fix(api-space): force Whisper decode scores via GenerationConfig; strip dangling grazie tail after .?!

HF Whisper ignored output_scores when return_dict_in_generate stayed false on the
config; deep-copy generation_config and set both flags.

Extend dangling Italian tail cleanup to drop hallucinated GRAZIE after sentence
punctuation only (not standalone grazie or comma-led grazie).

Co-authored-by: Cursor <cursoragent@cursor.com>

main.py

@@ -10,6 +10,7 @@ Response JSON: {"text": "......."}
 
 from __future__ import annotations
 
+import copy
 import math
 import os
 import re
@@ -440,12 +441,19 @@ def _clean_text_artifacts(text: str) -> str:
 
 
 def _strip_dangling_italian_tail(text: str) -> str:
-    """Remove only a final dangling Italian conjunction token (E/ED)."""
+    """Remove only lightweight dangling tails (no broad phrase blocklists).
+
+    - Final lone conjunction ``E`` / ``ED`` (after punctuation or EOS).
+    - Final standalone ``GRAZIE`` / ``GRAZIE.`` only after sentence-ending ``.?!``
+      (Whisper often hallucinates a polite tail after a complete clause — not injecting text).
+    """
     t = (text or "").strip()
     if not t:
         return ""
     t = re.sub(r"(?i)([.!?])\s+(e|ed)\s*$", r"\1", t)
     t = re.sub(r"(?i)\s+(e|ed)\s*$", "", t)
+    # "? grazie." / "... grazie" hallucination tail — keep bare "GRAZIE" or "SÌ, GRAZIE" untouched.
+    t = re.sub(r"(?i)([.!?])\s+GRAZIE\.?\s*$", r"\1", t)
     return t.strip()
 
 
@@ -1023,9 +1031,24 @@ class WhisperASRSingle:
             if logprob_threshold is not None and no_speech_threshold is not None:
                 gen_common["logprob_threshold"] = logprob_threshold
 
-            meta_kw = {"return_dict_in_generate": True, "output_scores": True}
+            # Whisper may ignore kwargs unless applied on GenerationConfig; without this,
+            # output_scores is dropped and avg_logprob stays None (see transformers warning).
+            try:
+                gen_cfg = copy.deepcopy(model.generation_config)
+                gen_cfg.return_dict_in_generate = True
+                gen_cfg.output_scores = True
+            except Exception:
+                gen_cfg = None
+
             try:
-                out = self._generate_ids(model, input_features, {**gen_common, **meta_kw})
+                if gen_cfg is not None:
+                    out = self._generate_ids(model, input_features, {**gen_common, "generation_config": gen_cfg})
+                else:
+                    out = self._generate_ids(
+                        model,
+                        input_features,
+                        {**gen_common, "return_dict_in_generate": True, "output_scores": True},
+                    )
             except TypeError:
                 try:
                     out = self._generate_ids(model, input_features, dict(gen_common))