[Feature][Quant] Auto-detect quantization format from model files (#6645)

## Summary

- Add automatic quantization format detection, eliminating the need to
manually specify `--quantization` when serving quantized models.
- The detection inspects only lightweight JSON files
(`quant_model_description.json` and `config.json`) at engine
initialization time, with no `.safetensors` reads.
- User-explicit `--quantization` flags are always respected;
auto-detection only applies when the flag is omitted.

## Details

**Detection priority:**
1. `quant_model_description.json` exists → `quantization="ascend"`
(ModelSlim)
2. `config.json` contains `"quant_method": "compressed-tensors"` →
`quantization="compressed-tensors"` (LLM-Compressor)
3. Neither → default float behavior

**Technical approach:**
Hooked into `NPUPlatform.check_and_update_config()` to run detection
after `VllmConfig.__post_init__`. Since `quant_config` is already `None`
at that point, we explicitly recreate it via
`VllmConfig._get_quantization_config()` to trigger the full quantization
initialization pipeline.

## Files Changed

| File | Description |
|------|-------------|
| `vllm_ascend/quantization/utils.py` | Added
`detect_quantization_method()` and `maybe_auto_detect_quantization()` |
| `vllm_ascend/platform.py` | Integrated auto-detection in
`check_and_update_config()` |
| `vllm_ascend/quantization/modelslim_config.py` | Improved error
handling for weight loading |
- vLLM version: v0.15.0
- vLLM main:
d7e17aaacd

---------

Signed-off-by: SlightwindSec <slightwindsec@gmail.com>

tests/ut/test_platform.py

@@ -116,13 +116,14 @@ class TestNPUPlatform(TestBase):
         self.assertIsNone(self.platform.inference_mode())
         mock_inference_mode.assert_called_once()
 
+    @patch("vllm_ascend.quantization.utils.maybe_auto_detect_quantization")
     @patch("vllm_ascend.ascend_config.init_ascend_config")
     @patch("vllm_ascend.utils.update_aclgraph_sizes")
     @patch("vllm_ascend.utils.get_ascend_device_type", return_value=AscendDeviceType.A3)
     @patch("os.environ", {})
     @patch("vllm_ascend.core.recompute_scheduler.RecomputeSchedulerConfig.initialize_from_config")
     def test_check_and_update_config_basic_config_update(
-        self, mock_init_recompute, mock_soc_version, mock_update_acl, mock_init_ascend
+        self, mock_init_recompute, mock_soc_version, mock_update_acl, mock_init_ascend, mock_auto_detect
     ):
         mock_init_ascend.return_value = TestNPUPlatform.mock_vllm_ascend_config()
         vllm_config = TestNPUPlatform.mock_vllm_config()
@@ -146,11 +147,12 @@ class TestNPUPlatform(TestBase):
 
         mock_init_ascend.assert_called_once_with(vllm_config)
 
+    @patch("vllm_ascend.quantization.utils.maybe_auto_detect_quantization")
     @patch("vllm_ascend.utils.get_ascend_device_type", return_value=AscendDeviceType.A3)
     @patch("vllm_ascend.ascend_config.init_ascend_config")
     @patch("vllm_ascend.core.recompute_scheduler.RecomputeSchedulerConfig.initialize_from_config")
     def test_check_and_update_config_no_model_config_warning(
-        self, mock_init_recompute, mock_init_ascend, mock_soc_version
+        self, mock_init_recompute, mock_init_ascend, mock_soc_version, mock_auto_detect
     ):
         mock_init_ascend.return_value = TestNPUPlatform.mock_vllm_ascend_config()
         vllm_config = TestNPUPlatform.mock_vllm_config()
@@ -172,10 +174,11 @@ class TestNPUPlatform(TestBase):
 
         self.assertTrue("Model config is missing" in cm.output[0])
 
+    @patch("vllm_ascend.quantization.utils.maybe_auto_detect_quantization")
     @patch("vllm_ascend.utils.get_ascend_device_type", return_value=AscendDeviceType.A3)
     @patch("vllm_ascend.ascend_config.init_ascend_config")
     @patch("vllm_ascend.core.recompute_scheduler.RecomputeSchedulerConfig.initialize_from_config")
-    def test_check_and_update_config_enforce_eager_mode(self, mock_init_recompute, mock_init_ascend, mock_soc_version):
+    def test_check_and_update_config_enforce_eager_mode(self, mock_init_recompute, mock_init_ascend, mock_soc_version, mock_auto_detect):
         mock_init_ascend.return_value = TestNPUPlatform.mock_vllm_ascend_config()
         vllm_config = TestNPUPlatform.mock_vllm_config()
         vllm_config.model_config.enforce_eager = True
@@ -206,11 +209,12 @@ class TestNPUPlatform(TestBase):
             CUDAGraphMode.NONE,
         )
 
