Johannes Janosovits commited on
Commit
18160d5
Β·
1 Parent(s): cef2bfc
Files changed (1) hide show
  1. app.py +24 -20
app.py CHANGED
@@ -58,9 +58,10 @@ tr.run-row.active td { background: var(--bg3); }
58
  _cache: dict = {"runs": None, "err": None}
59
 
60
 
61
- def _summarize(groups: dict, tasks: list) -> str:
62
- # Prefer group-level aggregates (e.g. "blimp: 0.778")
63
- parts = []
 
64
  for g_name, g_data in groups.items():
65
  if not isinstance(g_data, dict):
66
  continue
@@ -69,32 +70,31 @@ def _summarize(groups: dict, tasks: list) -> str:
69
  continue
70
  mname = k.split(",")[0]
71
  if not mname.endswith("_stderr"):
72
- parts.append(f"{g_name}: {v:.3f}")
 
73
  break
74
- if parts:
75
- return "; ".join(parts)
76
 
77
  if not tasks:
78
- return "β€”"
79
 
80
- # Few tasks: show each by name so the benchmark is identifiable
81
  if len(tasks) <= 5:
82
- task_parts = []
83
  for t in tasks:
 
84
  first_m = next(iter(t["metrics"]), None)
85
- if first_m:
86
- task_parts.append(f"{t['name']}: {t['metrics'][first_m]['value']:.3f}")
87
- else:
88
- task_parts.append(t["name"])
89
- return "; ".join(task_parts)
90
 
91
- # Many tasks without a group: avg of the primary metric
92
  first_m = next(iter(tasks[0]["metrics"]), None)
93
  if first_m:
94
  vals = [t["metrics"][first_m]["value"] for t in tasks if first_m in t["metrics"]]
95
  if vals:
96
- return f"avg {first_m}: {sum(vals)/len(vals):.3f} (n={len(tasks)})"
97
- return "β€”"
98
 
99
 
100
  def load_runs(force: bool = False) -> tuple[list, str | None]:
@@ -156,6 +156,7 @@ def load_runs(force: bool = False) -> tuple[list, str | None]:
156
  "samples_file": samples_f if samples_f in sample_files else None,
157
  })
158
 
 
159
  runs.append({
160
  "model": model,
161
  "date": date_str,
@@ -163,7 +164,8 @@ def load_runs(force: bool = False) -> tuple[list, str | None]:
163
  "ts": ts,
164
  "tasks": tasks,
165
  "n_tasks": len(tasks),
166
- "summary": _summarize(data.get("groups", {}), tasks),
 
167
  })
168
 
169
  _cache.update(runs=runs, err=None)
@@ -238,7 +240,8 @@ def server(input, output, session):
238
  ui.tags.td(ui.span({"class": "model"}, run["model"])),
239
  ui.tags.td(run["date"]),
240
  ui.tags.td(str(run["n_tasks"])),
241
- ui.tags.td(run["summary"]),
 
242
  ui.tags.td(
243
  ui.tags.a(
244
  {"class": "a", "href": f"{HF_BLOB_URL}/{run['filename']}",
@@ -255,7 +258,8 @@ def server(input, output, session):
255
  ui.tags.th("Model"),
256
  ui.tags.th("Date"),
257
  ui.tags.th("Tasks"),
258
- ui.tags.th("Summary"),
 
259
  ui.tags.th("File"),
260
  )
261
  ),
 
58
  _cache: dict = {"runs": None, "err": None}
59
 
60
 
61
+ def _names_and_metrics(groups: dict, tasks: list) -> tuple[str, str]:
62
+ """Returns (name_str, metrics_str) for the two summary columns."""
63
+ # Prefer group-level aggregates
64
+ g_names, g_metrics = [], []
65
  for g_name, g_data in groups.items():
66
  if not isinstance(g_data, dict):
67
  continue
 
70
  continue
71
  mname = k.split(",")[0]
72
  if not mname.endswith("_stderr"):
73
+ g_names.append(g_name)
74
+ g_metrics.append(f"{v:.3f}")
75
  break
76
+ if g_names:
77
+ return "; ".join(g_names), "; ".join(g_metrics)
78
 
79
  if not tasks:
80
+ return "β€”", "β€”"
81
 
82
+ # Few tasks: name column lists task names, metrics column lists values
83
  if len(tasks) <= 5:
84
+ names, metrics = [], []
85
  for t in tasks:
86
+ names.append(t["name"])
87
  first_m = next(iter(t["metrics"]), None)
88
+ metrics.append(f"{t['metrics'][first_m]['value']:.3f}" if first_m else "β€”")
89
+ return "; ".join(names), "; ".join(metrics)
 
 
 
90
 
91
+ # Many tasks without a group
92
  first_m = next(iter(tasks[0]["metrics"]), None)
93
  if first_m:
94
  vals = [t["metrics"][first_m]["value"] for t in tasks if first_m in t["metrics"]]
95
  if vals:
96
+ return f"({len(tasks)} tasks)", f"avg {first_m}: {sum(vals)/len(vals):.3f}"
97
+ return f"({len(tasks)} tasks)", "β€”"
98
 
99
 
100
  def load_runs(force: bool = False) -> tuple[list, str | None]:
 
156
  "samples_file": samples_f if samples_f in sample_files else None,
157
  })
158
 
159
+ task_names, metrics = _names_and_metrics(data.get("groups", {}), tasks)
160
  runs.append({
161
  "model": model,
162
  "date": date_str,
 
164
  "ts": ts,
165
  "tasks": tasks,
166
  "n_tasks": len(tasks),
167
+ "task_names": task_names,
168
+ "metrics": metrics,
169
  })
170
 
171
  _cache.update(runs=runs, err=None)
 
240
  ui.tags.td(ui.span({"class": "model"}, run["model"])),
241
  ui.tags.td(run["date"]),
242
  ui.tags.td(str(run["n_tasks"])),
243
+ ui.tags.td(run["task_names"]),
244
+ ui.tags.td(run["metrics"]),
245
  ui.tags.td(
246
  ui.tags.a(
247
  {"class": "a", "href": f"{HF_BLOB_URL}/{run['filename']}",
 
258
  ui.tags.th("Model"),
259
  ui.tags.th("Date"),
260
  ui.tags.th("Tasks"),
261
+ ui.tags.th("Name"),
262
+ ui.tags.th("Metrics"),
263
  ui.tags.th("File"),
264
  )
265
  ),