Encode generation prompt via tokenizer.encode (apply_chat_template returns Encodings)

2026-09-23 22:36:08 -05:00
parent fcefb971f8
commit f5419183bc
1 changed files with 3 additions and 2 deletions
@@ -112,12 +112,13 @@ class SemIfEngine:
            import numpy

            engine = model.engine
-            prompt_ids = tokenizer.apply_chat_template(
+            prompt_text = tokenizer.apply_chat_template(
                messages,
-                tokenize=True,
+                tokenize=False,
                add_generation_prompt=True,
                enable_thinking=False,
            )
+            prompt_ids = tokenizer.encode(prompt_text, add_special_tokens=False)
            engine.clear()
            logits = engine._decode(prompt_ids, 0, 0, True)
            generated: list[int] = []