Lint fixes, adjusting qwen2.5 vl prompt

olmocr/bench/prompts.py

@@ -1,5 +1,5 @@
 def build_basic_prompt() -> str:
-    return "Just return the markdown representation of this document as if you were reading it naturally. Convert equations to markdown using \( \) for inline math, and \[ \] otherwise."
+    return "Just return the plain text representation of this document as if you were reading it naturally. Convert equations to LaTeX using \( \) for inline math, and \[ \] otherwise. Convert all tables to markdown."
 
 
 def claude_response_format_schema() -> dict:

olmocr/bench/runners/run_rolmocr.py

@@ -1,5 +1,3 @@
-import json
-from typing import Literal
 
 import httpx
 

olmocr/bench/runners/run_server.py

@@ -45,8 +45,6 @@ async def run_server(
         prompt = build_finetuning_prompt(anchor_text)
     elif prompt_template == "basic":
         prompt = build_basic_prompt()
-    elif prompt_template == "rolmocr":
-        prompt = build_rolmocr_prompt()
     else:
         raise ValueError("Unknown prompt template")
 

olmocr/bench/scripts/convert_all.sh

@@ -260,7 +260,7 @@ pip install --upgrade vllm==0.8.3
 
 
 start_server vllm "Qwen/Qwen2.5-VL-7B-Instruct" --max-model-len 8192
-python -m olmocr.bench.convert --dir "$BENCH_DIR" server:name=qwen25vl_prompt3:model=Qwen/Qwen2.5-VL-7B-Instruct:temperature=0.1:prompt_template=basic:response_template=plain --parallel 50
+python -m olmocr.bench.convert --dir "$BENCH_DIR" server:name=qwen25vl_prompt4:model=Qwen/Qwen2.5-VL-7B-Instruct:temperature=0.1:prompt_template=basic:response_template=plain --parallel 50
 stop_server
 
 start_server vllm "reducto/RolmOCR" --max-model-len 8192

scripts/scan_dolmadocs.py

@@ -7,7 +7,6 @@ import os
 import random
 import re
 import sqlite3
-import string
 import tempfile
 from concurrent.futures import ThreadPoolExecutor
 from pathlib import Path