[Misc][Quark] Upstream Quark format to VLLM (#10765)

Signed-off-by: kewang-xlnx <kewang@xilinx.com>
Signed-off-by: kewang2 <kewang2@amd.com>
Co-authored-by: kewang2 <kewang2@amd.com>
Co-authored-by: Michael Goin <michael@neuralmagic.com>

tests/quantization/test_quark.py

@@ -0,0 +1,30 @@
+"""Test model set-up and weight loading for quark-quantized models.
+
+Run `pytest tests/quantization/test_quark.py`.
+"""
+
+import torch
+
+from vllm.model_executor.layers.quantization.quark.quark import (  # noqa: E501
+    QuarkLinearMethod, QuarkW8A8Fp8)
+
+
+def test_quark_fp8(vllm_runner):
+    model_path = "amd/Llama-3.1-8B-Instruct-FP8-KV-Quark-test"
+    with vllm_runner(model_path) as llm:
+        model = llm.model.llm_engine.model_executor.driver_worker.model_runner.model  # noqa: E501
+        layer = model.model.layers[0]
+
+        qkv_proj = layer.self_attn.qkv_proj
+
+        assert isinstance(qkv_proj.quant_method, QuarkLinearMethod)
+        assert isinstance(qkv_proj.scheme, QuarkW8A8Fp8)
+
+        if isinstance(qkv_proj.scheme, QuarkW8A8Fp8):
+            assert len(qkv_proj.input_scale.shape) == 0
+            assert qkv_proj.weight.dtype is torch.float8_e4m3fn
+            #assert qkv_proj.weight.dtype is torch.float8_e4m3fnuz
+            assert len(qkv_proj.weight_scale.shape) == 0
+
+        output = llm.generate_greedy("Hello my name is", max_tokens=20)
+        assert output