Strip size suffixes from model names before arXiv search

Queries like all:"OLMo 2 32B" return no results because papers use the
base name ("OLMo 2"), not the size-qualified form. A regex now strips
trailing parameter/context suffixes (7B, 32B, 1.4B, 4K, 160M) while
leaving MoE names like "Mixtral 8x7B" intact.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: erwardenaar

src/openllm_selector/database.py

@@ -2,6 +2,7 @@
 
 import json
 import pathlib
+import re
 import xml.etree.ElementTree as ET
 
 import requests
@@ -508,10 +509,14 @@ def fetch_recent_papers(model_name: str, max_results: int = 3) -> list[dict]:
     >>> papers[0].keys()
     dict_keys(['title', 'authors', 'summary', 'published', 'arxiv_url'])
     """
+    # Strip trailing size/context suffixes (e.g. "7B", "32B", "1.4B", "4K") so
+    # "OLMo 2 32B" searches as "OLMo 2". Mixture-of-experts names like "8x7B"
+    # are kept because they are part of the canonical model identity.
+    search_name = re.sub(r"\s+\d+(\.\d+)?[BbMmKk]$", "", model_name)
     response = requests.get(
         "https://export.arxiv.org/api/query",
         params={
-            "search_query": f'all:"{model_name}"',
+            "search_query": f'all:"{search_name}"',
             "sortBy": "submittedDate",
             "sortOrder": "descending",
             "max_results": max_results,