fix triton cache directory

src/config.py

@@ -1,7 +1,7 @@
 from pathlib import Path
 import pathlib,tempfile
 # Root folder for DB
-#PROJ = Path(__file__).parent # For MAC
+#PROJ = Path(__file__).parent # For MAC
 PROJ = pathlib.Path(tempfile.gettempdir()) # For Space
 MAX_RESULTS  = 10 #default number of results
 

src/helpers.py

@@ -10,10 +10,16 @@ def render_rows(rows):
     """
     blocks = []
     for t, a, txt, pub in rows:
+        # Handle None values by converting them to empty strings
+        title = html.escape(t) if t is not None else ""
+        authors = html.escape(a) if a is not None else ""
+        summary = html.escape(txt) if txt is not None else ""
+        published = pub[:10] if pub is not None else ""
+        
         blocks += [
-            f"<h3>{html.escape(t)}</h3>",
-            f"<p><b>Authors:</b> {html.escape(a)} <br><i>{pub[:10]}</i></p>",
-            f"<pre style='white-space:pre-wrap'>{html.escape(txt)}</pre>",
+            f"<h3>{title}</h3>",
+            f"<p><b>Authors:</b> {authors} <br><i>{published}</i></p>",
+            f"<pre style='white-space:pre-wrap'>{summary}</pre>",
             "<hr>"
         ]
     return "\n".join(blocks) or "<p>No matching papers found.</p>"

src/summarise.py

@@ -2,7 +2,9 @@ from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
 from db import get_conn
 from config import MODEL_NAME
 from helpers    import rows_by_tag 
-
+import os
+import tempfile
+import pathlib
 """ 
 Summarise the abstract of a paper using a LLM. Further versions should instead summarise the full paper.
 """
@@ -16,12 +18,14 @@ PROMPT = (
 
 # ---------------------------------------------------------------------- #
 def load_pipe():
+    cache_dir = pathlib.Path(tempfile.gettempdir()) / "hf_cache"
     model = AutoModelForCausalLM.from_pretrained(
         MODEL_NAME, 
-        #load_in_4bit=True, 
+        cache_dir=cache_dir,
+        load_in_4bit=True, 
         device_map="auto"
     )
-    tok   = AutoTokenizer.from_pretrained(MODEL_NAME)
+    tok   = AutoTokenizer.from_pretrained(MODEL_NAME, cache_dir=cache_dir)
     tok.pad_token = tok.eos_token
     return pipeline(
         "text-generation",