paths

app.py

@@ -1,6 +1,7 @@
 import json
 import os
 from datetime import datetime
+from pathlib import PurePosixPath
 
 from huggingface_hub import HfApi, hf_hub_download
 from shiny import App, reactive, render, ui
@@ -89,8 +90,14 @@ def load_runs(force: bool = False) -> tuple[list, str | None]:
         _cache.update(runs=[], err=str(e))
         return [], str(e)
 
-    result_files = [f for f in all_files if f.startswith("results_") and f.endswith(".json")]
-    sample_files = {f for f in all_files if f.startswith("samples_") and f.endswith(".jsonl")}
+    result_files = [
+        f for f in all_files
+        if PurePosixPath(f).name.startswith("results_") and f.endswith(".json")
+    ]
+    sample_files = {
+        f for f in all_files
+        if PurePosixPath(f).name.startswith("samples_") and f.endswith(".jsonl")
+    }
 
     runs = []
     for filename in sorted(result_files, reverse=True):
@@ -101,7 +108,9 @@ def load_runs(force: bool = False) -> tuple[list, str | None]:
         except Exception:
             continue
 
-        ts = filename.removeprefix("results_").removesuffix(".json")
+        p = PurePosixPath(filename)
+        folder = p.parent.as_posix()  # e.g. "Qwen__Qwen2.5-7B-Instruct"
+        ts = p.name.removeprefix("results_").removesuffix(".json")
         model = data.get("model_name", "unknown")
         date_ts = data.get("date")
         date_str = datetime.fromtimestamp(date_ts).strftime("%Y-%m-%d %H:%M") if date_ts else "—"
@@ -121,7 +130,7 @@ def load_runs(force: bool = False) -> tuple[list, str | None]:
                             "value": v,
                             "stderr": td.get(f"{mname}_stderr,{filt}"),
                         }
-            samples_f = f"samples_{task_name}_{ts}.jsonl"
+            samples_f = f"{folder}/samples_{task_name}_{ts}.jsonl"
             tasks.append({
                 "name": task_name,
                 "n": td.get("sample_len", 0),