change the models, add tracing.

app.py

@@ -10,6 +10,36 @@ import basic_agent
 # --- Constants ---
 DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
 
+# Tracing to Lanfuse
+from opentelemetry.sdk.trace import TracerProvider
+from openinference.instrumentation.smolagents import SmolagentsInstrumentor
+from opentelemetry.exporter.otlp.proto.http.trace_exporter import OTLPSpanExporter
+from opentelemetry.sdk.trace.export import SimpleSpanProcessor
+
+LANGFUSE_PUBLIC_KEY = os.environ.get("LANGFUSE_PUBLIC_KEY")
+LANGFUSE_SECRET_KEY = os.environ.get("LANGFUSE_SECRET_KEY")
+
+LANGFUSE_AUTH = base64.b64encode(
+    f"{LANGFUSE_PUBLIC_KEY}:{LANGFUSE_SECRET_KEY}".encode()
+).decode()
+
+os.environ["OTEL_EXPORTER_OTLP_ENDPOINT"] = os.environ.get("LANGFUSE_HOST") + "/api/public/otel"
+os.environ["OTEL_EXPORTER_OTLP_HEADERS"] = f"Authorization=Basic {LANGFUSE_AUTH}"
+
+# Create a TracerProvider for OpenTelemetry
+trace_provider = TracerProvider()
+
+# Add a SimpleSpanProcessor with the OTLPSpanExporter to send traces
+trace_provider.add_span_processor(SimpleSpanProcessor(OTLPSpanExporter()))
+
+# Set the global default tracer provider
+from opentelemetry import trace
+trace.set_tracer_provider(trace_provider)
+tracer = trace.get_tracer(__name__)
+
+# Instrument smolagents with the configured provider
+SmolagentsInstrumentor().instrument(tracer_provider=trace_provider)
+
 
 def run_and_submit_all(profile: gr.OAuthProfile | None):
   """

basic_agent.py

@@ -5,6 +5,7 @@ import time
 
 import smolagents
 
+# Logs appear to be swallowed.
 LOG = logging.getLogger(__name__)
 
 SYSTEM_PROMPT = """
@@ -13,44 +14,37 @@ You are a general AI assistant. I will ask you a question. Report your thoughts,
 The current date is April 30, 2025.
 
 Take the time to plan the steps to reach the solution. Show the steps and then execute the steps.
+
+The question is below:
+"""
+
+SUFFIX_PROMPT = """
+Remember - The FINAL ANSWER must be a string, a number or a comma separated list that is only strings and numbers. No other output is allowed for a FINAL ANSWER! If any other output is provided other than a string, number or comma separated list that contains only strings and numbers, you fail. Don't fail.
 """
 
 
 class BasicAgent:
+  def check_results(self, final_answer, agent_memory):
+    if isinstance(final_answer, str) or isinstance(final_answer, int) or isinstance(final_answer, float):
+      return True
+    failure = "FAILED! The final answer must be a string, a number or a comma separated list of strings and numbers."
+    print(f"Feedback: {failure}")
+    raise Exception(failure)
 
   def __init__(self):
     print("BasicAgent initializing.")
-    # Logs appear to be swallowed.
-    LOG.warning("logging BasicAgent initialized.")
 
-    self.manager_model_id = "Qwen/Qwen3-32B"
+# too big
+#    self.manager_model_id = "google/gemma-3-27B-it"
+#    self.manager_model_id = "OpenGVLab/InternVL3-14B"
+#    self.manager_model_id = "AIDC-AI/Ovis2-34B"
+
+    self.manager_model_id = "meta-llama/Llama-3.3-70B-Instruct"
     self.manager_model = None
 
     self.code_model_id = "Qwen/Qwen2.5-Coder-32B-Instruct"
     self.code_model = None
 
