tests to reproduce issue 820

tests/generate/test_integration_llamacpp.py

@@ -247,3 +247,19 @@ def test_llamacpp_cfg(model):
     prompt = "<|im_start|>user\nOutput a short and valid JSON object with two keys.<|im_end|>\n><|im_start|>assistant\n"
     result = generate.cfg(model, grammars.arithmetic)(prompt, seed=11)
     assert isinstance(result, str)
+
+
+@pytest.mark.parametrize(
+    "repo",
+    [
+        ("Qwen/Qwen1.5-0.5B-Chat-GGUF", "*q2*.gguf"),
+        ("QuantFactory/Meta-Llama-3-8B-Instruct-GGUF", "*Q2*.gguf"),
+        ("TheBloke/phi-2-GGUF", "*Q2*.gguf"),
+        ("QuantFactory/Phi-3-mini-4k-instruct-GGUF-v2", "*Q2*.gguf"),
+    ],
+)
+def test_byte_tokenizer_regression(repo, model_path):
+    """Reproduce https://github.com/outlines-dev/outlines/issues/820"""
+    model = models.llamacpp(repo, model_path)
+    generator = generate.choice(model, ["skirt", "dress", "pen", "jacket"])
+    generator("Pick the odd word out: skirt, dress, pen, jacket")