Update to new format

app.py

@@ -43,24 +43,19 @@ demo = gr.Blocks(css=".typewriter {font-family: 'JMH Typewriter', sans-serif;}")
 runs: dict[int, list[Run]] = {}
 
 
-
 @dataclass
 class LeaderboardEntry:
     uid: int
+    winner: bool
     model: str
     score: float
     similarity: float
     hotkey: str
-    rank: int
     baseline_generation_time: float
-    baseline_size: int
-    baseline_vram_used: float
-    baseline_watts_used: float
     generation_time: float
     size: int
     vram_used: float
     watts_used: float
-    previous_day_winner: bool
 
 
 @dataclass
@@ -94,7 +89,7 @@ def is_valid_run(run: Run):
 
 
 def calculate_score(baseline_generation_time: float, generation_time: float, similarity_score: float) -> float:
-    return max(0.0, baseline_generation_time - generation_time) * similarity_score
+    return (baseline_generation_time - generation_time) * similarity_score
 
 
 def get_graph_entries(runs: list[Run]) -> dict[int, GraphEntry]:
@@ -116,7 +111,7 @@ def get_graph_entries(runs: list[Run]) -> dict[int, GraphEntry]:
                 generation_time = value["generation_time"]
                 similarity = min(1, value["similarity"])
                 score = calculate_score(baseline_generation_time, generation_time, similarity)
-                model = value["model"]
+                model = run.summary["submissions"][str(uid)]["repository"]
 
                 if uid not in entries:
                     entries[uid] = GraphEntry([date], [baseline_generation_time], [generation_time], [similarity], [score], [model], generation_time)
@@ -191,51 +186,40 @@ def create_leaderboard(runs: list[Run]) -> list[tuple]:
             for key, value in summary_value.items():
                 has_data = True
 
-                try:
-                    uid = int(key)
-                    generation_time = value["generation_time"]
-                    baseline_generation_time = value["baseline_generation_time"]
-                    similarity = min(1, value["similarity"])
-
-                    entries[uid] = LeaderboardEntry(
-                        uid=uid,
-                        model=value["model"],
-                        score=calculate_score(baseline_generation_time, generation_time, similarity),
-                        similarity=similarity,
-                        hotkey=value["hotkey"],
-                        rank=value["rank"],
-                        baseline_generation_time=baseline_generation_time,
-                        baseline_size=value["baseline_size"],
-                        baseline_vram_used=value["baseline_vram_used"],
-                        baseline_watts_used=value["baseline_watts_used"],
-                        generation_time=generation_time,
-                        size=value["size"],
-                        vram_used=value["vram_used"],
-                        watts_used=value["watts_used"],
-                        previous_day_winner=value["multiday_winner"],
-                    )
-                except Exception:
-                    continue
+                uid = int(key)
+                generation_time = value["generation_time"]
+                baseline_generation_time = value["baseline_generation_time"]
+                similarity = min(1, value["similarity"])
+
+                entries[uid] = LeaderboardEntry(
+                    uid=uid,
+                    winner="winner" in value,
+                    model=run.summary["submissions"][str(uid)]["repository"],
+                    score=calculate_score(baseline_generation_time, generation_time, similarity),
+                    similarity=similarity,
+                    baseline_generation_time=baseline_generation_time,
+                    generation_time=generation_time,
+                    size=value["size"],
+                    vram_used=value["vram_used"],
+                    watts_used=value["watts_used"],
+                    hotkey=value["hotkey"],
+                )
 
         if has_data:
             break
 
-    leaderboard: list[tuple] = [
-        (entry.rank + 1,
-         entry.uid,
-         entry.model,
-         entry.score,
-         f"{entry.generation_time:.3f}s",
-         f"{entry.similarity:.3f}",
-         f"{entry.size / 1_000_000_000:.3f}GB",
-         f"{entry.vram_used / 1_000_000_000:.3f}GB",
-         f"{entry.watts_used:.3f}W",
-         entry.hotkey,
-         entry.previous_day_winner
-         ) for entry in sorted(entries.values(), key=lambda entry: (entry.score, entry.rank), reverse=True)
-    ]
-
-    return leaderboard
+    return [(
+        entry.uid,
+        f"<span style='color: {'springgreen' if entry.winner else 'red'}'>{entry.winner}</span>",
+        entry.model,
+        round(entry.score, 3),
+        f"{entry.generation_time:.3f}s",
+        f"{entry.similarity:.3f}",
+        f"{entry.size / 1_000_000_000:.3f}GB",
+        f"{entry.vram_used / 1_000_000_000:.3f}GB",
+        f"{entry.watts_used:.3f}W",
+        entry.hotkey,
+    ) for entry in sorted(entries.values(), key=lambda entry: (entry.winner, entry.score), reverse=True)]
 
 
 def get_run_validator_uid(run: Run) -> int:
@@ -251,7 +235,7 @@ def fetch_wandb_data():
         order="-created_at",
     )
 
-    wandb_runs = [run for run in wandb_runs if len(dict(run.summary)) > 1]
+    wandb_runs = [run for run in wandb_runs if "benchmarks" in run.summary]
 
     global runs
     runs.clear()
@@ -290,6 +274,7 @@ def refresh():
     fetch_wandb_data()
     demo.clear()
     now = datetime.now(tz=ZoneInfo("America/New_York"))
+
     with demo:
         gr.Image(
             "cover.png",
@@ -321,11 +306,9 @@ def refresh():
 
             leaderboard = gr.components.Dataframe(
                 create_leaderboard(runs[parse_uid(dropdown.value)]),
-                headers=["Rank", "Uid", "Model", "Score", "Generation Time", "Similarity", "Size", "VRAM Usage", "Power Usage", "Hotkey", "Previous day winner"],
-                datatype=["number", "number", "markdown", "number", "markdown", "number", "markdown", "markdown", "markdown", "markdown", "markdown"],
+                headers=["Uid", "Winner", "Model", "Score", "Gen Time", "Similarity", "Size", "VRAM Usage", "Power Usage", "Hotkey"],
+                datatype=["number", "markdown", "markdown", "number", "markdown", "number", "markdown", "markdown", "markdown", "markdown"],
                 elem_id="leaderboard-table",
-                interactive=False,
-                visible=True,
             )
 
             demo.load(lambda uid: create_graph(runs[parse_uid(uid)]), [dropdown], [graph])