Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -24,13 +24,10 @@ class BasicAgent:
|
|
| 24 |
def __init__(self):
|
| 25 |
print("BasicAgent initialized.")
|
| 26 |
|
| 27 |
-
|
| 28 |
-
new_hf_token = os.getenv("HF_TOKEN")
|
| 29 |
if not new_hf_token:
|
| 30 |
-
raise ValueError("
|
| 31 |
|
| 32 |
-
# 2. Connect to Hugging Face's free Serverless Inference API
|
| 33 |
-
# Qwen2.5-Coder-32B-Instruct is the recommended model for this course
|
| 34 |
self.model = InferenceClientModel(
|
| 35 |
model_id="Qwen/Qwen2.5-Coder-32B-Instruct",
|
| 36 |
token=new_hf_token,
|
|
@@ -46,8 +43,9 @@ class BasicAgent:
|
|
| 46 |
self.agent = CodeAgent(
|
| 47 |
tools=self.tools,
|
| 48 |
model=self.model,
|
| 49 |
-
max_steps=
|
| 50 |
-
|
|
|
|
| 51 |
)
|
| 52 |
print("BasicAgent ready with Hugging Face (Qwen2.5-Coder-32B).")
|
| 53 |
|
|
@@ -65,7 +63,7 @@ class BasicAgent:
|
|
| 65 |
err_msg = str(e).lower()
|
| 66 |
if "429" in err_msg or "rate limit" in err_msg or "too many requests" in err_msg:
|
| 67 |
wait_time = 20 * (attempt + 1)
|
| 68 |
-
print(f"Rate limit hit! Pausing for {wait_time} seconds before retrying
|
| 69 |
time.sleep(wait_time)
|
| 70 |
else:
|
| 71 |
print(f"Agent error processing question: {e}")
|
|
@@ -111,11 +109,31 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 111 |
for i, item in enumerate(questions_data):
|
| 112 |
task_id = item.get("task_id")
|
| 113 |
question_text = item.get("question")
|
|
|
|
|
|
|
|
|
|
|
|
|
| 114 |
if not task_id or not question_text:
|
| 115 |
continue
|
| 116 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 117 |
try:
|
| 118 |
-
|
|
|
|
| 119 |
answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
|
| 120 |
results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": submitted_answer})
|
| 121 |
except Exception as e:
|
|
@@ -150,12 +168,12 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 150 |
|
| 151 |
# --- Build Gradio UI ---
|
| 152 |
with gr.Blocks() as demo:
|
| 153 |
-
gr.Markdown("# Basic Agent Evaluation Runner
|
| 154 |
gr.Markdown(
|
| 155 |
"""
|
| 156 |
**Instructions:**
|
| 157 |
-
1.
|
| 158 |
-
2. Log in
|
| 159 |
3. Click 'Run Evaluation & Submit' to start.
|
| 160 |
"""
|
| 161 |
)
|
|
|
|
| 24 |
def __init__(self):
|
| 25 |
print("BasicAgent initialized.")
|
| 26 |
|
| 27 |
+
new_hf_token = os.getenv("NEW_HF_TOKEN")
|
|
|
|
| 28 |
if not new_hf_token:
|
| 29 |
+
raise ValueError("NEW_HF_TOKEN environment variable not set in Space Secrets.")
|
| 30 |
|
|
|
|
|
|
|
| 31 |
self.model = InferenceClientModel(
|
| 32 |
model_id="Qwen/Qwen2.5-Coder-32B-Instruct",
|
| 33 |
token=new_hf_token,
|
|
|
|
| 43 |
self.agent = CodeAgent(
|
| 44 |
tools=self.tools,
|
| 45 |
model=self.model,
|
| 46 |
+
max_steps=10,
|
| 47 |
+
# CRITICAL FIX: Added pandas and requests so the agent can download and read Excel/CSV files!
|
| 48 |
+
additional_authorized_imports=["datetime", "re", "json", "math", "collections", "pandas", "requests"],
|
| 49 |
)
|
| 50 |
print("BasicAgent ready with Hugging Face (Qwen2.5-Coder-32B).")
|
| 51 |
|
|
|
|
| 63 |
err_msg = str(e).lower()
|
| 64 |
if "429" in err_msg or "rate limit" in err_msg or "too many requests" in err_msg:
|
| 65 |
wait_time = 20 * (attempt + 1)
|
| 66 |
+
print(f"Rate limit hit! Pausing for {wait_time} seconds before retrying...")
|
| 67 |
time.sleep(wait_time)
|
| 68 |
else:
|
| 69 |
print(f"Agent error processing question: {e}")
|
|
|
|
| 109 |
for i, item in enumerate(questions_data):
|
| 110 |
task_id = item.get("task_id")
|
| 111 |
question_text = item.get("question")
|
| 112 |
+
|
| 113 |
+
# CRITICAL FIX 1: Grab the hidden file URL if the server provides one
|
| 114 |
+
file_url = item.get("file_url")
|
| 115 |
+
|
| 116 |
if not task_id or not question_text:
|
| 117 |
continue
|
| 118 |
|
| 119 |
+
# CRITICAL FIX 2: Inject the file URL into the agent's prompt so it can download it
|
| 120 |
+
if file_url:
|
| 121 |
+
question_text += f"\n\n[IMPORTANT: This task requires analyzing an attached file. You MUST download or read it directly from this URL: {file_url} using your Python tool.]"
|
| 122 |
+
|
| 123 |
+
# CRITICAL FIX 3: Threaten the agent to act like a strict robot to pass the automated grader
|
| 124 |
+
strict_prompt = (
|
| 125 |
+
f"{question_text}\n\n"
|
| 126 |
+
"CRITICAL SUBMISSION INSTRUCTIONS:\n"
|
| 127 |
+
"The system evaluating your answer is a strict automated parser.\n"
|
| 128 |
+
"1. You MUST output ONLY the final requested answer.\n"
|
| 129 |
+
"2. DO NOT include any conversational text, explanations, or reasoning in your final output.\n"
|
| 130 |
+
"3. If the answer is a name, number, or short string, output ONLY that exact string.\n"
|
| 131 |
+
"4. For numbers, do not include symbols unless explicitly requested."
|
| 132 |
+
)
|
| 133 |
+
|
| 134 |
try:
|
| 135 |
+
# We pass the strict prompt instead of the raw question
|
| 136 |
+
submitted_answer = agent(strict_prompt)
|
| 137 |
answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
|
| 138 |
results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": submitted_answer})
|
| 139 |
except Exception as e:
|
|
|
|
| 168 |
|
| 169 |
# --- Build Gradio UI ---
|
| 170 |
with gr.Blocks() as demo:
|
| 171 |
+
gr.Markdown("# Basic Agent Evaluation Runner")
|
| 172 |
gr.Markdown(
|
| 173 |
"""
|
| 174 |
**Instructions:**
|
| 175 |
+
1. Ensure your `NEW_HF_TOKEN` is set.
|
| 176 |
+
2. Log in below.
|
| 177 |
3. Click 'Run Evaluation & Submit' to start.
|
| 178 |
"""
|
| 179 |
)
|