[MM][Bugfix] Add error log for VL models when enabling FLASHCOMM (#4272)

### What this PR does / why we need it?
Add error log for VL models when enabling
`VLLM_ASCEND_ENABLE_FLASHCOMM1=1` or `VLLM_ASCEND_ENABLE_FLASHCOMM=1`
(for backward compatibility).

This is a temporary fix for
https://github.com/vllm-project/vllm-ascend/issues/4132.

### Does this PR introduce _any_ user-facing change?

### How was this patch tested?

- vLLM version: v0.11.0
- vLLM main:
2918c1b49c

Signed-off-by: shen-shanshan <467638484@qq.com>

vllm_ascend/platform.py

@@ -31,7 +31,7 @@ from vllm_ascend.ascend_config import (check_ascend_config, get_ascend_config,
 from vllm_ascend.torchair.utils import (check_torchair_cache_exist,
                                         delete_torchair_cache_file)
 from vllm_ascend.utils import (ASCEND_QUANTIZATION_METHOD, enable_sp, is_310p,
-                               prefill_context_parallel_enable,
+                               is_vl_model, prefill_context_parallel_enable,
                                update_aclgraph_sizes,
                                update_cudagraph_capture_sizes,
                                update_default_aclgraph_sizes, vllm_version_is)
@@ -380,6 +380,14 @@ class NPUPlatform(Platform):
                 "needs to be equal if use cp or dcp > 1 in P/D disaggregate scenario."
             )
 
+        if is_vl_model(vllm_config):
+            if bool(int(os.getenv("VLLM_ASCEND_ENABLE_FLASHCOMM", '0'))) or \
+               bool(int(os.getenv("VLLM_ASCEND_ENABLE_FLASHCOMM1", '0'))):
+                raise ValueError(
+                    "Currently, VL models doesn't support "
+                    "FLASHCOMM in vllm-ascend. We will fix this in the future. "
+                    "Please set VLLM_ASCEND_ENABLE_FLASHCOMM1=0.")
+
     @classmethod
     def import_kernels(cls) -> None:
         # Directly importing vllm_ascend_C prevents ASCEND_RT_VISIBLE_DEVICES

vllm_ascend/utils.py

@@ -57,6 +57,7 @@ _ASCEND_CUSTOMOP_IS_REIGISTERED = False
 _DEFAULT_BUFFER_SIZE = 200
 _MIN_DP_BUFFER_SIZE = 50
 _IS_MOE_MODEL = None
+_IS_VL_MODEL = None
 _ENABLE_SP = None
 _HAS_LAYER_IDX = None
 _SUBSCRIBED_COMPUTE_STREAMS = set()
@@ -829,6 +830,15 @@ def _is_contain_expert(config: Any):
     return False
 
 
+def is_vl_model(vllm_config: VllmConfig):
+    """Checks if the model is a VL model by config"""
+    global _IS_VL_MODEL
+    if _IS_VL_MODEL is None:
+        model_configs = vllm_config.model_config.hf_config.to_dict()
+        _IS_VL_MODEL = "VL" in model_configs["architectures"][0]
+    return _IS_VL_MODEL
+
+
 def weak_ref_tensor(tensor: Any) -> Any:
     """
     Create a weak reference to a tensor.