Switch model backend from Gemini to Groq

Gemini free tier caps at 20 requests/day for gemini-3.6-flash - too low for a multi-step agent benchmark. Groq's llama-3.3-70b-versatile free tier allows 1000/day, 30 RPM. Uses GROQ_API_KEY.

app.py

@@ -24,7 +24,7 @@ DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
 @spaces.GPU
 def _zerogpu_startup_check():
     # This Space runs on ZeroGPU hardware but the agent below only makes
-    # network calls (Gemini API, web search) and never touches CUDA.
+    # network calls (Groq API, web search) and never touches CUDA.
     # ZeroGPU requires at least one @spaces.GPU function to be declared,
     # so this no-op satisfies that check without spending any GPU quota
     # (it is never actually invoked).
@@ -43,11 +43,13 @@ class RateLimiter:
     """
     Enforces a minimum delay between successive agent LLM calls (one call per
     step) as a step_callback, to stay under the model provider's
-    requests-per-minute limit. Gemini's free tier for gemini-3.6-flash caps
-    at 5 RPM, so the default here (12s + margin) targets that; override via
-    RATE_LIMIT_SECONDS_BETWEEN_CALLS if your tier/model allows more.
+    requests-per-minute limit. Groq's free tier for llama-3.3-70b-versatile
+    caps at 30 RPM, so the default here (2.5s) targets that with a small
+    margin; override via RATE_LIMIT_SECONDS_BETWEEN_CALLS for other tiers.
+    Note this only protects against per-minute limits - free tiers also
+    often cap total requests/day, which this can't work around.
     """
-    def __init__(self, min_seconds_between_calls: float = 13.0):
+    def __init__(self, min_seconds_between_calls: float = 2.5):
         self.min_seconds_between_calls = min_seconds_between_calls
         self._last_call_at: float | None = None
 
@@ -65,10 +67,10 @@ class RateLimiter:
 # ----- THIS IS WERE YOU CAN BUILD WHAT YOU WANT ------
 class BasicAgent:
     def __init__(self):
-        model_id = os.getenv("AGENT_MODEL_ID", "gemini/gemini-3.6-flash")
-        api_key = os.getenv("GEMINI_API_KEY")
+        model_id = os.getenv("AGENT_MODEL_ID", "groq/llama-3.3-70b-versatile")
+        api_key = os.getenv("GROQ_API_KEY")
         if not api_key:
-            print("Warning: GEMINI_API_KEY is not set - the agent will fail to call the model.")
+            print("Warning: GROQ_API_KEY is not set - the agent will fail to call the model.")
 
         self.model = LiteLLMModel(model_id=model_id, api_key=api_key, temperature=0)
         self.agent = CodeAgent(
@@ -80,7 +82,7 @@ class BasicAgent:
                 "collections", "statistics", "datetime", "io", "openpyxl", "PIL",
             ],
             max_steps=12,
-            step_callbacks=[RateLimiter(float(os.getenv("RATE_LIMIT_SECONDS_BETWEEN_CALLS", "13.0")))],
+            step_callbacks=[RateLimiter(float(os.getenv("RATE_LIMIT_SECONDS_BETWEEN_CALLS", "2.5")))],
         )
         print("BasicAgent initialized.")
 
@@ -253,7 +255,7 @@ with gr.Blocks() as demo:
         Once clicking on the "submit button, it can take quite some time ( this is the time for the agent to go through all the questions).
         This space provides a basic setup and is intentionally sub-optimal to encourage you to develop your own, more robust solution. For instance for the delay process of the submit button, a solution could be to cache the answers and submit in a seperate action or even to answer the questions in async.
 
-        **Setup:** This agent calls Google's Gemini via `smolagents`. Get a free key at https://aistudio.google.com/apikey and set it as the `GEMINI_API_KEY` secret in this Space's settings before running.
+        **Setup:** This agent calls Groq (Llama 3.3 70B) via `smolagents`. Get a free key at https://console.groq.com/keys and set it as the `GROQ_API_KEY` secret in this Space's settings before running.
         """
     )