Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -24,17 +24,15 @@ class BasicAgent:
|
|
| 24 |
def __init__(self):
|
| 25 |
print("BasicAgent initialized.")
|
| 26 |
|
| 27 |
-
# 1. Fetch the
|
| 28 |
-
|
| 29 |
-
if not
|
| 30 |
-
raise ValueError("
|
| 31 |
|
| 32 |
-
#
|
| 33 |
-
# The first tells LiteLLM to connect to OpenRouter.
|
| 34 |
-
# The second tells OpenRouter to use its dynamic, load-balanced free pool.
|
| 35 |
self.model = LiteLLMModel(
|
| 36 |
-
model_id="
|
| 37 |
-
api_key=
|
| 38 |
)
|
| 39 |
|
| 40 |
self.tools = [
|
|
@@ -47,10 +45,10 @@ class BasicAgent:
|
|
| 47 |
self.agent = CodeAgent(
|
| 48 |
tools=self.tools,
|
| 49 |
model=self.model,
|
| 50 |
-
max_steps=
|
| 51 |
additional_authorized_imports=["datetime", "re", "json", "math", "collections"],
|
| 52 |
)
|
| 53 |
-
print("BasicAgent ready with
|
| 54 |
|
| 55 |
def __call__(self, question: str) -> str:
|
| 56 |
print(f"\nAgent received question: {question[:80]}...")
|
|
@@ -58,14 +56,14 @@ class BasicAgent:
|
|
| 58 |
|
| 59 |
for attempt in range(max_retries):
|
| 60 |
try:
|
| 61 |
-
# A
|
| 62 |
time.sleep(2)
|
| 63 |
answer = self.agent.run(question)
|
| 64 |
print(f"Agent answer: {str(answer)[:200]}")
|
| 65 |
return str(answer)
|
| 66 |
except Exception as e:
|
| 67 |
err_msg = str(e).lower()
|
| 68 |
-
# If we hit a rate limit, pause and retry
|
| 69 |
if "429" in err_msg or "rate limit" in err_msg or "too many requests" in err_msg:
|
| 70 |
wait_time = 20 * (attempt + 1)
|
| 71 |
print(f"Rate limit hit! Pausing for {wait_time} seconds before retrying (Attempt {attempt+1}/{max_retries})...")
|
|
@@ -125,7 +123,7 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 125 |
print(f"Error on task {task_id}: {e}")
|
| 126 |
results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": f"ERROR: {e}"})
|
| 127 |
|
| 128 |
-
# CRITICAL FIX: Give the API token bucket
|
| 129 |
print("Cooling down for 15 seconds to prevent token exhaustion...")
|
| 130 |
time.sleep(15)
|
| 131 |
|
|
@@ -154,11 +152,11 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 154 |
|
| 155 |
# --- Build Gradio UI ---
|
| 156 |
with gr.Blocks() as demo:
|
| 157 |
-
gr.Markdown("# Basic Agent Evaluation Runner")
|
| 158 |
gr.Markdown(
|
| 159 |
"""
|
| 160 |
**Instructions:**
|
| 161 |
-
1. Set `
|
| 162 |
2. Log in with your Hugging Face account below.
|
| 163 |
3. Click 'Run Evaluation & Submit' to start.
|
| 164 |
"""
|
|
|
|
| 24 |
def __init__(self):
|
| 25 |
print("BasicAgent initialized.")
|
| 26 |
|
| 27 |
+
# 1. Fetch the Groq API Key
|
| 28 |
+
groq_api_key = os.getenv("GROQ_API_KEY")
|
| 29 |
+
if not groq_api_key:
|
| 30 |
+
raise ValueError("GROQ_API_KEY environment variable not set in Space Secrets.")
|
| 31 |
|
| 32 |
+
# 2. Use LiteLLM to connect to Groq's Llama 3.3 70B (Extremely smart, fast, and free)
|
|
|
|
|
|
|
| 33 |
self.model = LiteLLMModel(
|
| 34 |
+
model_id="groq/llama-3.3-70b-versatile",
|
| 35 |
+
api_key=groq_api_key,
|
| 36 |
)
|
| 37 |
|
| 38 |
self.tools = [
|
|
|
|
| 45 |
self.agent = CodeAgent(
|
| 46 |
tools=self.tools,
|
| 47 |
model=self.model,
|
| 48 |
+
max_steps=7, # Capped at 7 to give it enough thought cycles without looping infinitely
|
| 49 |
additional_authorized_imports=["datetime", "re", "json", "math", "collections"],
|
| 50 |
)
|
| 51 |
+
print("BasicAgent ready with Groq (Llama-3.3-70b-versatile).")
|
| 52 |
|
| 53 |
def __call__(self, question: str) -> str:
|
| 54 |
print(f"\nAgent received question: {question[:80]}...")
|
|
|
|
| 56 |
|
| 57 |
for attempt in range(max_retries):
|
| 58 |
try:
|
| 59 |
+
# A small 2-second buffer per thought step
|
| 60 |
time.sleep(2)
|
| 61 |
answer = self.agent.run(question)
|
| 62 |
print(f"Agent answer: {str(answer)[:200]}")
|
| 63 |
return str(answer)
|
| 64 |
except Exception as e:
|
| 65 |
err_msg = str(e).lower()
|
| 66 |
+
# If we hit a Groq rate limit, pause and retry
|
| 67 |
if "429" in err_msg or "rate limit" in err_msg or "too many requests" in err_msg:
|
| 68 |
wait_time = 20 * (attempt + 1)
|
| 69 |
print(f"Rate limit hit! Pausing for {wait_time} seconds before retrying (Attempt {attempt+1}/{max_retries})...")
|
|
|
|
| 123 |
print(f"Error on task {task_id}: {e}")
|
| 124 |
results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": f"ERROR: {e}"})
|
| 125 |
|
| 126 |
+
# CRITICAL FIX: Give the API token bucket 15 seconds to cool down between questions
|
| 127 |
print("Cooling down for 15 seconds to prevent token exhaustion...")
|
| 128 |
time.sleep(15)
|
| 129 |
|
|
|
|
| 152 |
|
| 153 |
# --- Build Gradio UI ---
|
| 154 |
with gr.Blocks() as demo:
|
| 155 |
+
gr.Markdown("# Basic Agent Evaluation Runner (Groq Edition)")
|
| 156 |
gr.Markdown(
|
| 157 |
"""
|
| 158 |
**Instructions:**
|
| 159 |
+
1. Set `GROQ_API_KEY` in your Space Secrets.
|
| 160 |
2. Log in with your Hugging Face account below.
|
| 161 |
3. Click 'Run Evaluation & Submit' to start.
|
| 162 |
"""
|