Merge pull request #54 from ansari-project/fix-build

Fix continuous testing.
ansari-project · Oct 19, 2024 · cc24e3a · cc24e3a
2 parents fba0a22 + f38f13c
commit cc24e3a
Show file tree

Hide file tree

Showing 3 changed files with 18 additions and 20 deletions.
diff --git a/.gitignore b/.gitignore
@@ -7,6 +7,7 @@ share/
 pyvenv.cfg
 example_projects/
 abandoned/
+build/
 data/
 datasources/
 etc/

diff --git a/agents/ansari.py b/agents/ansari.py
@@ -85,12 +85,13 @@ def replace_message_history(self, message_history ):
 
     @observe(capture_input = False,capture_output = False)
     def process_message_history(self):
-        langfuse_context.update_current_trace(
-            user_id = self.message_logger.user_id,
-            session_id = str(self.message_logger.thread_id),
-            tags = ['debug', 'replace_message_history'],
-            input = self.message_history
-        )
+        if self.message_logger is not None:
+            langfuse_context.update_current_trace(
+                user_id = self.message_logger.user_id,
+                session_id = str(self.message_logger.thread_id),
+                tags = ['debug', 'replace_message_history'],
+                input = self.message_history
+            )
         # Keep processing the user input until we get something from the assistant
         self.start_time = datetime.now()
         count = 0
@@ -121,11 +122,6 @@ def process_message_history(self):
 
     @observe(as_type="generator")
     def process_one_round(self, use_function=True):
-        langfuse_context.update_current_trace(
-            user_id = self.message_logger.user_id,
-            session_id = str(self.message_logger.thread_id),
-            tags = ['debug', 'replace_message_history']
-        )
         response = None
         failures = 0
         while not response:
@@ -136,6 +132,7 @@ def process_one_round(self, use_function=True):
                             model=self.model,
                             messages=self.message_history,
                             stream=True,
+                            stream_options = {"include_usage": True}, 
                             functions=self.functions,
                             timeout=30.0,
                             temperature=0.0,
@@ -148,6 +145,7 @@ def process_one_round(self, use_function=True):
                             model=self.model,
                             messages=self.message_history,
                             stream=True,
+                            stream_options = {"include_usage": True}, 
                             functions=self.functions,
                             timeout=30.0,
                             temperature=0.0,
@@ -160,6 +158,7 @@ def process_one_round(self, use_function=True):
                             model=self.model,
                             messages=self.message_history,
                             stream=True,
+                            stream_options = {"include_usage": True}, 
                             timeout=30.0,
                             temperature=0.0,
                             response_format={"type": "json_object"},
@@ -171,6 +170,7 @@ def process_one_round(self, use_function=True):
                             model=self.model,
                             messages=self.message_history,
                             stream=True,
+                            stream_options = {"include_usage": True}, 
                             timeout=30.0,
                             temperature=0.0,
                             metadata={"generation-name": "ansari"},
@@ -193,7 +193,9 @@ def process_one_round(self, use_function=True):
         function_arguments = ""
         response_mode = ""  # words or fn
         for tok in response:
-            logger.debug(f"Tok is {tok}")
+            if len(tok.choices) == 0: # in case usage is defind.q 
+                logging.warning(f"Token has no choices: {tok}")
+                langfuse_context.update_current_observation(usage = tok.usage)  
             delta = tok.choices[0].delta
             if not response_mode:
                 # This code should only trigger the first
@@ -250,11 +252,6 @@ def process_one_round(self, use_function=True):
 
     @observe()
     def process_fn_call(self, orig_question, function_name, function_arguments):
-        langfuse_context.update_current_trace(
-            user_id = self.message_logger.user_id,
-            session_id = str(self.message_logger.thread_id),
-            tags = ['debug', 'replace_message_history']
-        )
         if function_name in self.tools.keys():
             args = json.loads(function_arguments)
             query = args["query"]

diff --git a/config.py b/config.py
@@ -51,9 +51,9 @@ class Settings(BaseSettings):
     VECTARA_CORPUS_ID: str
     DISCORD_TOKEN: Optional[SecretStr] = Field(default=None)
     SENDGRID_API_KEY: Optional[SecretStr] = Field(default=None)
-    LANGFUSE_SECRET_KEY: SecretStr = Field(default=None)
-    LANGFUSE_PUBLIC_KEY: SecretStr = Field(default=None)
-    LANGFUSE_HOST: str = Field(default=None) 
+    LANGFUSE_SECRET_KEY: Optional[SecretStr] = Field(default=None)
+    LANGFUSE_PUBLIC_KEY: Optional[SecretStr] = Field(default=None)
+    LANGFUSE_HOST: Optional[str] = Field(default=None) 
 
     template_dir: DirectoryPath = Field(default="resources/templates")
     diskcache_dir: str = Field(default="diskcache_dir")