Adjust the error return value and markdown marigin style

app.py

@@ -1,31 +1,19 @@
 import gradio as gr
-from huggingface_hub import login
 
-import re
-
-# from vllm import LLM, SamplingParams
-import pandas as pd
 from dataclasses import dataclass
 from concurrent.futures import ThreadPoolExecutor, TimeoutError
-import os
-from typing import Dict, Any, List
 
-# code execution
 import os
 import re
-import signal
 import subprocess
 import tempfile
-from contextlib import contextmanager
-from typing import Tuple
-from tqdm import tqdm
-import time
+import json
+from typing import Tuple, Dict, Any, List
 from sympy import N, simplify
 from sympy.parsing.latex import parse_latex
-import random
-from pathlib import Path
 from openai import OpenAI
 
+
 client = OpenAI(
     base_url=os.environ.get("SERVER_URL"),
     api_key=os.environ.get("HF_TOKEN"),
@@ -442,15 +430,6 @@ def validate_answer_is_numeric(x: str | int | float) -> int:
     return x
 
 
-def get_majority_vote(responses: List[int]) -> int:
-    if len(responses) < 1:
-        return 0
-    else:
-        c = Counter(responses)
-        value, count = c.most_common()[0]
-        return value
-
-
 def filter_answers(answers: List[str]) -> List[int]:
     formatted_answers = [validate_answer_is_numeric(a) for a in answers]
 
@@ -540,17 +519,45 @@ print(f"=== Running submission with config ===\n\n{config}")
 
 
 def generate(message, temperature):
-    chat_completion = client.chat.completions.create(
+    """
+    Generates a chat completion response by streaming data from the client chat model.
+
+    This function streams the response from the client chat model and yields the content
+    of the response chunk by chunk. If an error occurs, it yields the error message.
+
+    Parameters:
+    message (str): The input message to be sent to the chat model.
+    temperature (float): The sampling temperature to use. Higher values mean the model will take more risks.
+
+    Yields:
+    tuple: A tuple containing the content of the response and a boolean flag indicating if an error occurred.
+           If no error occurred, the boolean flag will be False and the content will be the response text.
+           If an error occurred, the boolean flag will be True and the content will be the error message.
+    """
+    stream = client.chat.completions.create(
         model="tgi",
         messages=message,
         stream=True,
-        max_tokens=1024,
+        max_tokens=1200,
         stop=["```output\n"],
         temperature=temperature,
     )
 
-    for message in chat_completion:
-        yield message.choices[0].delta.content
+    response = stream.response
+
+    # The reason why the library method is not used here is that if an error occurs,
+    #    the returned data will not be a stream, and using the official library will result in an error.
+    for chunk in response.iter_bytes():
+        chunk = chunk.decode("utf-8")
+        chune_json = json.loads(chunk.replace("data:", ""))
+
+        if "error" in chune_json and chune_json["error"]:
+            yield chune_json["error"], True
+            break
+
+        content = chune_json["choices"][0]["delta"]["content"]
+        if content is not None:
+            yield content, False
 
 
 def get_majority_text(data):
@@ -645,10 +652,6 @@ def process_code(
     return example
 
 
-# load the vllm instance and set sampling parameters
-# vllm = build_vllm(config)
-
-
 def solve_problem(problem, temperature, progress=gr.Progress()):
     problem = apply_template({"prompt": problem}, prompt=config.system_prompt)
     print(f"Problem: {problem}")
@@ -656,8 +659,8 @@ def solve_problem(problem, temperature, progress=gr.Progress()):
     sample = {
         "problem": problem,  # not used for the submission TODO Remove
         "ground_truth": "unknown",  # not used for the submission TODO Remove
-        "text": "### Solution:\n",
-        "gen_texts": "### Solution:\n",  # used to store all the generated text
+        "text": "## Solution:\n",
+        "gen_texts": "## Solution:\n",  # used to store all the generated text
         "should_prune": False,
         "problem_index": -1,  # not used for the submission TODO Remove
         "model_answers": "-1",
@@ -676,11 +679,14 @@ def solve_problem(problem, temperature, progress=gr.Progress()):
             {"role": "assistant", "content": sample["gen_texts"]},
         ]
 
-        for reponse_message in generate(messages, temperature):
+        for reponse_message, error in generate(messages, temperature):
             if reponse_message is not None:
                 step_reponse += reponse_message
                 yield step_reponse
 
+                if error:
+                    return
+
         sample["gen_texts"] = step_reponse
 
         # TODO: Maybe it should just return the result of running the code
@@ -705,13 +711,13 @@ def solve_problem(problem, temperature, progress=gr.Progress()):
     yield sample["gen_texts"]
 
 
-with gr.Blocks() as demo:
+with gr.Blocks(css=".top-margin { margin-top: 20px; }") as demo:
     with gr.Row():
         inp = gr.Textbox(placeholder="Problem", label="Problem", lines=5)
     with gr.Accordion("Advanced Options", open=False):
-        temperature = gr.Slider(minimum=0.0, maximum=1.0, value=0.8, step=0.1, label="Temperature")
+        temperature = gr.Slider(minimum=0.0, maximum=1.0, value=0.1, step=0.1, label="Temperature")
     with gr.Row():
-        out = gr.Markdown()
+        out = gr.Markdown(elem_classes=["top-margin"])
 
     btn = gr.Button("Run")