chore: import upstream snapshot with attribution

tests/python/loader/test_awq.py

@@ -0,0 +1,39 @@
+from pathlib import Path
+from typing import Union
+
+import pytest
+import tvm
+
+from mlc_llm.loader import HuggingFaceLoader
+from mlc_llm.model import MODEL_PRESETS, MODELS
+from mlc_llm.quantization import QUANTIZATION
+from mlc_llm.support import logging, tqdm
+
+logging.enable_logging()
+
+
+@pytest.mark.parametrize(
+    "param_path",
+    [
+        "./dist/models/llama-2-7b-w4-g128-awq.pt",
+        "./dist/models/Llama-2-7B-AWQ/model.safetensors",
+    ],
+)
+def test_load_llama(param_path: Union[str, Path]):
+    path_params = Path(param_path)
+
+    model = MODELS["llama"]
+    quantization = QUANTIZATION["q4f16_awq"]
+    config = model.config.from_dict(MODEL_PRESETS["llama2_7b"])
+    loader = HuggingFaceLoader(
+        path=path_params,
+        extern_param_map=model.source["awq"](config, quantization),
+    )
+    with tqdm.redirect():
+        for _name, _param in loader.load(tvm.device("cpu")):
+            ...
+
+
+if __name__ == "__main__":
+    test_load_llama(param_path="./dist/models/llama-2-7b-w4-g128-awq.pt")
+    test_load_llama(param_path="./dist/models/Llama-2-7B-AWQ/model.safetensors")

tests/python/loader/test_huggingface.py

@@ -0,0 +1,68 @@
+from pathlib import Path
+from typing import Union
+
+import pytest
+import tvm
+
+from mlc_llm.loader import HuggingFaceLoader
+from mlc_llm.model import MODELS
+from mlc_llm.support import logging, tqdm
+
+logging.enable_logging()
+
+
+@pytest.mark.parametrize(
+    "base_path",
+    [
+        "./dist/models/Llama-2-7b-hf",
+        "./dist/models/Llama-2-13b-hf",
+        "./dist/models/Llama-2-70b-hf",
+    ],
+)
+def test_load_torch_llama(base_path: Union[str, Path]):
+    base_path = Path(base_path)
+    path_config = base_path / "config.json"
+    path_params = base_path / "pytorch_model.bin.index.json"
+
+    model = MODELS["llama"]
+    config = model.config.from_file(path_config)
+    loader = HuggingFaceLoader(
+        path=path_params,
+        extern_param_map=model.source["huggingface-torch"](config, None),
+    )
+    with tqdm.redirect():
+        for _name, _param in loader.load(device=tvm.device("cpu")):
+            return  # To reduce the time of the test
+
+
+@pytest.mark.parametrize(
+    "base_path",
+    [
+        "./dist/models/Llama-2-7b-hf",
+        "./dist/models/Llama-2-13b-hf",
+        "./dist/models/Llama-2-70b-hf",
+    ],
+)
+def test_load_safetensor_llama(base_path: Union[str, Path]):
+    base_path = Path(base_path)
+    path_config = base_path / "config.json"
+    path_params = base_path / "model.safetensors.index.json"
+
+    model = MODELS["llama"]
+    config = model.config.from_file(path_config)
+    loader = HuggingFaceLoader(
+        path=path_params,
+        extern_param_map=model.source["huggingface-safetensor"](config, None),
+    )
+    with tqdm.redirect():
+        for _name, _param in loader.load(device=tvm.device("cpu")):
+            return  # To reduce the time of the test
+
+
+if __name__ == "__main__":
+    test_load_torch_llama(base_path="./dist/models/Llama-2-7b-hf")
+    test_load_torch_llama(base_path="./dist/models/Llama-2-13b-hf")
+    test_load_torch_llama(base_path="./dist/models/Llama-2-70b-hf")
+    test_load_safetensor_llama(base_path="./dist/models/Llama-2-7b-hf")
+    test_load_safetensor_llama(base_path="./dist/models/Llama-2-13b-hf")
+    test_load_safetensor_llama(base_path="./dist/models/Llama-2-70b-hf")