chore: fix ruff lint warnings and signal handling edge cases

- Fix pre-existing ruff lint warnings (F401, F541, F841, E741)
- Exclude .md/.json/.yml from ruff format check
- Unblock SIGTERM/SIGINT via pthread_sigmask in early signal handler
- Do not restore SIG_DFL on unregister to prevent pending signal kills

scripts/eval/analyze_weights.py

@@ -143,7 +143,7 @@ def print_layer_grid(results: dict[str, dict]):
     widths = [6] + [10] * len(comps)
     metric = "er_99_norm"
 
-    print(f"\n--- Per-Layer Effective Rank (99% energy) ---")
+    print("\n--- Per-Layer Effective Rank (99% energy) ---")
     print(format_header(["Layer"] + comps, widths))
     print("-" * sum(widths))
 
@@ -173,7 +173,7 @@ def print_layer_grid(results: dict[str, dict]):
 def print_weight_stats(results: dict[str, dict]):
     groups = group_by_component(results)
     widths = [20, 12, 12, 12, 12]
-    print(f"\n--- Weight Value Statistics ---")
+    print("\n--- Weight Value Statistics ---")
     print(format_header(["Component", "Mean", "Std", "Min", "Max"], widths))
     print("-" * sum(widths))
 
@@ -265,7 +265,7 @@ def main():
             )
         print(f"{'=' * 70}")
 
-        print(f"Loading weights...")
+        print("Loading weights...")
         sd = safetensors.torch.load_file(str(weights_path))
         print(f"  {len(sd)} keys loaded")
 

scripts/eval/evaluate_mmlu.py

@@ -215,7 +215,6 @@ def _permute_choices(item: dict, rng: random.Random) -> tuple[dict, str]:
     positional bias (e.g. always picking B).
     """
     letters = ("A", "B", "C", "D")
-    contents = [item[k] for k in letters]
     perm = list(letters)
     rng.shuffle(perm)
     permuted = {"question": item["question"]}

scripts/eval/evaluate_ppl.py

@@ -148,7 +148,7 @@ class LossAccumulator:
         self.total += sum(losses)
         self.count += len(losses)
         if self.stream:
-            clamped = [min(max(l, 0.0), self._HIST_MAX) for l in losses]
+            clamped = [min(max(v, 0.0), self._HIST_MAX) for v in losses]
             idx = torch.tensor(clamped) / self._HIST_MAX * (self._HIST_BINS - 1)
             self.hist += torch.bincount(
                 idx.long().clamp(0, self._HIST_BINS - 1),
@@ -315,7 +315,7 @@ def print_stats(label: str, stats: Dict):
             )
     by_type = stats.get("by_token_type", {})
     if by_type:
-        print(f"\n  by token type:")
+        print("\n  by token type:")
         print(f"  {'type':<12} {'count':>8} {'mean_loss':>10} {'ppl':>8}")
         print(f"  {'-' * 12} {'-' * 8} {'-' * 10} {'-' * 8}")
         for ttype, s in by_type.items():