fix: improve prompt generation logging and add padding token ID

app_nobatching.py

@@ -94,7 +94,6 @@ def generate(submission: list[dict[str, str]], team_id: str) -> list[dict[str, s
 
     # Format prompts using chat template
     for i, prompt in enumerate(prompts):
-        print(f"Generating response for prompt {i + 1}/{len(prompts)}")
         start_time = time.perf_counter()
 
         messages = [{"role": "user", "content": prompt}]
@@ -111,6 +110,7 @@ def generate(submission: list[dict[str, str]], team_id: str) -> list[dict[str, s
                 do_sample=False,
                 temperature=None,
                 repetition_penalty=REPETITION_PENALTY,
+                pad_token_id=chat_tokenizer.pad_token_id,
                 eos_token_id=chat_tokenizer.eos_token_id,
             )
 
@@ -121,7 +121,7 @@ def generate(submission: list[dict[str, str]], team_id: str) -> list[dict[str, s
 
         generation_time = time.perf_counter() - start_time
         times.append(generation_time)
-        print(f"✓ Prompt {i + 1} generated in {generation_time:.3f}s")
+        print(f"✓ Prompt {i + 1:02}/25 generated in {generation_time:.3f}s")
 
     print(
         f"✓ Generation completed: {sum(times):.3f}s ({len(prompts) / sum(times):.1f} prompts/s) (average {sum(times) / len(prompts):.3f}s per prompt)"