+    @patch("vllm_ascend.quantization.utils.maybe_auto_detect_quantization")
     @patch("vllm_ascend.utils.get_ascend_device_type", return_value=AscendDeviceType.A3)
     @patch("vllm_ascend.ascend_config.init_ascend_config")
     @patch("vllm_ascend.core.recompute_scheduler.RecomputeSchedulerConfig.initialize_from_config")
     def test_check_and_update_config_unsupported_compilation_level(
-        self, mock_init_recompute, mock_init_ascend, mock_soc_version
+        self, mock_init_recompute, mock_init_ascend, mock_soc_version, mock_auto_detect
     ):
         mock_init_ascend.return_value = TestNPUPlatform.mock_vllm_ascend_config()
         vllm_config = TestNPUPlatform.mock_vllm_config()
@@ -244,9 +248,10 @@ class TestNPUPlatform(TestBase):
             )
 
     @pytest.mark.skip("Revert me when vllm support setting cudagraph_mode on oot platform")
+    @patch("vllm_ascend.quantization.utils.maybe_auto_detect_quantization")
     @patch("vllm_ascend.utils.get_ascend_device_type", return_value=AscendDeviceType.A3)
     @patch("vllm_ascend.ascend_config.init_ascend_config")
-    def test_check_and_update_config_unsupported_cudagraph_mode(self, mock_init_ascend, mock_soc_version):
+    def test_check_and_update_config_unsupported_cudagraph_mode(self, mock_init_ascend, mock_soc_version, mock_auto_detect):
         mock_init_ascend.return_value = TestNPUPlatform.mock_vllm_ascend_config()
         vllm_config = TestNPUPlatform.mock_vllm_config()
         vllm_config.model_config.enforce_eager = False
@@ -268,11 +273,12 @@ class TestNPUPlatform(TestBase):
                 CUDAGraphMode.NONE,
             )
 
+    @patch("vllm_ascend.quantization.utils.maybe_auto_detect_quantization")
     @patch("vllm_ascend.utils.get_ascend_device_type", return_value=AscendDeviceType.A3)
     @patch("vllm_ascend.ascend_config.init_ascend_config")
     @patch("vllm_ascend.core.recompute_scheduler.RecomputeSchedulerConfig.initialize_from_config")
     def test_check_and_update_config_cache_config_block_size(
-        self, mock_init_recompute, mock_init_ascend, mock_soc_version
+        self, mock_init_recompute, mock_init_ascend, mock_soc_version, mock_auto_detect
     ):
         mock_init_ascend.return_value = TestNPUPlatform.mock_vllm_ascend_config()
         vllm_config = TestNPUPlatform.mock_vllm_config()
@@ -292,11 +298,12 @@ class TestNPUPlatform(TestBase):
 
         self.assertEqual(vllm_config.cache_config.block_size, 128)
 
+    @patch("vllm_ascend.quantization.utils.maybe_auto_detect_quantization")
     @patch("vllm_ascend.utils.get_ascend_device_type", return_value=AscendDeviceType.A3)
     @patch("vllm_ascend.ascend_config.init_ascend_config")
     @patch("vllm_ascend.core.recompute_scheduler.RecomputeSchedulerConfig.initialize_from_config")
     def test_check_and_update_config_v1_worker_class_selection(
-        self, mock_init_recompute, mock_init_ascend, mock_soc_version
+        self, mock_init_recompute, mock_init_ascend, mock_soc_version, mock_auto_detect
     ):
         mock_init_ascend.return_value = TestNPUPlatform.mock_vllm_ascend_config()
         vllm_config = TestNPUPlatform.mock_vllm_config()
@@ -327,10 +334,11 @@ class TestNPUPlatform(TestBase):
             "vllm_ascend.xlite.xlite_worker.XliteWorker",
         )
 
+    @patch("vllm_ascend.quantization.utils.maybe_auto_detect_quantization")
     @patch("vllm_ascend.ascend_config.init_ascend_config")
     @patch("vllm_ascend.utils.get_ascend_device_type", return_value=AscendDeviceType._310P)
     @patch("vllm_ascend.core.recompute_scheduler.RecomputeSchedulerConfig.initialize_from_config")
-    def test_check_and_update_config_310p_no_custom_ops(self, mock_init_recompute, mock_soc_version, mock_init_ascend):
+    def test_check_and_update_config_310p_no_custom_ops(self, mock_init_recompute, mock_soc_version, mock_init_ascend, mock_auto_detect):
         mock_init_ascend.return_value = TestNPUPlatform.mock_vllm_ascend_config()
         vllm_config = TestNPUPlatform.mock_vllm_config()
         vllm_config.compilation_config.custom_ops = []