9 @@ def unary_tokenize(
         # Currently, we send a batch request of len(x)==1, so we expect one response back
         with alog.ContextTimer(log.trace, "TGIS request duration: "):
             try:
-                batch_response = self.tgis_client.Tokenize(request)
+                batch_response = self.tgis_client.Tokenize(
+                    request, timeout=self.tgis_req_timeout
+                )
             except grpc.RpcError as err:
                 raise_caikit_core_exception(err)
 

tests/fixtures/__init__.py

@@ -198,17 +198,17 @@ class StubTGISClient:
     def __init__(self, base_model_name):
         pass
 
-    def Generate(self, request):
+    def Generate(self, request, **kwargs):
         return StubTGISClient.unary_generate(request)
 
-    def GenerateStream(self, request):
+    def GenerateStream(self, request, **kwargs):
         return StubTGISClient.stream_generate(request)
 
-    def Tokenize(self, request):
+    def Tokenize(self, request, **kwargs):
         return StubTGISClient.tokenize(request)
 
     @staticmethod
-    def unary_generate(request):
+    def unary_generate(request, **kwargs):
         fake_response = mock.Mock()
         fake_result = mock.Mock()
         fake_result.stop_reason = 5
@@ -229,7 +229,7 @@ def unary_generate(request):
         return fake_response
 
     @staticmethod
-    def stream_generate(request):
+    def stream_generate(request, **kwargs):
         fake_stream = mock.Mock()
         fake_stream.stop_reason = 5
         fake_stream.generated_token_count = 1
@@ -250,7 +250,7 @@ def stream_generate(request):
             yield fake_stream
 
     @staticmethod
-    def tokenize(request):
+    def tokenize(request, **kwargs):
         fake_response = mock.Mock()
         fake_result = mock.Mock()
         fake_result.token_count = 1

tests/toolkit/text_generation/test_tgis_utils.py

@@ -54,22 +54,19 @@ def _maybe_raise(self, error_type: Type[grpc.RpcError], *args):
             )
 
     def Generate(
-        self,
-        request: generation_pb2.BatchedGenerationRequest,
+        self, request: generation_pb2.BatchedGenerationRequest, **kwargs
     ) -> generation_pb2.BatchedGenerationResponse:
         self._maybe_raise(grpc._channel._InactiveRpcError)
         return generation_pb2.BatchedGenerationResponse()
 
     def GenerateStream(
-        self,
-        request: generation_pb2.SingleGenerationRequest,
+        self, request: generation_pb2.SingleGenerationRequest, **kwargs
     ) -> Iterable[generation_pb2.GenerationResponse]:
         self._maybe_raise(grpc._channel._MultiThreadedRendezvous, None, None, None)
         yield generation_pb2.GenerationResponse()
 
     def Tokenize(
-        self,
-        request: generation_pb2.BatchedTokenizeRequest,
+        self, request: generation_pb2.BatchedTokenizeRequest, **kwargs
     ) -> generation_pb2.BatchedTokenizeResponse:
         self._maybe_raise(grpc._channel._InactiveRpcError)
         return generation_pb2.BatchedTokenizeResponse()