chore: import upstream snapshot with attribution

tests/quantization/test_configs.py

@@ -0,0 +1,77 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+"""Tests whether Marlin models can be loaded from the autogptq config.
+
+Run `pytest tests/quantization/test_configs.py --forked`.
+"""
+
+from dataclasses import dataclass
+
+import pytest
+
+from vllm.config import ModelConfig
+from vllm.platforms import current_platform
+
+
+@dataclass
+class ModelPair:
+    model_marlin: str
+    model_gptq: str
+
+
+# Model Id // Quantization Arg // Expected Type
+MODEL_ARG_EXPTYPES = [
+    # AUTOGPTQ
+    # compat: autogptq <=0.7.1 is_marlin_format: bool
+    # Model Serialized in Exllama Format.
+    ("TheBloke/Llama-2-7B-Chat-GPTQ", None, "auto_gptq"),
+    (
+        "TheBloke/Llama-2-7B-Chat-GPTQ",
+        "marlin",
+        "auto_gptq" if current_platform.is_cuda_alike() else "ERROR",
+    ),
+    ("TheBloke/Llama-2-7B-Chat-GPTQ", "gptq", "auto_gptq"),
+    ("TheBloke/Llama-2-7B-Chat-GPTQ", "awq", "ERROR"),
+    # compat: autogptq >=0.8.0 use checkpoint_format: str
+    # Model Serialized in Exllama Format.
+    ("LnL-AI/TinyLlama-1.1B-Chat-v1.0-GPTQ-4bit", None, "auto_gptq"),
+    (
+        "LnL-AI/TinyLlama-1.1B-Chat-v1.0-GPTQ-4bit",
+        "marlin",
+        "auto_gptq" if current_platform.is_cuda_alike() else "ERROR",
+    ),
+    ("LnL-AI/TinyLlama-1.1B-Chat-v1.0-GPTQ-4bit", "gptq", "auto_gptq"),
+    ("LnL-AI/TinyLlama-1.1B-Chat-v1.0-GPTQ-4bit", "awq", "ERROR"),
+    # AUTOAWQ
+    # AutoAWQConfig.override_quantization_method() returns "auto_awq" for AWQ models
+    # when user_quant is None, "awq", "awq_marlin", "marlin", or "auto_awq"
+    (
+        "TheBloke/OpenHermes-2.5-Mistral-7B-AWQ",
+        None,
+        "auto_awq",
+    ),
+    ("TheBloke/OpenHermes-2.5-Mistral-7B-AWQ", "awq", "auto_awq"),
+    (
+        "TheBloke/OpenHermes-2.5-Mistral-7B-AWQ",
+        "marlin",
+        "auto_awq" if current_platform.is_cuda_alike() else "ERROR",
+    ),
+    ("TheBloke/OpenHermes-2.5-Mistral-7B-AWQ", "gptq", "ERROR"),
+]
+
+
+@pytest.mark.parametrize("model_arg_exptype", MODEL_ARG_EXPTYPES)
+def test_auto_gptq(model_arg_exptype: tuple[str, None, str]) -> None:
+    model_path, quantization_arg, expected_type = model_arg_exptype
+
+    try:
+        model_config = ModelConfig(model_path, quantization=quantization_arg)
+        found_quantization_type = model_config.quantization
+    except ValueError:
+        found_quantization_type = "ERROR"
+
+    assert found_quantization_type == expected_type, (
+        f"Expected quant_type == {expected_type} for {model_path}, "
+        f"but found {found_quantization_type} "
+        f"for no --quantization {quantization_arg} case"
+    )