-    """
-    # Handle the special cases
-    if self.model_id.lower() == "google":
-      self.model_id = "google"
-
-      # Use Google gemini free tier
-      GEM_KEY=os.environ["GOOGLE_API_KEY"]
-      self.model = smolagents.OpenAIServerModel(
-          model_id="gemini-2.0-flash",
-          api_base="https://generativelanguage.googleapis.com/v1beta/openai/",
-          api_key=GEM_KEY,
-          temperature=0.3)
-    elif self.model_id.lower() == "local":
-      self.model_id = "Qwen/Qwen3-4B-FP8"
-      # Run locally.
-      self.model = smolagents.TransformersModel(
-        model_id=self.model_id,
-        max_new_tokens=32000,
-        temperature=0.3
-        )
-     """
-
     if not self.manager_model:
       self.manager_model = smolagents.HfApiModel(
           max_tokens=32000,
@@ -67,7 +61,7 @@ class BasicAgent:
           custom_role_conversions=None,
       )
 
-    print(f"NEW3: BasicAgent {self.manager_model_id=} {self.code_model_id=}")
+    print(f"NEW4: BasicAgent {self.manager_model_id=} {self.code_model_id=}")
 
     self.math_agent = smolagents.CodeAgent(
         name="date_time_math_agent",
@@ -86,7 +80,6 @@ class BasicAgent:
         smolagents.FinalAnswerTool()
     ]
 
-    print("BasicAgent making search tool.")
     self.web_search_agent = smolagents.CodeAgent(
         name="web_search_agent",
         description="Search the web",
@@ -115,10 +108,6 @@ class BasicAgent:
         additional_authorized_imports=["wikipedia-api"],
     )
 
-    print("BasicAgent making wikipedia search tool.")
-
-
-    print("BasicAgent making manager.")
     self.manager_agent = smolagents.CodeAgent(
         name="manager_agent",
         description="Manger of other agents",
@@ -129,11 +118,13 @@ class BasicAgent:
         verbosity_level=2,
         planning_interval=None,
         additional_authorized_imports=["duckduckgo_search", "wikipedia-api"],
-        managed_agents=[self.web_search_agent, self.wikipedia_search_agent, self.math_agent])
+        managed_agents=[self.web_search_agent, self.wikipedia_search_agent, self.math_agent],
+        final_answer_checks=[self.check_results],
+        )
 
   def __call__(self, question: str) -> str:
-    print(f"NEW Agent received question (first 50 chars): {question[:50]}...")
-    prompt = f"{SYSTEM_PROMPT}\n\n{question}"
+    print(f"Agent received question (first 50 chars): {question[:50]}...")
+    prompt = f'{SYSTEM_PROMPT}\n\n{question}\n\n {SUFFIX_PROMPT}'
     answer = self.manager_agent.run(prompt)
-    print(f"NEW {answer=}")
+    print(f"{answer=}")
     return answer

basic_agent_test.py

@@ -6,6 +6,31 @@ import basic_agent
 
 LOG = logging.getLogger(__name__)
 
+
+# Local:
+#  https://huggingface.co/docs/smolagents/tutorials/inspect_runs
+#  https://cobusgreyling.medium.com/introduce-inspectability-to-huggingface-smolagents-571bd3f8da4c
+#
+# pip install arize-phoenix opentelemetry-sdk opentelemetry-exporter-otlp openinference-instrumentation-smolagents
+#
+# pip install 'smolagents[telemetry]'
+#
+# python -m phoenix.server.main serve
+
+from opentelemetry import trace
+from opentelemetry.sdk.trace import TracerProvider
+from opentelemetry.sdk.trace.export import BatchSpanProcessor
+
+from openinference.instrumentation.smolagents import SmolagentsInstrumentor
+from opentelemetry.exporter.otlp.proto.http.trace_exporter import OTLPSpanExporter
+from opentelemetry.sdk.trace.export import ConsoleSpanExporter, SimpleSpanProcessor
+
+endpoint = "http://0.0.0.0:6006/v1/traces"
+trace_provider = TracerProvider()
+trace_provider.add_span_processor(SimpleSpanProcessor(OTLPSpanExporter(endpoint)))
+
+SmolagentsInstrumentor().instrument(tracer_provider=trace_provider)
+
 ba = basic_agent.BasicAgent()
 
 """

requirements.txt

@@ -5,6 +5,15 @@ smolagents
 smolagents[openai]
 wikipedia-api
 
+# For tracing/monitoring
+smolagents[telemetry]
+opentelemetry-sdk
+opentelemetry-exporter-otlp
+openinference-instrumentation-smolagents
+langfuse
+datasets
+smolagents[gradio]
+
 # Only needed for running locally
 #compressed-tensors
 #smolagents[transformers]