fix bug when resuming generation with OpenAI models

guidance/models/_openai.py

@@ -257,6 +257,9 @@ def _generator(self, prompt, temperature):
                         assert (
                             role_name == "assistant"
                         ), "Bad chat format! Last role before gen needs to be assistant!"
+                        btext = prompt[pos:]
+                        pos = len(prompt)
+                        messages.append({"role": role_name, "content": btext.decode("utf8")})
                         break
                     btext = prompt[pos : pos + end_pos]
                     pos += end_pos + len(role_end)

tests/models/test_openai.py

@@ -106,3 +106,21 @@ def test_openai_chat_loop():
 
         with assistant():
             lm += gen(name="answer", max_tokens=2)
+
+
+def test_openai_prefill():
+    try:
+        model = guidance.models.OpenAI("gpt-3.5-turbo", echo=False)
+    except:
+        pytest.skip("Skipping OpenAI test because we can't load the model!")
+
+    with system():
+        lm = model + "You are a helpful assistant."
+
+    with user():
+        lm += "What is the capital of France?"
+
+    with assistant():
+        lm += "The capital of France is " + gen(name="capital", suffix=".")
+
+    assert lm["capital"] == "Paris"