Upload folder using huggingface_hub

code/aubin_mini/__init__.py

@@ -5,7 +5,7 @@ and the option scores of a question are turned into probabilities with a fitted
 Runs on CPU with ONNX Runtime only (no torch, no GPU).
 
     from aubin_mini import AubinMini
-    m = AubinMini.from_pretrained("emrevrg/AUBIN-Mini")          # phone build, 255 MB
+    m = AubinMini.from_pretrained("emrevrg/AUBIN-Mini")          # phone build, 267 MB
     m.decide({"ticket": "Package arrived broken, customer wants money back"},
              {"route": {"type": "choice", "instructions": "Which team?",
                         "criteria": {"billing": "refunds", "shipping": "delivery", "tech": "software"}},
@@ -20,8 +20,8 @@ __version__ = "0.1.0"
 __all__ = ["AubinMini", "options_of"]
 
 VARIANTS = {                      # file name, size on disk, accuracy on the same 300 Kev development questions
-    "phone": "aubin_mini_nb4_g8.onnx",   # 4-bit block weights + 8-bit embeddings, 255 MB, dev 0.833
-    "nb4": "aubin_mini_nb4.onnx",        # 4-bit block weights, 402 MB, dev 0.840
+    "phone": "aubin_mini_nb4_g8.onnx",   # 4-bit block weights + 8-bit embeddings, 267 MB, locked Kev test 0.847
+    "nb4": "aubin_mini_nb4.onnx",        # 4-bit block weights, 422 MB, locked Kev test 0.846
 }
 
 
@@ -113,8 +113,10 @@ class AubinMini:
 
     def choose(self, question, options, state=""):
         """options: dict {key: description} or list of texts → {answer, confidence, probabilities}."""
-        q = {"type": "choice", "criteria": options}
-        keys, texts = options_of(q)
+        if isinstance(options, (list, tuple)):              # plain list: the answer is the option text itself
+            keys = texts = [str(o) for o in options]
+        else:
+            keys, texts = options_of({"type": "choice", "criteria": options})
         return self._pack(keys, self.log_probs(question, texts, state))
 
     def decide(self, state, questions):